feat: use json schema for tool argument serialization

- Replace Python representation with JsonSchema for tool arguments
  - Remove deprecated PydanticSchemaParser in favor of direct schema generation
  - Add handling for VAR_POSITIONAL and VAR_KEYWORD parameters
  - Improve tool argument schema collection

.env.test

@@ -0,0 +1,161 @@
+# =============================================================================
+# Test Environment Variables
+# =============================================================================
+# This file contains all environment variables needed to run tests locally
+# in a way that mimics the GitHub Actions CI environment.
+
+# =============================================================================
+
+# -----------------------------------------------------------------------------
+# LLM Provider API Keys
+# -----------------------------------------------------------------------------
+OPENAI_API_KEY=fake-api-key
+ANTHROPIC_API_KEY=fake-anthropic-key
+GEMINI_API_KEY=fake-gemini-key
+AZURE_API_KEY=fake-azure-key
+OPENROUTER_API_KEY=fake-openrouter-key
+
+# -----------------------------------------------------------------------------
+# AWS Credentials
+# -----------------------------------------------------------------------------
+AWS_ACCESS_KEY_ID=fake-aws-access-key
+AWS_SECRET_ACCESS_KEY=fake-aws-secret-key
+AWS_DEFAULT_REGION=us-east-1
+AWS_REGION_NAME=us-east-1
+
+# -----------------------------------------------------------------------------
+# Azure OpenAI Configuration
+# -----------------------------------------------------------------------------
+AZURE_ENDPOINT=https://fake-azure-endpoint.openai.azure.com
+AZURE_OPENAI_ENDPOINT=https://fake-azure-endpoint.openai.azure.com
+AZURE_OPENAI_API_KEY=fake-azure-openai-key
+AZURE_API_VERSION=2024-02-15-preview
+OPENAI_API_VERSION=2024-02-15-preview
+
+# -----------------------------------------------------------------------------
+# Google Cloud Configuration
+# -----------------------------------------------------------------------------
+#GOOGLE_CLOUD_PROJECT=fake-gcp-project
+#GOOGLE_CLOUD_LOCATION=us-central1
+
+# -----------------------------------------------------------------------------
+# OpenAI Configuration
+# -----------------------------------------------------------------------------
+OPENAI_BASE_URL=https://api.openai.com/v1
+OPENAI_API_BASE=https://api.openai.com/v1
+
+# -----------------------------------------------------------------------------
+# Search & Scraping Tool API Keys
+# -----------------------------------------------------------------------------
+SERPER_API_KEY=fake-serper-key
+EXA_API_KEY=fake-exa-key
+BRAVE_API_KEY=fake-brave-key
+FIRECRAWL_API_KEY=fake-firecrawl-key
+TAVILY_API_KEY=fake-tavily-key
+SERPAPI_API_KEY=fake-serpapi-key
+SERPLY_API_KEY=fake-serply-key
+LINKUP_API_KEY=fake-linkup-key
+PARALLEL_API_KEY=fake-parallel-key
+
+# -----------------------------------------------------------------------------
+# Exa Configuration
+# -----------------------------------------------------------------------------
+EXA_BASE_URL=https://api.exa.ai
+
+# -----------------------------------------------------------------------------
+# Web Scraping & Automation
+# -----------------------------------------------------------------------------
+BRIGHT_DATA_API_KEY=fake-brightdata-key
+BRIGHT_DATA_ZONE=fake-zone
+BRIGHTDATA_API_URL=https://api.brightdata.com
+BRIGHTDATA_DEFAULT_TIMEOUT=600
+BRIGHTDATA_DEFAULT_POLLING_INTERVAL=1
+
+OXYLABS_USERNAME=fake-oxylabs-user
+OXYLABS_PASSWORD=fake-oxylabs-pass
+
+SCRAPFLY_API_KEY=fake-scrapfly-key
+SCRAPEGRAPH_API_KEY=fake-scrapegraph-key
+
+BROWSERBASE_API_KEY=fake-browserbase-key
+BROWSERBASE_PROJECT_ID=fake-browserbase-project
+
+HYPERBROWSER_API_KEY=fake-hyperbrowser-key
+MULTION_API_KEY=fake-multion-key
+APIFY_API_TOKEN=fake-apify-token
+
+# -----------------------------------------------------------------------------
+# Database & Vector Store Credentials
+# -----------------------------------------------------------------------------
+SINGLESTOREDB_URL=mysql://fake:fake@localhost:3306/fake
+SINGLESTOREDB_HOST=localhost
+SINGLESTOREDB_PORT=3306
+SINGLESTOREDB_USER=fake-user
+SINGLESTOREDB_PASSWORD=fake-password
+SINGLESTOREDB_DATABASE=fake-database
+SINGLESTOREDB_CONNECT_TIMEOUT=30
+
+SNOWFLAKE_USER=fake-snowflake-user
+SNOWFLAKE_PASSWORD=fake-snowflake-password
+SNOWFLAKE_ACCOUNT=fake-snowflake-account
+SNOWFLAKE_WAREHOUSE=fake-snowflake-warehouse
+SNOWFLAKE_DATABASE=fake-snowflake-database
+SNOWFLAKE_SCHEMA=fake-snowflake-schema
+
+WEAVIATE_URL=http://localhost:8080
+WEAVIATE_API_KEY=fake-weaviate-key
+
+EMBEDCHAIN_DB_URI=sqlite:///test.db
+
+# Databricks Credentials
+DATABRICKS_HOST=https://fake-databricks.cloud.databricks.com
+DATABRICKS_TOKEN=fake-databricks-token
+DATABRICKS_CONFIG_PROFILE=fake-profile
+
+# MongoDB Credentials
+MONGODB_URI=mongodb://fake:fake@localhost:27017/fake
+
+# -----------------------------------------------------------------------------
+# CrewAI Platform & Enterprise
+# -----------------------------------------------------------------------------
+# setting CREWAI_PLATFORM_INTEGRATION_TOKEN causes these test to fail:
+#=========================== short test summary info ============================
+#FAILED tests/test_context.py::TestPlatformIntegrationToken::test_platform_context_manager_basic_usage - AssertionError: assert 'fake-platform-token' is None
+# +  where 'fake-platform-token' = get_platform_integration_token()
+#FAILED tests/test_context.py::TestPlatformIntegrationToken::test_context_var_isolation_between_tests - AssertionError: assert 'fake-platform-token' is None
+# +  where 'fake-platform-token' = get_platform_integration_token()
+#FAILED tests/test_context.py::TestPlatformIntegrationToken::test_multiple_sequential_context_managers - AssertionError: assert 'fake-platform-token' is None
+# +  where 'fake-platform-token' = get_platform_integration_token()
+#CREWAI_PLATFORM_INTEGRATION_TOKEN=fake-platform-token
+CREWAI_PERSONAL_ACCESS_TOKEN=fake-personal-token
+CREWAI_PLUS_URL=https://fake.crewai.com
+
+# -----------------------------------------------------------------------------
+# Other Service API Keys
+# -----------------------------------------------------------------------------
+ZAPIER_API_KEY=fake-zapier-key
+PATRONUS_API_KEY=fake-patronus-key
+MINDS_API_KEY=fake-minds-key
+HF_TOKEN=fake-hf-token
+
+# -----------------------------------------------------------------------------
+# Feature Flags/Testing Modes
+# -----------------------------------------------------------------------------
+CREWAI_DISABLE_TELEMETRY=true
+OTEL_SDK_DISABLED=true
+CREWAI_TESTING=true
+CREWAI_TRACING_ENABLED=false
+
+# -----------------------------------------------------------------------------
+# Testing/CI Configuration
+# -----------------------------------------------------------------------------
+# VCR recording mode: "none" (default), "new_episodes", "all", "once"
+PYTEST_VCR_RECORD_MODE=none
+
+# Set to "true" by GitHub when running in GitHub Actions
+# GITHUB_ACTIONS=false
+
+# -----------------------------------------------------------------------------
+# Python Configuration
+# -----------------------------------------------------------------------------
+PYTHONUNBUFFERED=1

.github/workflows/tests.yml

@@ -5,18 +5,6 @@ on: [pull_request]
 permissions:
   contents: read
 
-env:
-  OPENAI_API_KEY: fake-api-key
-  PYTHONUNBUFFERED: 1
-  BRAVE_API_KEY: fake-brave-key
-  SNOWFLAKE_USER: fake-snowflake-user
-  SNOWFLAKE_PASSWORD: fake-snowflake-password
-  SNOWFLAKE_ACCOUNT: fake-snowflake-account
-  SNOWFLAKE_WAREHOUSE: fake-snowflake-warehouse
-  SNOWFLAKE_DATABASE: fake-snowflake-database
-  SNOWFLAKE_SCHEMA: fake-snowflake-schema
-  EMBEDCHAIN_DB_URI: sqlite:///test.db
-
 jobs:
   tests:
     name: tests (${{ matrix.python-version }})
@@ -84,26 +72,20 @@ jobs:
           # fi
 
           cd lib/crewai && uv run pytest \
-            --block-network \
-            --timeout=30 \
             -vv \
             --splits 8 \
             --group ${{ matrix.group }} \
             $DURATIONS_ARG \
             --durations=10 \
-            -n auto \
             --maxfail=3
 
       - name: Run tool tests (group ${{ matrix.group }} of 8)
         run: |
           cd lib/crewai-tools && uv run pytest \
-            --block-network \
-            --timeout=30 \
             -vv \
             --splits 8 \
             --group ${{ matrix.group }} \
             --durations=10 \
-            -n auto \
             --maxfail=3
 
 

conftest.py

@@ -0,0 +1,197 @@
+"""Pytest configuration for crewAI workspace."""
+
+from collections.abc import Generator
+import os
+from pathlib import Path
+import tempfile
+from typing import Any
+
+from dotenv import load_dotenv
+import pytest
+from vcr.request import Request  # type: ignore[import-untyped]
+
+
+env_test_path = Path(__file__).parent / ".env.test"
+load_dotenv(env_test_path, override=True)
+load_dotenv(override=True)
+
+
+@pytest.fixture(autouse=True, scope="function")
+def cleanup_event_handlers() -> Generator[None, Any, None]:
+    """Clean up event bus handlers after each test to prevent test pollution."""
+    yield
+
+    try:
+        from crewai.events.event_bus import crewai_event_bus
+
+        with crewai_event_bus._rwlock.w_locked():
+            crewai_event_bus._sync_handlers.clear()
+            crewai_event_bus._async_handlers.clear()
+    except Exception:  # noqa: S110
+        pass
+
+
+@pytest.fixture(autouse=True, scope="function")
+def setup_test_environment() -> Generator[None, Any, None]:
+    """Setup test environment for crewAI workspace."""
+    with tempfile.TemporaryDirectory() as temp_dir:
+        storage_dir = Path(temp_dir) / "crewai_test_storage"
+        storage_dir.mkdir(parents=True, exist_ok=True)
+
+        if not storage_dir.exists() or not storage_dir.is_dir():
+            raise RuntimeError(
+                f"Failed to create test storage directory: {storage_dir}"
+            )
+
+        try:
+            test_file = storage_dir / ".permissions_test"
+            test_file.touch()
+            test_file.unlink()
+        except (OSError, IOError) as e:
+            raise RuntimeError(
+                f"Test storage directory {storage_dir} is not writable: {e}"
+            ) from e
+
+        os.environ["CREWAI_STORAGE_DIR"] = str(storage_dir)
+        os.environ["CREWAI_TESTING"] = "true"
+
+        try:
+            yield
+        finally:
+            os.environ.pop("CREWAI_TESTING", "true")
+            os.environ.pop("CREWAI_STORAGE_DIR", None)
+            os.environ.pop("CREWAI_DISABLE_TELEMETRY", "true")
+            os.environ.pop("OTEL_SDK_DISABLED", "true")
+            os.environ.pop("OPENAI_BASE_URL", "https://api.openai.com/v1")
+            os.environ.pop("OPENAI_API_BASE", "https://api.openai.com/v1")
+
+
+HEADERS_TO_FILTER = {
+    "authorization": "AUTHORIZATION-XXX",
+    "content-security-policy": "CSP-FILTERED",
+    "cookie": "COOKIE-XXX",
+    "set-cookie": "SET-COOKIE-XXX",
+    "permissions-policy": "PERMISSIONS-POLICY-XXX",
+    "referrer-policy": "REFERRER-POLICY-XXX",
+    "strict-transport-security": "STS-XXX",
+    "x-content-type-options": "X-CONTENT-TYPE-XXX",
+    "x-frame-options": "X-FRAME-OPTIONS-XXX",
+    "x-permitted-cross-domain-policies": "X-PERMITTED-XXX",
+    "x-request-id": "X-REQUEST-ID-XXX",
+    "x-runtime": "X-RUNTIME-XXX",
+    "x-xss-protection": "X-XSS-PROTECTION-XXX",
+    "x-stainless-arch": "X-STAINLESS-ARCH-XXX",
+    "x-stainless-os": "X-STAINLESS-OS-XXX",
+    "x-stainless-read-timeout": "X-STAINLESS-READ-TIMEOUT-XXX",
+    "cf-ray": "CF-RAY-XXX",
+    "etag": "ETAG-XXX",
+    "Strict-Transport-Security": "STS-XXX",
+    "access-control-expose-headers": "ACCESS-CONTROL-XXX",
+    "openai-organization": "OPENAI-ORG-XXX",
+    "openai-project": "OPENAI-PROJECT-XXX",
+    "x-ratelimit-limit-requests": "X-RATELIMIT-LIMIT-REQUESTS-XXX",
+    "x-ratelimit-limit-tokens": "X-RATELIMIT-LIMIT-TOKENS-XXX",
+    "x-ratelimit-remaining-requests": "X-RATELIMIT-REMAINING-REQUESTS-XXX",
+    "x-ratelimit-remaining-tokens": "X-RATELIMIT-REMAINING-TOKENS-XXX",
+    "x-ratelimit-reset-requests": "X-RATELIMIT-RESET-REQUESTS-XXX",
+    "x-ratelimit-reset-tokens": "X-RATELIMIT-RESET-TOKENS-XXX",
+    "x-goog-api-key": "X-GOOG-API-KEY-XXX",
+    "api-key": "X-API-KEY-XXX",
+    "User-Agent": "X-USER-AGENT-XXX",
+    "apim-request-id:": "X-API-CLIENT-REQUEST-ID-XXX",
+    "azureml-model-session": "AZUREML-MODEL-SESSION-XXX",
+    "x-ms-client-request-id": "X-MS-CLIENT-REQUEST-ID-XXX",
+    "x-ms-region": "X-MS-REGION-XXX",
+    "apim-request-id": "APIM-REQUEST-ID-XXX",
+    "x-api-key": "X-API-KEY-XXX",
+    "anthropic-organization-id": "ANTHROPIC-ORGANIZATION-ID-XXX",
+    "request-id": "REQUEST-ID-XXX",
+    "anthropic-ratelimit-input-tokens-limit": "ANTHROPIC-RATELIMIT-INPUT-TOKENS-LIMIT-XXX",
+    "anthropic-ratelimit-input-tokens-remaining": "ANTHROPIC-RATELIMIT-INPUT-TOKENS-REMAINING-XXX",
+    "anthropic-ratelimit-input-tokens-reset": "ANTHROPIC-RATELIMIT-INPUT-TOKENS-RESET-XXX",
+    "anthropic-ratelimit-output-tokens-limit": "ANTHROPIC-RATELIMIT-OUTPUT-TOKENS-LIMIT-XXX",
+    "anthropic-ratelimit-output-tokens-remaining": "ANTHROPIC-RATELIMIT-OUTPUT-TOKENS-REMAINING-XXX",
+    "anthropic-ratelimit-output-tokens-reset": "ANTHROPIC-RATELIMIT-OUTPUT-TOKENS-RESET-XXX",
+    "anthropic-ratelimit-tokens-limit": "ANTHROPIC-RATELIMIT-TOKENS-LIMIT-XXX",
+    "anthropic-ratelimit-tokens-remaining": "ANTHROPIC-RATELIMIT-TOKENS-REMAINING-XXX",
+    "anthropic-ratelimit-tokens-reset": "ANTHROPIC-RATELIMIT-TOKENS-RESET-XXX",
+    "x-amz-date": "X-AMZ-DATE-XXX",
+    "amz-sdk-invocation-id": "AMZ-SDK-INVOCATION-ID-XXX",
+    "accept-encoding": "ACCEPT-ENCODING-XXX",
+    "x-amzn-requestid": "X-AMZN-REQUESTID-XXX",
+    "x-amzn-RequestId": "X-AMZN-REQUESTID-XXX",
+}
+
+
+def _filter_request_headers(request: Request) -> Request:  # type: ignore[no-any-unimported]
+    """Filter sensitive headers from request before recording."""
+    for header_name, replacement in HEADERS_TO_FILTER.items():
+        for variant in [header_name, header_name.upper(), header_name.title()]:
+            if variant in request.headers:
+                request.headers[variant] = [replacement]
+
+    request.method = request.method.upper()
+    return request
+
+
+def _filter_response_headers(response: dict[str, Any]) -> dict[str, Any]:
+    """Filter sensitive headers from response before recording."""
+    # Remove Content-Encoding to prevent decompression issues on replay
+    for encoding_header in ["Content-Encoding", "content-encoding"]:
+        response["headers"].pop(encoding_header, None)
+
+    for header_name, replacement in HEADERS_TO_FILTER.items():
+        for variant in [header_name, header_name.upper(), header_name.title()]:
+            if variant in response["headers"]:
+                response["headers"][variant] = [replacement]
+    return response
+
+
+@pytest.fixture(scope="module")
+def vcr_cassette_dir(request: Any) -> str:
+    """Generate cassette directory path based on test module location.
+
+    Organizes cassettes to mirror test directory structure within each package:
+    lib/crewai/tests/llms/google/test_google.py -> lib/crewai/tests/cassettes/llms/google/
+    lib/crewai-tools/tests/tools/test_search.py -> lib/crewai-tools/tests/cassettes/tools/
+    """
+    test_file = Path(request.fspath)
+
+    for parent in test_file.parents:
+        if parent.name in ("crewai", "crewai-tools") and parent.parent.name == "lib":
+            package_root = parent
+            break
+    else:
+        package_root = test_file.parent
+
+    tests_root = package_root / "tests"
+    test_dir = test_file.parent
+
+    if test_dir != tests_root:
+        relative_path = test_dir.relative_to(tests_root)
+        cassette_dir = tests_root / "cassettes" / relative_path
+    else:
+        cassette_dir = tests_root / "cassettes"
+
+    cassette_dir.mkdir(parents=True, exist_ok=True)
+
+    return str(cassette_dir)
+
+
+@pytest.fixture(scope="module")
+def vcr_config(vcr_cassette_dir: str) -> dict[str, Any]:
+    """Configure VCR with organized cassette storage."""
+    config = {
+        "cassette_library_dir": vcr_cassette_dir,
+        "record_mode": os.getenv("PYTEST_VCR_RECORD_MODE", "once"),
+        "filter_headers": [(k, v) for k, v in HEADERS_TO_FILTER.items()],
+        "before_record_request": _filter_request_headers,
+        "before_record_response": _filter_response_headers,
+        "filter_query_parameters": ["key"],
+        "match_on": ["method", "scheme", "host", "port", "path"],
+    }
+
+    if os.getenv("GITHUB_ACTIONS") == "true":
+        config["record_mode"] = "none"
+
+    return config

docs/docs.json

@@ -253,7 +253,8 @@
                     "pages": [
                       "en/tools/integration/overview",
                       "en/tools/integration/bedrockinvokeagenttool",
-                      "en/tools/integration/crewaiautomationtool"
+                      "en/tools/integration/crewaiautomationtool",
+                      "en/tools/integration/mergeagenthandlertool"
                     ]
                   },
                   {

docs/en/concepts/crews.mdx

@@ -307,12 +307,27 @@ print(result)
 
 ### Different Ways to Kick Off a Crew
 
-Once your crew is assembled, initiate the workflow with the appropriate kickoff method. CrewAI provides several methods for better control over the kickoff process: `kickoff()`, `kickoff_for_each()`, `kickoff_async()`, and `kickoff_for_each_async()`.
+Once your crew is assembled, initiate the workflow with the appropriate kickoff method. CrewAI provides several methods for better control over the kickoff process.
+
+#### Synchronous Methods
 
 - `kickoff()`: Starts the execution process according to the defined process flow.
 - `kickoff_for_each()`: Executes tasks sequentially for each provided input event or item in the collection.
-- `kickoff_async()`: Initiates the workflow asynchronously.
-- `kickoff_for_each_async()`: Executes tasks concurrently for each provided input event or item, leveraging asynchronous processing.
+
+#### Asynchronous Methods
+
+CrewAI offers two approaches for async execution:
+
+| Method | Type | Description |
+|--------|------|-------------|
+| `akickoff()` | Native async | True async/await throughout the entire execution chain |
+| `akickoff_for_each()` | Native async | Native async execution for each input in a list |
+| `kickoff_async()` | Thread-based | Wraps synchronous execution in `asyncio.to_thread` |
+| `kickoff_for_each_async()` | Thread-based | Thread-based async for each input in a list |
+
+<Note>
+For high-concurrency workloads, `akickoff()` and `akickoff_for_each()` are recommended as they use native async for task execution, memory operations, and knowledge retrieval.
+</Note>
 
 ```python Code
 # Start the crew's task execution
@@ -325,19 +340,30 @@ results = my_crew.kickoff_for_each(inputs=inputs_array)
 for result in results:
     print(result)
 
-# Example of using kickoff_async
+# Example of using native async with akickoff
+inputs = {'topic': 'AI in healthcare'}
+async_result = await my_crew.akickoff(inputs=inputs)
+print(async_result)
+
+# Example of using native async with akickoff_for_each
+inputs_array = [{'topic': 'AI in healthcare'}, {'topic': 'AI in finance'}]
+async_results = await my_crew.akickoff_for_each(inputs=inputs_array)
+for async_result in async_results:
+    print(async_result)
+
+# Example of using thread-based kickoff_async
 inputs = {'topic': 'AI in healthcare'}
 async_result = await my_crew.kickoff_async(inputs=inputs)
 print(async_result)
 
-# Example of using kickoff_for_each_async
+# Example of using thread-based kickoff_for_each_async
 inputs_array = [{'topic': 'AI in healthcare'}, {'topic': 'AI in finance'}]
 async_results = await my_crew.kickoff_for_each_async(inputs=inputs_array)
 for async_result in async_results:
     print(async_result)
 ```
 
-These methods provide flexibility in how you manage and execute tasks within your crew, allowing for both synchronous and asynchronous workflows tailored to your needs.
+These methods provide flexibility in how you manage and execute tasks within your crew, allowing for both synchronous and asynchronous workflows tailored to your needs. For detailed async examples, see the [Kickoff Crew Asynchronously](/en/learn/kickoff-async) guide.
 
 ### Streaming Crew Execution
 

docs/en/concepts/llms.mdx

@@ -283,11 +283,54 @@ In this section, you'll find detailed examples that help you select, configure,
     )
     ```
 
+    **Extended Thinking (Claude Sonnet 4 and Beyond):**
+
+    CrewAI supports Anthropic's Extended Thinking feature, which allows Claude to think through problems in a more human-like way before responding. This is particularly useful for complex reasoning, analysis, and problem-solving tasks.
+
+    ```python Code
+    from crewai import LLM
+
+    # Enable extended thinking with default settings
+    llm = LLM(
+        model="anthropic/claude-sonnet-4",
+        thinking={"type": "enabled"},
+        max_tokens=10000
+    )
+
+    # Configure thinking with budget control
+    llm = LLM(
+        model="anthropic/claude-sonnet-4",
+        thinking={
+            "type": "enabled",
+            "budget_tokens": 5000  # Limit thinking tokens
+        },
+        max_tokens=10000
+    )
+    ```
+
+    **Thinking Configuration Options:**
+    - `type`: Set to `"enabled"` to activate extended thinking mode
+    - `budget_tokens` (optional): Maximum tokens to use for thinking (helps control costs)
+
+    **Models Supporting Extended Thinking:**
+    - `claude-sonnet-4` and newer models
+    - `claude-3-7-sonnet` (with extended thinking capabilities)
+
+    **When to Use Extended Thinking:**
+    - Complex reasoning and multi-step problem solving
+    - Mathematical calculations and proofs
+    - Code analysis and debugging
+    - Strategic planning and decision making
+    - Research and analytical tasks
+
+    **Note:** Extended thinking consumes additional tokens but can significantly improve response quality for complex tasks.
+
     **Supported Environment Variables:**
     - `ANTHROPIC_API_KEY`: Your Anthropic API key (required)
 
     **Features:**
     - Native tool use support for Claude 3+ models
+    - Extended Thinking support for Claude Sonnet 4+
     - Streaming support for real-time responses
     - Automatic system message handling
     - Stop sequences for controlled output
@@ -305,6 +348,7 @@ In this section, you'll find detailed examples that help you select, configure,
 
     | Model                        | Context Window | Best For                                      |
     |------------------------------|----------------|-----------------------------------------------|
+    | claude-sonnet-4              | 200,000 tokens | Latest with extended thinking capabilities    |
     | claude-3-7-sonnet            | 200,000 tokens | Advanced reasoning and agentic tasks          |
     | claude-3-5-sonnet-20241022   | 200,000 tokens | Latest Sonnet with best performance           |
     | claude-3-5-haiku             | 200,000 tokens | Fast, compact model for quick responses       |
@@ -1089,6 +1133,50 @@ CrewAI supports streaming responses from LLMs, allowing your application to rece
   </Tab>
 </Tabs>
 
+## Async LLM Calls
+
+CrewAI supports asynchronous LLM calls for improved performance and concurrency in your AI workflows. Async calls allow you to run multiple LLM requests concurrently without blocking, making them ideal for high-throughput applications and parallel agent operations.
+
+<Tabs>
+  <Tab title="Basic Usage">
+    Use the `acall` method for asynchronous LLM requests:
+
+    ```python
+    import asyncio
+    from crewai import LLM
+
+    async def main():
+        llm = LLM(model="openai/gpt-4o")
+
+        # Single async call
+        response = await llm.acall("What is the capital of France?")
+        print(response)
+
+    asyncio.run(main())
+    ```
+
+    The `acall` method supports all the same parameters as the synchronous `call` method, including messages, tools, and callbacks.
+  </Tab>
+
+  <Tab title="With Streaming">
+    Combine async calls with streaming for real-time concurrent responses:
+
+    ```python
+    import asyncio
+    from crewai import LLM
+
+    async def stream_async():
+        llm = LLM(model="openai/gpt-4o", stream=True)
+
+        response = await llm.acall("Write a short story about AI")
+
+        print(response)
+
+    asyncio.run(stream_async())
+    ```
+  </Tab>
+</Tabs>
+
 ## Structured LLM Calls
 
 CrewAI supports structured responses from LLM calls by allowing you to define a `response_format` using a Pydantic model. This enables the framework to automatically parse and validate the output, making it easier to integrate the response into your application without manual post-processing.

docs/en/concepts/memory.mdx

@@ -515,8 +515,7 @@ crew = Crew(
         "provider": "huggingface",
         "config": {
             "api_key": "your-hf-token",  # Optional for public models
-            "model": "sentence-transformers/all-MiniLM-L6-v2",
-            "api_url": "https://api-inference.huggingface.co"  # or your custom endpoint
+            "model": "sentence-transformers/all-MiniLM-L6-v2"
         }
     }
 )

docs/en/enterprise/guides/deploy-crew.mdx

@@ -187,6 +187,97 @@ You can also deploy your crews directly through the CrewAI AOP web interface by
 
 </Steps>
 
+## Option 3: Redeploy Using API (CI/CD Integration)
+
+For automated deployments in CI/CD pipelines, you can use the CrewAI API to trigger redeployments of existing crews. This is particularly useful for GitHub Actions, Jenkins, or other automation workflows.
+
+<Steps>
+  <Step title="Get Your Personal Access Token">
+
+    Navigate to your CrewAI AOP account settings to generate an API token:
+
+    1. Go to [app.crewai.com](https://app.crewai.com)
+    2. Click on **Settings** → **Account** → **Personal Access Token**
+    3. Generate a new token and copy it securely
+    4. Store this token as a secret in your CI/CD system
+
+  </Step>
+
+  <Step title="Find Your Automation UUID">
+
+    Locate the unique identifier for your deployed crew:
+
+    1. Go to **Automations** in your CrewAI AOP dashboard
+    2. Select your existing automation/crew
+    3. Click on **Additional Details**
+    4. Copy the **UUID** - this identifies your specific crew deployment
+
+  </Step>
+
+  <Step title="Trigger Redeployment via API">
+
+    Use the Deploy API endpoint to trigger a redeployment:
+
+    ```bash
+    curl -i -X POST \
+         -H "Authorization: Bearer YOUR_PERSONAL_ACCESS_TOKEN" \
+         https://app.crewai.com/crewai_plus/api/v1/crews/YOUR-AUTOMATION-UUID/deploy
+
+    # HTTP/2 200
+    # content-type: application/json
+    #
+    # {
+    #   "uuid": "your-automation-uuid",
+    #   "status": "Deploy Enqueued",
+    #   "public_url": "https://your-crew-deployment.crewai.com",
+    #   "token": "your-bearer-token"
+    # }
+    ```
+
+    <Info>
+    If your automation was first created connected to Git, the API will automatically pull the latest changes from your repository before redeploying.
+    </Info>
+
+
+  </Step>
+
+  <Step title="GitHub Actions Integration Example">
+
+    Here's a GitHub Actions workflow with more complex deployment triggers:
+
+    ```yaml
+    name: Deploy CrewAI Automation
+
+    on:
+      push:
+        branches: [ main ]
+      pull_request:
+        types: [ labeled ]
+      release:
+        types: [ published ]
+
+    jobs:
+      deploy:
+        runs-on: ubuntu-latest
+        if: |
+          (github.event_name == 'push' && github.ref == 'refs/heads/main') ||
+          (github.event_name == 'pull_request' && contains(github.event.pull_request.labels.*.name, 'deploy')) ||
+          (github.event_name == 'release')
+        steps:
+          - name: Trigger CrewAI Redeployment
+            run: |
+              curl -X POST \
+                   -H "Authorization: Bearer ${{ secrets.CREWAI_PAT }}" \
+                   https://app.crewai.com/crewai_plus/api/v1/crews/${{ secrets.CREWAI_AUTOMATION_UUID }}/deploy
+    ```
+
+    <Tip>
+    Add `CREWAI_PAT` and `CREWAI_AUTOMATION_UUID` as repository secrets. For PR deployments, add a "deploy" label to trigger the workflow.
+    </Tip>
+
+    </Step>
+  </Steps>
+
 ## ⚠️ Environment Variable Security Requirements
 
 <Warning>

docs/en/learn/create-custom-tools.mdx

@@ -66,5 +66,55 @@ def my_cache_strategy(arguments: dict, result: str) -> bool:
 cached_tool.cache_function = my_cache_strategy
 ```
 
+### Creating Async Tools
+
+CrewAI supports async tools for non-blocking I/O operations. This is useful when your tool needs to make HTTP requests, database queries, or other I/O-bound operations.
+
+#### Using the `@tool` Decorator with Async Functions
+
+The simplest way to create an async tool is using the `@tool` decorator with an async function:
+
+```python Code
+import aiohttp
+from crewai.tools import tool
+
+@tool("Async Web Fetcher")
+async def fetch_webpage(url: str) -> str:
+    """Fetch content from a webpage asynchronously."""
+    async with aiohttp.ClientSession() as session:
+        async with session.get(url) as response:
+            return await response.text()
+```
+
+#### Subclassing `BaseTool` with Async Support
+
+For more control, subclass `BaseTool` and implement both `_run` (sync) and `_arun` (async) methods:
+
+```python Code
+import requests
+import aiohttp
+from crewai.tools import BaseTool
+from pydantic import BaseModel, Field
+
+class WebFetcherInput(BaseModel):
+    """Input schema for WebFetcher."""
+    url: str = Field(..., description="The URL to fetch")
+
+class WebFetcherTool(BaseTool):
+    name: str = "Web Fetcher"
+    description: str = "Fetches content from a URL"
+    args_schema: type[BaseModel] = WebFetcherInput
+
+    def _run(self, url: str) -> str:
+        """Synchronous implementation."""
+        return requests.get(url).text
+
+    async def _arun(self, url: str) -> str:
+        """Asynchronous implementation for non-blocking I/O."""
+        async with aiohttp.ClientSession() as session:
+            async with session.get(url) as response:
+                return await response.text()
+```
+
 By adhering to these guidelines and incorporating new functionalities and collaboration tools into your tool creation and management processes,
 you can leverage the full capabilities of the CrewAI framework, enhancing both the development experience and the efficiency of your AI agents.

docs/en/learn/kickoff-async.mdx

@@ -7,17 +7,28 @@ mode: "wide"
 
 ## Introduction
 
-CrewAI provides the ability to kickoff a crew asynchronously, allowing you to start the crew execution in a non-blocking manner. 
+CrewAI provides the ability to kickoff a crew asynchronously, allowing you to start the crew execution in a non-blocking manner.
 This feature is particularly useful when you want to run multiple crews concurrently or when you need to perform other tasks while the crew is executing.
 
-## Asynchronous Crew Execution
+CrewAI offers two approaches for async execution:
 
-To kickoff a crew asynchronously, use the `kickoff_async()` method. This method initiates the crew execution in a separate thread, allowing the main thread to continue executing other tasks.
+| Method | Type | Description |
+|--------|------|-------------|
+| `akickoff()` | Native async | True async/await throughout the entire execution chain |
+| `kickoff_async()` | Thread-based | Wraps synchronous execution in `asyncio.to_thread` |
+
+<Note>
+For high-concurrency workloads, `akickoff()` is recommended as it uses native async for task execution, memory operations, and knowledge retrieval.
+</Note>
+
+## Native Async Execution with `akickoff()`
+
+The `akickoff()` method provides true native async execution, using async/await throughout the entire execution chain including task execution, memory operations, and knowledge queries.
 
 ### Method Signature
 
 ```python Code
-def kickoff_async(self, inputs: dict) -> CrewOutput:
+async def akickoff(self, inputs: dict) -> CrewOutput:
 ```
 
 ### Parameters
@@ -28,23 +39,13 @@ def kickoff_async(self, inputs: dict) -> CrewOutput:
 
 - `CrewOutput`: An object representing the result of the crew execution.
 
-## Potential Use Cases
-
-- **Parallel Content Generation**: Kickoff multiple independent crews asynchronously, each responsible for generating content on different topics. For example, one crew might research and draft an article on AI trends, while another crew generates social media posts about a new product launch. Each crew operates independently, allowing content production to scale efficiently.
-
-- **Concurrent Market Research Tasks**: Launch multiple crews asynchronously to conduct market research in parallel. One crew might analyze industry trends, while another examines competitor strategies, and yet another evaluates consumer sentiment. Each crew independently completes its task, enabling faster and more comprehensive insights.
-
-- **Independent Travel Planning Modules**: Execute separate crews to independently plan different aspects of a trip. One crew might handle flight options, another handles accommodation, and a third plans activities. Each crew works asynchronously, allowing various components of the trip to be planned simultaneously and independently for faster results.
-
-## Example: Single Asynchronous Crew Execution
-
-Here's an example of how to kickoff a crew asynchronously using asyncio and awaiting the result:
+### Example: Native Async Crew Execution
 
 ```python Code
 import asyncio
 from crewai import Crew, Agent, Task
 
-# Create an agent with code execution enabled
+# Create an agent
 coding_agent = Agent(
     role="Python Data Analyst",
     goal="Analyze data and provide insights using Python",
@@ -52,37 +53,165 @@ coding_agent = Agent(
     allow_code_execution=True
 )
 
-# Create a task that requires code execution
+# Create a task
 data_analysis_task = Task(
     description="Analyze the given dataset and calculate the average age of participants. Ages: {ages}",
     agent=coding_agent,
     expected_output="The average age of the participants."
 )
 
-# Create a crew and add the task
+# Create a crew
 analysis_crew = Crew(
     agents=[coding_agent],
     tasks=[data_analysis_task]
 )
 
-# Async function to kickoff the crew asynchronously
-async def async_crew_execution():
-    result = await analysis_crew.kickoff_async(inputs={"ages": [25, 30, 35, 40, 45]})
+# Native async execution
+async def main():
+    result = await analysis_crew.akickoff(inputs={"ages": [25, 30, 35, 40, 45]})
     print("Crew Result:", result)
 
-# Run the async function
-asyncio.run(async_crew_execution())
+asyncio.run(main())
 ```
 
-## Example: Multiple Asynchronous Crew Executions
+### Example: Multiple Native Async Crews
 
-In this example, we'll show how to kickoff multiple crews asynchronously and wait for all of them to complete using `asyncio.gather()`:
+Run multiple crews concurrently using `asyncio.gather()` with native async:
+
+```python Code
+import asyncio
+from crewai import Crew, Agent, Task
+
+coding_agent = Agent(
+    role="Python Data Analyst",
+    goal="Analyze data and provide insights using Python",
+    backstory="You are an experienced data analyst with strong Python skills.",
+    allow_code_execution=True
+)
+
+task_1 = Task(
+    description="Analyze the first dataset and calculate the average age. Ages: {ages}",
+    agent=coding_agent,
+    expected_output="The average age of the participants."
+)
+
+task_2 = Task(
+    description="Analyze the second dataset and calculate the average age. Ages: {ages}",
+    agent=coding_agent,
+    expected_output="The average age of the participants."
+)
+
+crew_1 = Crew(agents=[coding_agent], tasks=[task_1])
+crew_2 = Crew(agents=[coding_agent], tasks=[task_2])
+
+async def main():
+    results = await asyncio.gather(
+        crew_1.akickoff(inputs={"ages": [25, 30, 35, 40, 45]}),
+        crew_2.akickoff(inputs={"ages": [20, 22, 24, 28, 30]})
+    )
+
+    for i, result in enumerate(results, 1):
+        print(f"Crew {i} Result:", result)
+
+asyncio.run(main())
+```
+
+### Example: Native Async for Multiple Inputs
+
+Use `akickoff_for_each()` to execute your crew against multiple inputs concurrently with native async:
+
+```python Code
+import asyncio
+from crewai import Crew, Agent, Task
+
+coding_agent = Agent(
+    role="Python Data Analyst",
+    goal="Analyze data and provide insights using Python",
+    backstory="You are an experienced data analyst with strong Python skills.",
+    allow_code_execution=True
+)
+
+data_analysis_task = Task(
+    description="Analyze the dataset and calculate the average age. Ages: {ages}",
+    agent=coding_agent,
+    expected_output="The average age of the participants."
+)
+
+analysis_crew = Crew(
+    agents=[coding_agent],
+    tasks=[data_analysis_task]
+)
+
+async def main():
+    datasets = [
+        {"ages": [25, 30, 35, 40, 45]},
+        {"ages": [20, 22, 24, 28, 30]},
+        {"ages": [30, 35, 40, 45, 50]}
+    ]
+
+    results = await analysis_crew.akickoff_for_each(datasets)
+
+    for i, result in enumerate(results, 1):
+        print(f"Dataset {i} Result:", result)
+
+asyncio.run(main())
+```
+
+## Thread-Based Async with `kickoff_async()`
+
+The `kickoff_async()` method provides async execution by wrapping the synchronous `kickoff()` in a thread. This is useful for simpler async integration or backward compatibility.
+
+### Method Signature
+
+```python Code
+async def kickoff_async(self, inputs: dict) -> CrewOutput:
+```
+
+### Parameters
+
+- `inputs` (dict): A dictionary containing the input data required for the tasks.
+
+### Returns
+
+- `CrewOutput`: An object representing the result of the crew execution.
+
+### Example: Thread-Based Async Execution
+
+```python Code
+import asyncio
+from crewai import Crew, Agent, Task
+
+coding_agent = Agent(
+    role="Python Data Analyst",
+    goal="Analyze data and provide insights using Python",
+    backstory="You are an experienced data analyst with strong Python skills.",
+    allow_code_execution=True
+)
+
+data_analysis_task = Task(
+    description="Analyze the given dataset and calculate the average age of participants. Ages: {ages}",
+    agent=coding_agent,
+    expected_output="The average age of the participants."
+)
+
+analysis_crew = Crew(
+    agents=[coding_agent],
+    tasks=[data_analysis_task]
+)
+
+async def async_crew_execution():
+    result = await analysis_crew.kickoff_async(inputs={"ages": [25, 30, 35, 40, 45]})
+    print("Crew Result:", result)
+
+asyncio.run(async_crew_execution())
+```
+
+### Example: Multiple Thread-Based Async Crews
 
 ```python Code
 import asyncio
 from crewai import Crew, Agent, Task
 
-# Create an agent with code execution enabled
 coding_agent = Agent(
     role="Python Data Analyst",
     goal="Analyze data and provide insights using Python",
@@ -90,7 +219,6 @@ coding_agent = Agent(
     allow_code_execution=True
 )
 
-# Create tasks that require code execution
 task_1 = Task(
     description="Analyze the first dataset and calculate the average age of participants. Ages: {ages}",
     agent=coding_agent,
@@ -103,22 +231,76 @@ task_2 = Task(
     expected_output="The average age of the participants."
 )
 
-# Create two crews and add tasks
 crew_1 = Crew(agents=[coding_agent], tasks=[task_1])
 crew_2 = Crew(agents=[coding_agent], tasks=[task_2])
 
-# Async function to kickoff multiple crews asynchronously and wait for all to finish
 async def async_multiple_crews():
-    # Create coroutines for concurrent execution
     result_1 = crew_1.kickoff_async(inputs={"ages": [25, 30, 35, 40, 45]})
     result_2 = crew_2.kickoff_async(inputs={"ages": [20, 22, 24, 28, 30]})
 
-    # Wait for both crews to finish
     results = await asyncio.gather(result_1, result_2)
 
     for i, result in enumerate(results, 1):
         print(f"Crew {i} Result:", result)
 
-# Run the async function
 asyncio.run(async_multiple_crews())
 ```
+
+## Async Streaming
+
+Both async methods support streaming when `stream=True` is set on the crew:
+
+```python Code
+import asyncio
+from crewai import Crew, Agent, Task
+
+agent = Agent(
+    role="Researcher",
+    goal="Research and summarize topics",
+    backstory="You are an expert researcher."
+)
+
+task = Task(
+    description="Research the topic: {topic}",
+    agent=agent,
+    expected_output="A comprehensive summary of the topic."
+)
+
+crew = Crew(
+    agents=[agent],
+    tasks=[task],
+    stream=True  # Enable streaming
+)
+
+async def main():
+    streaming_output = await crew.akickoff(inputs={"topic": "AI trends in 2024"})
+
+    # Async iteration over streaming chunks
+    async for chunk in streaming_output:
+        print(f"Chunk: {chunk.content}")
+
+    # Access final result after streaming completes
+    result = streaming_output.result
+    print(f"Final result: {result.raw}")
+
+asyncio.run(main())
+```
+
+## Potential Use Cases
+
+- **Parallel Content Generation**: Kickoff multiple independent crews asynchronously, each responsible for generating content on different topics. For example, one crew might research and draft an article on AI trends, while another crew generates social media posts about a new product launch.
+
+- **Concurrent Market Research Tasks**: Launch multiple crews asynchronously to conduct market research in parallel. One crew might analyze industry trends, while another examines competitor strategies, and yet another evaluates consumer sentiment.
+
+- **Independent Travel Planning Modules**: Execute separate crews to independently plan different aspects of a trip. One crew might handle flight options, another handles accommodation, and a third plans activities.
+
+## Choosing Between `akickoff()` and `kickoff_async()`
+
+| Feature | `akickoff()` | `kickoff_async()` |
+|---------|--------------|-------------------|
+| Execution model | Native async/await | Thread-based wrapper |
+| Task execution | Async with `aexecute_sync()` | Sync in thread pool |
+| Memory operations | Async | Sync in thread pool |
+| Knowledge retrieval | Async | Sync in thread pool |
+| Best for | High-concurrency, I/O-bound workloads | Simple async integration |
+| Streaming support | Yes | Yes |

docs/en/learn/streaming-crew-execution.mdx

@@ -95,7 +95,11 @@ print(f"Final result: {streaming.result.raw}")
 
 ## Asynchronous Streaming
 
-For async applications, use `kickoff_async()` with async iteration:
+For async applications, you can use either `akickoff()` (native async) or `kickoff_async()` (thread-based) with async iteration:
+
+### Native Async with `akickoff()`
+
+The `akickoff()` method provides true native async execution throughout the entire chain:
 
 ```python Code
 import asyncio
@@ -107,7 +111,35 @@ async def stream_crew():
         stream=True
     )
 
-    # Start async streaming
+    # Start native async streaming
+    streaming = await crew.akickoff(inputs={"topic": "AI"})
+
+    # Async iteration over chunks
+    async for chunk in streaming:
+        print(chunk.content, end="", flush=True)
+
+    # Access final result
+    result = streaming.result
+    print(f"\n\nFinal output: {result.raw}")
+
+asyncio.run(stream_crew())
+```
+
+### Thread-Based Async with `kickoff_async()`
+
+For simpler async integration or backward compatibility:
+
+```python Code
+import asyncio
+
+async def stream_crew():
+    crew = Crew(
+        agents=[researcher],
+        tasks=[task],
+        stream=True
+    )
+
+    # Start thread-based async streaming
     streaming = await crew.kickoff_async(inputs={"topic": "AI"})
 
     # Async iteration over chunks
@@ -121,6 +153,10 @@ async def stream_crew():
 asyncio.run(stream_crew())
 ```
 
+<Note>
+For high-concurrency workloads, `akickoff()` is recommended as it uses native async for task execution, memory operations, and knowledge retrieval. See the [Kickoff Crew Asynchronously](/en/learn/kickoff-async) guide for more details.
+</Note>
+
 ## Streaming with kickoff_for_each
 
 When executing a crew for multiple inputs with `kickoff_for_each()`, streaming works differently depending on whether you use sync or async:

docs/en/tools/integration/mergeagenthandlertool.mdx

@@ -0,0 +1,367 @@
+---
+title: Merge Agent Handler Tool
+description: Enables CrewAI agents to securely access third-party integrations like Linear, GitHub, Slack, and more through Merge's Agent Handler platform
+icon: diagram-project
+mode: "wide"
+---
+
+# `MergeAgentHandlerTool`
+
+The `MergeAgentHandlerTool` enables CrewAI agents to securely access third-party integrations through [Merge's Agent Handler](https://www.merge.dev/products/merge-agent-handler) platform. Agent Handler provides pre-built, secure connectors to popular tools like Linear, GitHub, Slack, Notion, and hundreds more—all with built-in authentication, permissions, and monitoring.
+
+## Installation
+
+```bash
+uv pip install 'crewai[tools]'
+```
+
+## Requirements
+
+- Merge Agent Handler account with a configured Tool Pack
+- Agent Handler API key
+- At least one registered user linked to your Tool Pack
+- Third-party integrations configured in your Tool Pack
+
+## Getting Started with Agent Handler
+
+1. **Sign up** for a Merge Agent Handler account at [ah.merge.dev/signup](https://ah.merge.dev/signup)
+2. **Create a Tool Pack** and configure the integrations you need
+3. **Register users** who will authenticate with the third-party services
+4. **Get your API key** from the Agent Handler dashboard
+5. **Set environment variable**: `export AGENT_HANDLER_API_KEY='your-key-here'`
+6. **Start building** with the MergeAgentHandlerTool in CrewAI
+
+## Notes
+
+- Tool Pack IDs and Registered User IDs can be found in your Agent Handler dashboard or created via API
+- The tool uses the Model Context Protocol (MCP) for communication with Agent Handler
+- Session IDs are automatically generated but can be customized for context persistence
+- All tool calls are logged and auditable through the Agent Handler platform
+- Tool parameters are dynamically discovered from the Agent Handler API and validated automatically
+
+## Usage
+
+### Single Tool Usage
+
+Here's how to use a specific tool from your Tool Pack:
+
+```python {2, 4-9}
+from crewai import Agent, Task, Crew
+from crewai_tools import MergeAgentHandlerTool
+
+# Create a tool for Linear issue creation
+linear_create_tool = MergeAgentHandlerTool.from_tool_name(
+    tool_name="linear__create_issue",
+    tool_pack_id="134e0111-0f67-44f6-98f0-597000290bb3",
+    registered_user_id="91b2b905-e866-40c8-8be2-efe53827a0aa"
+)
+
+# Create a CrewAI agent that uses the tool
+project_manager = Agent(
+    role='Project Manager',
+    goal='Manage project tasks and issues efficiently',
+    backstory='I am an expert at tracking project work and creating actionable tasks.',
+    tools=[linear_create_tool],
+    verbose=True
+)
+
+# Create a task for the agent
+create_issue_task = Task(
+    description="Create a new high-priority issue in Linear titled 'Implement user authentication' with a detailed description of the requirements.",
+    agent=project_manager,
+    expected_output="Confirmation that the issue was created with its ID"
+)
+
+# Create a crew with the agent
+crew = Crew(
+    agents=[project_manager],
+    tasks=[create_issue_task],
+    verbose=True
+)
+
+# Run the crew
+result = crew.kickoff()
+print(result)
+```
+
+### Loading Multiple Tools from a Tool Pack
+
+You can load all available tools from your Tool Pack at once:
+
+```python {2, 4-8}
+from crewai import Agent, Task, Crew
+from crewai_tools import MergeAgentHandlerTool
+
+# Load all tools from the Tool Pack
+tools = MergeAgentHandlerTool.from_tool_pack(
+    tool_pack_id="134e0111-0f67-44f6-98f0-597000290bb3",
+    registered_user_id="91b2b905-e866-40c8-8be2-efe53827a0aa"
+)
+
+# Create an agent with access to all tools
+automation_expert = Agent(
+    role='Automation Expert',
+    goal='Automate workflows across multiple platforms',
+    backstory='I can work with any tool in the toolbox to get things done.',
+    tools=tools,
+    verbose=True
+)
+
+automation_task = Task(
+    description="Check for any high-priority issues in Linear and post a summary to Slack.",
+    agent=automation_expert
+)
+
+crew = Crew(
+    agents=[automation_expert],
+    tasks=[automation_task],
+    verbose=True
+)
+
+result = crew.kickoff()
+```
+
+### Loading Specific Tools Only
+
+Load only the tools you need:
+
+```python {2, 4-10}
+from crewai import Agent, Task, Crew
+from crewai_tools import MergeAgentHandlerTool
+
+# Load specific tools from the Tool Pack
+selected_tools = MergeAgentHandlerTool.from_tool_pack(
+    tool_pack_id="134e0111-0f67-44f6-98f0-597000290bb3",
+    registered_user_id="91b2b905-e866-40c8-8be2-efe53827a0aa",
+    tool_names=["linear__create_issue", "linear__get_issues", "slack__post_message"]
+)
+
+developer_assistant = Agent(
+    role='Developer Assistant',
+    goal='Help developers track and communicate about their work',
+    backstory='I help developers stay organized and keep the team informed.',
+    tools=selected_tools,
+    verbose=True
+)
+
+daily_update_task = Task(
+    description="Get all issues assigned to the current user in Linear and post a summary to the #dev-updates Slack channel.",
+    agent=developer_assistant
+)
+
+crew = Crew(
+    agents=[developer_assistant],
+    tasks=[daily_update_task],
+    verbose=True
+)
+
+result = crew.kickoff()
+```
+
+## Tool Arguments
+
+### `from_tool_name()` Method
+
+| Argument | Type | Required | Default | Description |
+|:---------|:-----|:---------|:--------|:------------|
+| **tool_name** | `str` | Yes | None | Name of the specific tool to use (e.g., "linear__create_issue") |
+| **tool_pack_id** | `str` | Yes | None | UUID of your Agent Handler Tool Pack |
+| **registered_user_id** | `str` | Yes | None | UUID or origin_id of the registered user |
+| **base_url** | `str` | No | "https://ah-api.merge.dev" | Base URL for Agent Handler API |
+| **session_id** | `str` | No | Auto-generated | MCP session ID for maintaining context |
+
+### `from_tool_pack()` Method
+
+| Argument | Type | Required | Default | Description |
+|:---------|:-----|:---------|:--------|:------------|
+| **tool_pack_id** | `str` | Yes | None | UUID of your Agent Handler Tool Pack |
+| **registered_user_id** | `str` | Yes | None | UUID or origin_id of the registered user |
+| **tool_names** | `list[str]` | No | None | Specific tool names to load. If None, loads all available tools |
+| **base_url** | `str` | No | "https://ah-api.merge.dev" | Base URL for Agent Handler API |
+
+## Environment Variables
+
+```bash
+AGENT_HANDLER_API_KEY=your_api_key_here  # Required for authentication
+```
+
+## Advanced Usage
+
+### Multi-Agent Workflow with Different Tool Access
+
+```python {2, 4-20}
+from crewai import Agent, Task, Crew, Process
+from crewai_tools import MergeAgentHandlerTool
+
+# Create specialized tools for different agents
+github_tools = MergeAgentHandlerTool.from_tool_pack(
+    tool_pack_id="134e0111-0f67-44f6-98f0-597000290bb3",
+    registered_user_id="91b2b905-e866-40c8-8be2-efe53827a0aa",
+    tool_names=["github__create_pull_request", "github__get_pull_requests"]
+)
+
+linear_tools = MergeAgentHandlerTool.from_tool_pack(
+    tool_pack_id="134e0111-0f67-44f6-98f0-597000290bb3",
+    registered_user_id="91b2b905-e866-40c8-8be2-efe53827a0aa",
+    tool_names=["linear__create_issue", "linear__update_issue"]
+)
+
+slack_tool = MergeAgentHandlerTool.from_tool_name(
+    tool_name="slack__post_message",
+    tool_pack_id="134e0111-0f67-44f6-98f0-597000290bb3",
+    registered_user_id="91b2b905-e866-40c8-8be2-efe53827a0aa"
+)
+
+# Create specialized agents
+code_reviewer = Agent(
+    role='Code Reviewer',
+    goal='Review pull requests and ensure code quality',
+    backstory='I am an expert at reviewing code changes and providing constructive feedback.',
+    tools=github_tools
+)
+
+task_manager = Agent(
+    role='Task Manager',
+    goal='Track and update project tasks based on code changes',
+    backstory='I keep the project board up to date with the latest development progress.',
+    tools=linear_tools
+)
+
+communicator = Agent(
+    role='Team Communicator',
+    goal='Keep the team informed about important updates',
+    backstory='I make sure everyone knows what is happening in the project.',
+    tools=[slack_tool]
+)
+
+# Create sequential tasks
+review_task = Task(
+    description="Review all open pull requests in the 'api-service' repository and identify any that need attention.",
+    agent=code_reviewer,
+    expected_output="List of pull requests that need review or have issues"
+)
+
+update_task = Task(
+    description="Update Linear issues based on the pull request review findings. Mark completed PRs as done.",
+    agent=task_manager,
+    expected_output="Summary of updated Linear issues"
+)
+
+notify_task = Task(
+    description="Post a summary of today's code review and task updates to the #engineering Slack channel.",
+    agent=communicator,
+    expected_output="Confirmation that the message was posted"
+)
+
+# Create a crew with sequential processing
+crew = Crew(
+    agents=[code_reviewer, task_manager, communicator],
+    tasks=[review_task, update_task, notify_task],
+    process=Process.sequential,
+    verbose=True
+)
+
+result = crew.kickoff()
+```
+
+### Custom Session Management
+
+Maintain context across multiple tool calls using session IDs:
+
+```python {2, 4-17}
+from crewai import Agent, Task, Crew
+from crewai_tools import MergeAgentHandlerTool
+
+# Create tools with the same session ID to maintain context
+session_id = "project-sprint-planning-2024"
+
+create_tool = MergeAgentHandlerTool(
+    name="linear_create_issue",
+    description="Creates a new issue in Linear",
+    tool_name="linear__create_issue",
+    tool_pack_id="134e0111-0f67-44f6-98f0-597000290bb3",
+    registered_user_id="91b2b905-e866-40c8-8be2-efe53827a0aa",
+    session_id=session_id
+)
+
+update_tool = MergeAgentHandlerTool(
+    name="linear_update_issue",
+    description="Updates an existing issue in Linear",
+    tool_name="linear__update_issue",
+    tool_pack_id="134e0111-0f67-44f6-98f0-597000290bb3",
+    registered_user_id="91b2b905-e866-40c8-8be2-efe53827a0aa",
+    session_id=session_id
+)
+
+sprint_planner = Agent(
+    role='Sprint Planner',
+    goal='Plan and organize sprint tasks',
+    backstory='I help teams plan effective sprints with well-defined tasks.',
+    tools=[create_tool, update_tool],
+    verbose=True
+)
+
+planning_task = Task(
+    description="Create 5 sprint tasks for the authentication feature and set their priorities based on dependencies.",
+    agent=sprint_planner
+)
+
+crew = Crew(
+    agents=[sprint_planner],
+    tasks=[planning_task],
+    verbose=True
+)
+
+result = crew.kickoff()
+```
+
+## Use Cases
+
+### Unified Integration Access
+- Access hundreds of third-party tools through a single unified API without managing multiple SDKs
+- Enable agents to work with Linear, GitHub, Slack, Notion, Jira, Asana, and more from one integration point
+- Reduce integration complexity by letting Agent Handler manage authentication and API versioning
+
+### Secure Enterprise Workflows
+- Leverage built-in authentication and permission management for all third-party integrations
+- Maintain enterprise security standards with centralized access control and audit logging
+- Enable agents to access company tools without exposing API keys or credentials in code
+
+### Cross-Platform Automation
+- Build workflows that span multiple platforms (e.g., create GitHub issues from Linear tasks, sync Notion pages to Slack)
+- Enable seamless data flow between different tools in your tech stack
+- Create intelligent automation that understands context across different platforms
+
+### Dynamic Tool Discovery
+- Load all available tools at runtime without hardcoding integration logic
+- Enable agents to discover and use new tools as they're added to your Tool Pack
+- Build flexible agents that can adapt to changing tool availability
+
+### User-Specific Tool Access
+- Different users can have different tool permissions and access levels
+- Enable multi-tenant workflows where agents act on behalf of specific users
+- Maintain proper attribution and permissions for all tool actions
+
+## Available Integrations
+
+Merge Agent Handler supports hundreds of integrations across multiple categories:
+
+- **Project Management**: Linear, Jira, Asana, Monday.com, ClickUp
+- **Code Management**: GitHub, GitLab, Bitbucket
+- **Communication**: Slack, Microsoft Teams, Discord
+- **Documentation**: Notion, Confluence, Google Docs
+- **CRM**: Salesforce, HubSpot, Pipedrive
+- **And many more...**
+
+Visit the [Merge Agent Handler documentation](https://docs.ah.merge.dev/) for a complete list of available integrations.
+
+## Error Handling
+
+The tool provides comprehensive error handling:
+
+- **Authentication Errors**: Invalid or missing API keys
+- **Permission Errors**: User lacks permission for the requested action
+- **API Errors**: Issues communicating with Agent Handler or third-party services
+- **Validation Errors**: Invalid parameters passed to tool methods
+
+All errors are wrapped in `MergeAgentHandlerToolError` for consistent error handling.

docs/en/tools/integration/overview.mdx

@@ -10,6 +10,10 @@ Integration tools let your agents hand off work to other automation platforms an
 ## **Available Tools**
 
 <CardGroup cols={2}>
+  <Card title="Merge Agent Handler Tool" icon="diagram-project" href="/en/tools/integration/mergeagenthandlertool">
+    Securely access hundreds of third-party tools like Linear, GitHub, Slack, and more through Merge's unified API.
+  </Card>
+
   <Card title="CrewAI Run Automation Tool" icon="robot" href="/en/tools/integration/crewaiautomationtool">
     Invoke live CrewAI Platform automations, pass custom inputs, and poll for results directly from your agent.
   </Card>

docs/ko/concepts/crews.mdx

@@ -33,6 +33,7 @@ crewAI에서 crew는 일련의 작업을 달성하기 위해 함께 협력하는
 | **Planning** *(선택사항)*              | `planning`             | Crew에 계획 수립 기능을 추가. 활성화하면 각 Crew 반복 전에 모든 Crew 데이터를 AgentPlanner로 전송하여 작업계획을 세우고, 이 계획이 각 작업 설명에 추가됨.                                                     |
 | **Planning LLM** *(선택사항)*          | `planning_llm`         | 계획 과정에서 AgentPlanner가 사용하는 언어 모델.                                                                                                                                                                 |
 | **Knowledge Sources** _(선택사항)_     | `knowledge_sources`    | crew 수준에서 사용 가능한 지식 소스. 모든 agent가 접근 가능.                                                                                                                                                             |
+| **Stream** _(선택사항)_                | `stream`               | 스트리밍 출력을 활성화하여 crew 실행 중 실시간 업데이트를 받을 수 있습니다. 청크를 반복할 수 있는 `CrewStreamingOutput` 객체를 반환합니다. 기본값은 `False`.                                                                    |
 
 <Tip>
 **Crew Max RPM**: `max_rpm` 속성은 crew가 분당 처리할 수 있는 최대 요청 수를 설정하며, 개별 agent의 `max_rpm` 설정을 crew 단위로 지정할 경우 오버라이드합니다.
@@ -306,12 +307,27 @@ print(result)
 
 ### Crew를 시작하는 다양한 방법
 
-crew가 구성되면, 적절한 시작 방법으로 workflow를 시작하세요. CrewAI는 kickoff 프로세스를 더 잘 제어할 수 있도록 여러 방법을 제공합니다: `kickoff()`, `kickoff_for_each()`, `kickoff_async()`, 그리고 `kickoff_for_each_async()`.
+crew가 구성되면, 적절한 시작 방법으로 workflow를 시작하세요. CrewAI는 kickoff 프로세스를 더 잘 제어할 수 있도록 여러 방법을 제공합니다.
+
+#### 동기 메서드
 
 - `kickoff()`: 정의된 process flow에 따라 실행 프로세스를 시작합니다.
 - `kickoff_for_each()`: 입력 이벤트나 컬렉션 내 각 항목에 대해 순차적으로 task를 실행합니다.
-- `kickoff_async()`: 비동기적으로 workflow를 시작합니다.
-- `kickoff_for_each_async()`: 입력 이벤트나 각 항목에 대해 비동기 처리를 활용하여 task를 동시에 실행합니다.
+
+#### 비동기 메서드
+
+CrewAI는 비동기 실행을 위해 두 가지 접근 방식을 제공합니다:
+
+| 메서드 | 타입 | 설명 |
+|--------|------|-------------|
+| `akickoff()` | 네이티브 async | 전체 실행 체인에서 진정한 async/await 사용 |
+| `akickoff_for_each()` | 네이티브 async | 리스트의 각 입력에 대해 네이티브 async 실행 |
+| `kickoff_async()` | 스레드 기반 | 동기 실행을 `asyncio.to_thread`로 래핑 |
+| `kickoff_for_each_async()` | 스레드 기반 | 리스트의 각 입력에 대해 스레드 기반 async |
+
+<Note>
+고동시성 워크로드의 경우 `akickoff()` 및 `akickoff_for_each()`가 권장됩니다. 이들은 작업 실행, 메모리 작업, 지식 검색에 네이티브 async를 사용합니다.
+</Note>
 
 ```python Code
 # Start the crew's task execution
@@ -324,19 +340,53 @@ results = my_crew.kickoff_for_each(inputs=inputs_array)
 for result in results:
     print(result)
 
-# Example of using kickoff_async
+# Example of using native async with akickoff
+inputs = {'topic': 'AI in healthcare'}
+async_result = await my_crew.akickoff(inputs=inputs)
+print(async_result)
+
+# Example of using native async with akickoff_for_each
+inputs_array = [{'topic': 'AI in healthcare'}, {'topic': 'AI in finance'}]
+async_results = await my_crew.akickoff_for_each(inputs=inputs_array)
+for async_result in async_results:
+    print(async_result)
+
+# Example of using thread-based kickoff_async
 inputs = {'topic': 'AI in healthcare'}
 async_result = await my_crew.kickoff_async(inputs=inputs)
 print(async_result)
 
-# Example of using kickoff_for_each_async
+# Example of using thread-based kickoff_for_each_async
 inputs_array = [{'topic': 'AI in healthcare'}, {'topic': 'AI in finance'}]
 async_results = await my_crew.kickoff_for_each_async(inputs=inputs_array)
 for async_result in async_results:
     print(async_result)
 ```
 
-이러한 메서드는 crew 내에서 task를 관리하고 실행하는 데 유연성을 제공하며, 동기 및 비동기 workflow 모두 필요에 맞게 사용할 수 있도록 지원합니다.
+이러한 메서드는 crew 내에서 task를 관리하고 실행하는 데 유연성을 제공하며, 동기 및 비동기 workflow 모두 필요에 맞게 사용할 수 있도록 지원합니다. 자세한 비동기 예제는 [Crew 비동기 시작](/ko/learn/kickoff-async) 가이드를 참조하세요.
+
+### 스트리밍 Crew 실행
+
+crew 실행을 실시간으로 확인하려면 스트리밍을 활성화하여 출력이 생성되는 대로 받을 수 있습니다:
+
+```python Code
+# 스트리밍 활성화
+crew = Crew(
+    agents=[researcher],
+    tasks=[task],
+    stream=True
+)
+
+# 스트리밍 출력을 반복
+streaming = crew.kickoff(inputs={"topic": "AI"})
+for chunk in streaming:
+    print(chunk.content, end="", flush=True)
+
+# 최종 결과 접근
+result = streaming.result
+```
+
+스트리밍에 대한 자세한 내용은 [스트리밍 Crew 실행](/ko/learn/streaming-crew-execution) 가이드를 참조하세요.
 
 ### 특정 Task에서 다시 실행하기
 

docs/ko/concepts/memory.mdx

@@ -515,8 +515,7 @@ crew = Crew(
         "provider": "huggingface",
         "config": {
             "api_key": "your-hf-token",  # Optional for public models
-            "model": "sentence-transformers/all-MiniLM-L6-v2",
-            "api_url": "https://api-inference.huggingface.co"  # or your custom endpoint
+            "model": "sentence-transformers/all-MiniLM-L6-v2"
         }
     }
 )

docs/ko/learn/create-custom-tools.mdx

@@ -63,5 +63,55 @@ def my_cache_strategy(arguments: dict, result: str) -> bool:
 cached_tool.cache_function = my_cache_strategy
 ```
 
+### 비동기 도구 생성하기
+
+CrewAI는 논블로킹 I/O 작업을 위한 비동기 도구를 지원합니다. 이는 HTTP 요청, 데이터베이스 쿼리 또는 기타 I/O 바운드 작업이 필요한 경우에 유용합니다.
+
+#### `@tool` 데코레이터와 비동기 함수 사용하기
+
+비동기 도구를 만드는 가장 간단한 방법은 `@tool` 데코레이터와 async 함수를 사용하는 것입니다:
+
+```python Code
+import aiohttp
+from crewai.tools import tool
+
+@tool("Async Web Fetcher")
+async def fetch_webpage(url: str) -> str:
+    """Fetch content from a webpage asynchronously."""
+    async with aiohttp.ClientSession() as session:
+        async with session.get(url) as response:
+            return await response.text()
+```
+
+#### 비동기 지원으로 `BaseTool` 서브클래싱하기
+
+더 많은 제어를 위해 `BaseTool`을 상속하고 `_run`(동기) 및 `_arun`(비동기) 메서드를 모두 구현할 수 있습니다:
+
+```python Code
+import requests
+import aiohttp
+from crewai.tools import BaseTool
+from pydantic import BaseModel, Field
+
+class WebFetcherInput(BaseModel):
+    """Input schema for WebFetcher."""
+    url: str = Field(..., description="The URL to fetch")
+
+class WebFetcherTool(BaseTool):
+    name: str = "Web Fetcher"
+    description: str = "Fetches content from a URL"
+    args_schema: type[BaseModel] = WebFetcherInput
+
+    def _run(self, url: str) -> str:
+        """Synchronous implementation."""
+        return requests.get(url).text
+
+    async def _arun(self, url: str) -> str:
+        """Asynchronous implementation for non-blocking I/O."""
+        async with aiohttp.ClientSession() as session:
+            async with session.get(url) as response:
+                return await response.text()
+```
+
 이 가이드라인을 준수하고 새로운 기능과 협업 도구를 도구 생성 및 관리 프로세스에 통합함으로써,
-CrewAI 프레임워크의 모든 기능을 활용할 수 있으며, AI agent의 개발 경험과 효율성을 모두 높일 수 있습니다.
+CrewAI 프레임워크의 모든 기능을 활용할 수 있으며, AI agent의 개발 경험과 효율성을 모두 높일 수 있습니다.

docs/ko/learn/kickoff-async.mdx

@@ -7,17 +7,28 @@ mode: "wide"
 
 ## 소개
 
-CrewAI는 crew를 비동기적으로 시작할 수 있는 기능을 제공합니다. 이를 통해 crew 실행을 블로킹(blocking) 없이 시작할 수 있습니다.  
+CrewAI는 crew를 비동기적으로 시작할 수 있는 기능을 제공합니다. 이를 통해 crew 실행을 블로킹(blocking) 없이 시작할 수 있습니다.
 이 기능은 여러 개의 crew를 동시에 실행하거나 crew가 실행되는 동안 다른 작업을 수행해야 할 때 특히 유용합니다.
 
-## 비동기 Crew 실행
+CrewAI는 비동기 실행을 위해 두 가지 접근 방식을 제공합니다:
 
-Crew를 비동기적으로 시작하려면 `kickoff_async()` 메서드를 사용하세요. 이 메서드는 별도의 스레드에서 crew 실행을 시작하여, 메인 스레드가 다른 작업을 계속 실행할 수 있도록 합니다.
+| 메서드 | 타입 | 설명 |
+|--------|------|-------------|
+| `akickoff()` | 네이티브 async | 전체 실행 체인에서 진정한 async/await 사용 |
+| `kickoff_async()` | 스레드 기반 | 동기 실행을 `asyncio.to_thread`로 래핑 |
+
+<Note>
+고동시성 워크로드의 경우 `akickoff()`가 권장됩니다. 이는 작업 실행, 메모리 작업, 지식 검색에 네이티브 async를 사용합니다.
+</Note>
+
+## `akickoff()`를 사용한 네이티브 비동기 실행
+
+`akickoff()` 메서드는 작업 실행, 메모리 작업, 지식 쿼리를 포함한 전체 실행 체인에서 async/await를 사용하여 진정한 네이티브 비동기 실행을 제공합니다.
 
 ### 메서드 시그니처
 
 ```python Code
-def kickoff_async(self, inputs: dict) -> CrewOutput:
+async def akickoff(self, inputs: dict) -> CrewOutput:
 ```
 
 ### 매개변수
@@ -28,23 +39,13 @@ def kickoff_async(self, inputs: dict) -> CrewOutput:
 
 - `CrewOutput`: crew 실행 결과를 나타내는 객체입니다.
 
-## 잠재적 사용 사례
-
-- **병렬 콘텐츠 생성**: 여러 개의 독립적인 crew를 비동기적으로 시작하여, 각 crew가 다른 주제에 대한 콘텐츠 생성을 담당합니다. 예를 들어, 한 crew는 AI 트렌드에 대한 기사 조사 및 초안을 작성하는 반면, 또 다른 crew는 신제품 출시와 관련된 소셜 미디어 게시물을 생성할 수 있습니다. 각 crew는 독립적으로 운영되므로 콘텐츠 생산을 효율적으로 확장할 수 있습니다.
-
-- **동시 시장 조사 작업**: 여러 crew를 비동기적으로 시작하여 시장 조사를 병렬로 수행합니다. 한 crew는 업계 동향을 분석하고, 또 다른 crew는 경쟁사 전략을 조사하며, 또 다른 crew는 소비자 감정을 평가할 수 있습니다. 각 crew는 독립적으로 자신의 작업을 완료하므로 더 빠르고 포괄적인 인사이트를 얻을 수 있습니다.
-
-- **독립적인 여행 계획 모듈**: 각각 독립적으로 여행의 다양한 측면을 계획하도록 crew를 따로 실행합니다. 한 crew는 항공편 옵션을, 다른 crew는 숙박을, 세 번째 crew는 활동 계획을 담당할 수 있습니다. 각 crew는 비동기적으로 작업하므로 여행의 다양한 요소를 동시에 그리고 독립적으로 더 빠르게 계획할 수 있습니다.
-
-## 예시: 단일 비동기 crew 실행
-
-다음은 asyncio를 사용하여 crew를 비동기적으로 시작하고 결과를 await하는 방법의 예시입니다:
+### 예시: 네이티브 비동기 Crew 실행
 
 ```python Code
 import asyncio
 from crewai import Crew, Agent, Task
 
-# Create an agent with code execution enabled
+# 에이전트 생성
 coding_agent = Agent(
     role="Python Data Analyst",
     goal="Analyze data and provide insights using Python",
@@ -52,37 +53,165 @@ coding_agent = Agent(
     allow_code_execution=True
 )
 
-# Create a task that requires code execution
+# 작업 생성
 data_analysis_task = Task(
     description="Analyze the given dataset and calculate the average age of participants. Ages: {ages}",
     agent=coding_agent,
     expected_output="The average age of the participants."
 )
 
-# Create a crew and add the task
+# Crew 생성
 analysis_crew = Crew(
     agents=[coding_agent],
     tasks=[data_analysis_task]
 )
 
-# Async function to kickoff the crew asynchronously
-async def async_crew_execution():
-    result = await analysis_crew.kickoff_async(inputs={"ages": [25, 30, 35, 40, 45]})
+# 네이티브 비동기 실행
+async def main():
+    result = await analysis_crew.akickoff(inputs={"ages": [25, 30, 35, 40, 45]})
     print("Crew Result:", result)
 
-# Run the async function
-asyncio.run(async_crew_execution())
+asyncio.run(main())
 ```
 
-## 예제: 다중 비동기 Crew 실행
+### 예시: 여러 네이티브 비동기 Crew
 
-이 예제에서는 여러 Crew를 비동기적으로 시작하고 `asyncio.gather()`를 사용하여 모두 완료될 때까지 기다리는 방법을 보여줍니다:
+`asyncio.gather()`를 사용하여 네이티브 async로 여러 crew를 동시에 실행:
+
+```python Code
+import asyncio
+from crewai import Crew, Agent, Task
+
+coding_agent = Agent(
+    role="Python Data Analyst",
+    goal="Analyze data and provide insights using Python",
+    backstory="You are an experienced data analyst with strong Python skills.",
+    allow_code_execution=True
+)
+
+task_1 = Task(
+    description="Analyze the first dataset and calculate the average age. Ages: {ages}",
+    agent=coding_agent,
+    expected_output="The average age of the participants."
+)
+
+task_2 = Task(
+    description="Analyze the second dataset and calculate the average age. Ages: {ages}",
+    agent=coding_agent,
+    expected_output="The average age of the participants."
+)
+
+crew_1 = Crew(agents=[coding_agent], tasks=[task_1])
+crew_2 = Crew(agents=[coding_agent], tasks=[task_2])
+
+async def main():
+    results = await asyncio.gather(
+        crew_1.akickoff(inputs={"ages": [25, 30, 35, 40, 45]}),
+        crew_2.akickoff(inputs={"ages": [20, 22, 24, 28, 30]})
+    )
+
+    for i, result in enumerate(results, 1):
+        print(f"Crew {i} Result:", result)
+
+asyncio.run(main())
+```
+
+### 예시: 여러 입력에 대한 네이티브 비동기
+
+`akickoff_for_each()`를 사용하여 네이티브 async로 여러 입력에 대해 crew를 동시에 실행:
+
+```python Code
+import asyncio
+from crewai import Crew, Agent, Task
+
+coding_agent = Agent(
+    role="Python Data Analyst",
+    goal="Analyze data and provide insights using Python",
+    backstory="You are an experienced data analyst with strong Python skills.",
+    allow_code_execution=True
+)
+
+data_analysis_task = Task(
+    description="Analyze the dataset and calculate the average age. Ages: {ages}",
+    agent=coding_agent,
+    expected_output="The average age of the participants."
+)
+
+analysis_crew = Crew(
+    agents=[coding_agent],
+    tasks=[data_analysis_task]
+)
+
+async def main():
+    datasets = [
+        {"ages": [25, 30, 35, 40, 45]},
+        {"ages": [20, 22, 24, 28, 30]},
+        {"ages": [30, 35, 40, 45, 50]}
+    ]
+
+    results = await analysis_crew.akickoff_for_each(datasets)
+
+    for i, result in enumerate(results, 1):
+        print(f"Dataset {i} Result:", result)
+
+asyncio.run(main())
+```
+
+## `kickoff_async()`를 사용한 스레드 기반 비동기
+
+`kickoff_async()` 메서드는 동기 `kickoff()`를 스레드로 래핑하여 비동기 실행을 제공합니다. 이는 더 간단한 비동기 통합이나 하위 호환성에 유용합니다.
+
+### 메서드 시그니처
+
+```python Code
+async def kickoff_async(self, inputs: dict) -> CrewOutput:
+```
+
+### 매개변수
+
+- `inputs` (dict): 작업에 필요한 입력 데이터를 포함하는 딕셔너리입니다.
+
+### 반환
+
+- `CrewOutput`: crew 실행 결과를 나타내는 객체입니다.
+
+### 예시: 스레드 기반 비동기 실행
+
+```python Code
+import asyncio
+from crewai import Crew, Agent, Task
+
+coding_agent = Agent(
+    role="Python Data Analyst",
+    goal="Analyze data and provide insights using Python",
+    backstory="You are an experienced data analyst with strong Python skills.",
+    allow_code_execution=True
+)
+
+data_analysis_task = Task(
+    description="Analyze the given dataset and calculate the average age of participants. Ages: {ages}",
+    agent=coding_agent,
+    expected_output="The average age of the participants."
+)
+
+analysis_crew = Crew(
+    agents=[coding_agent],
+    tasks=[data_analysis_task]
+)
+
+async def async_crew_execution():
+    result = await analysis_crew.kickoff_async(inputs={"ages": [25, 30, 35, 40, 45]})
+    print("Crew Result:", result)
+
+asyncio.run(async_crew_execution())
+```
+
+### 예시: 여러 스레드 기반 비동기 Crew
 
 ```python Code
 import asyncio
 from crewai import Crew, Agent, Task
 
-# Create an agent with code execution enabled
 coding_agent = Agent(
     role="Python Data Analyst",
     goal="Analyze data and provide insights using Python",
@@ -90,7 +219,6 @@ coding_agent = Agent(
     allow_code_execution=True
 )
 
-# Create tasks that require code execution
 task_1 = Task(
     description="Analyze the first dataset and calculate the average age of participants. Ages: {ages}",
     agent=coding_agent,
@@ -103,22 +231,76 @@ task_2 = Task(
     expected_output="The average age of the participants."
 )
 
-# Create two crews and add tasks
 crew_1 = Crew(agents=[coding_agent], tasks=[task_1])
 crew_2 = Crew(agents=[coding_agent], tasks=[task_2])
 
-# Async function to kickoff multiple crews asynchronously and wait for all to finish
 async def async_multiple_crews():
-    # Create coroutines for concurrent execution
     result_1 = crew_1.kickoff_async(inputs={"ages": [25, 30, 35, 40, 45]})
     result_2 = crew_2.kickoff_async(inputs={"ages": [20, 22, 24, 28, 30]})
 
-    # Wait for both crews to finish
     results = await asyncio.gather(result_1, result_2)
 
     for i, result in enumerate(results, 1):
         print(f"Crew {i} Result:", result)
 
-# Run the async function
 asyncio.run(async_multiple_crews())
-```
+```
+
+## 비동기 스트리밍
+
+두 비동기 메서드 모두 crew에 `stream=True`가 설정된 경우 스트리밍을 지원합니다:
+
+```python Code
+import asyncio
+from crewai import Crew, Agent, Task
+
+agent = Agent(
+    role="Researcher",
+    goal="Research and summarize topics",
+    backstory="You are an expert researcher."
+)
+
+task = Task(
+    description="Research the topic: {topic}",
+    agent=agent,
+    expected_output="A comprehensive summary of the topic."
+)
+
+crew = Crew(
+    agents=[agent],
+    tasks=[task],
+    stream=True  # 스트리밍 활성화
+)
+
+async def main():
+    streaming_output = await crew.akickoff(inputs={"topic": "AI trends in 2024"})
+
+    # 스트리밍 청크에 대한 비동기 반복
+    async for chunk in streaming_output:
+        print(f"Chunk: {chunk.content}")
+
+    # 스트리밍 완료 후 최종 결과 접근
+    result = streaming_output.result
+    print(f"Final result: {result.raw}")
+
+asyncio.run(main())
+```
+
+## 잠재적 사용 사례
+
+- **병렬 콘텐츠 생성**: 여러 개의 독립적인 crew를 비동기적으로 시작하여, 각 crew가 다른 주제에 대한 콘텐츠 생성을 담당합니다. 예를 들어, 한 crew는 AI 트렌드에 대한 기사 조사 및 초안을 작성하는 반면, 또 다른 crew는 신제품 출시와 관련된 소셜 미디어 게시물을 생성할 수 있습니다.
+
+- **동시 시장 조사 작업**: 여러 crew를 비동기적으로 시작하여 시장 조사를 병렬로 수행합니다. 한 crew는 업계 동향을 분석하고, 또 다른 crew는 경쟁사 전략을 조사하며, 또 다른 crew는 소비자 감정을 평가할 수 있습니다.
+
+- **독립적인 여행 계획 모듈**: 각각 독립적으로 여행의 다양한 측면을 계획하도록 crew를 따로 실행합니다. 한 crew는 항공편 옵션을, 다른 crew는 숙박을, 세 번째 crew는 활동 계획을 담당할 수 있습니다.
+
+## `akickoff()`와 `kickoff_async()` 선택하기
+
+| 기능 | `akickoff()` | `kickoff_async()` |
+|---------|--------------|-------------------|
+| 실행 모델 | 네이티브 async/await | 스레드 기반 래퍼 |
+| 작업 실행 | `aexecute_sync()`로 비동기 | 스레드 풀에서 동기 |
+| 메모리 작업 | 비동기 | 스레드 풀에서 동기 |
+| 지식 검색 | 비동기 | 스레드 풀에서 동기 |
+| 적합한 용도 | 고동시성, I/O 바운드 워크로드 | 간단한 비동기 통합 |
+| 스트리밍 지원 | 예 | 예 |

docs/ko/learn/streaming-crew-execution.mdx

@@ -0,0 +1,356 @@
+---
+title: 스트리밍 Crew 실행
+description: CrewAI crew 실행에서 실시간 출력을 스트리밍하기
+icon: wave-pulse
+mode: "wide"
+---
+
+## 소개
+
+CrewAI는 crew 실행 중 실시간 출력을 스트리밍하는 기능을 제공하여, 전체 프로세스가 완료될 때까지 기다리지 않고 결과가 생성되는 대로 표시할 수 있습니다. 이 기능은 대화형 애플리케이션을 구축하거나, 사용자 피드백을 제공하거나, 장시간 실행되는 프로세스를 모니터링할 때 특히 유용합니다.
+
+## 스트리밍 작동 방식
+
+스트리밍이 활성화되면 CrewAI는 LLM 응답과 도구 호출을 실시간으로 캡처하여, 어떤 task와 agent가 실행 중인지에 대한 컨텍스트를 포함한 구조화된 청크로 패키징합니다. 이러한 청크를 실시간으로 반복 처리하고 실행이 완료되면 최종 결과에 접근할 수 있습니다.
+
+## 스트리밍 활성화
+
+스트리밍을 활성화하려면 crew를 생성할 때 `stream` 파라미터를 `True`로 설정하세요:
+
+```python Code
+from crewai import Agent, Crew, Task
+
+# 에이전트와 태스크 생성
+researcher = Agent(
+    role="Research Analyst",
+    goal="Gather comprehensive information on topics",
+    backstory="You are an experienced researcher with excellent analytical skills.",
+)
+
+task = Task(
+    description="Research the latest developments in AI",
+    expected_output="A detailed report on recent AI advancements",
+    agent=researcher,
+)
+
+# 스트리밍 활성화
+crew = Crew(
+    agents=[researcher],
+    tasks=[task],
+    stream=True  # 스트리밍 출력 활성화
+)
+```
+
+## 동기 스트리밍
+
+스트리밍이 활성화된 crew에서 `kickoff()`를 호출하면, 청크가 도착할 때마다 반복 처리할 수 있는 `CrewStreamingOutput` 객체가 반환됩니다:
+
+```python Code
+# 스트리밍 실행 시작
+streaming = crew.kickoff(inputs={"topic": "artificial intelligence"})
+
+# 청크가 도착할 때마다 반복
+for chunk in streaming:
+    print(chunk.content, end="", flush=True)
+
+# 스트리밍 완료 후 최종 결과 접근
+result = streaming.result
+print(f"\n\n최종 출력: {result.raw}")
+```
+
+### 스트림 청크 정보
+
+각 청크는 실행에 대한 풍부한 컨텍스트를 제공합니다:
+
+```python Code
+streaming = crew.kickoff(inputs={"topic": "AI"})
+
+for chunk in streaming:
+    print(f"Task: {chunk.task_name} (인덱스 {chunk.task_index})")
+    print(f"Agent: {chunk.agent_role}")
+    print(f"Content: {chunk.content}")
+    print(f"Type: {chunk.chunk_type}")  # TEXT 또는 TOOL_CALL
+    if chunk.tool_call:
+        print(f"Tool: {chunk.tool_call.tool_name}")
+        print(f"Arguments: {chunk.tool_call.arguments}")
+```
+
+### 스트리밍 결과 접근
+
+`CrewStreamingOutput` 객체는 여러 유용한 속성을 제공합니다:
+
+```python Code
+streaming = crew.kickoff(inputs={"topic": "AI"})
+
+# 청크 반복 및 수집
+for chunk in streaming:
+    print(chunk.content, end="", flush=True)
+
+# 반복 완료 후
+print(f"\n완료됨: {streaming.is_completed}")
+print(f"전체 텍스트: {streaming.get_full_text()}")
+print(f"전체 청크 수: {len(streaming.chunks)}")
+print(f"최종 결과: {streaming.result.raw}")
+```
+
+## 비동기 스트리밍
+
+비동기 애플리케이션의 경우, 비동기 반복과 함께 `akickoff()`(네이티브 async) 또는 `kickoff_async()`(스레드 기반)를 사용할 수 있습니다:
+
+### `akickoff()`를 사용한 네이티브 Async
+
+`akickoff()` 메서드는 전체 체인에서 진정한 네이티브 async 실행을 제공합니다:
+
+```python Code
+import asyncio
+
+async def stream_crew():
+    crew = Crew(
+        agents=[researcher],
+        tasks=[task],
+        stream=True
+    )
+
+    # 네이티브 async 스트리밍 시작
+    streaming = await crew.akickoff(inputs={"topic": "AI"})
+
+    # 청크에 대한 비동기 반복
+    async for chunk in streaming:
+        print(chunk.content, end="", flush=True)
+
+    # 최종 결과 접근
+    result = streaming.result
+    print(f"\n\n최종 출력: {result.raw}")
+
+asyncio.run(stream_crew())
+```
+
+### `kickoff_async()`를 사용한 스레드 기반 Async
+
+더 간단한 async 통합이나 하위 호환성을 위해:
+
+```python Code
+import asyncio
+
+async def stream_crew():
+    crew = Crew(
+        agents=[researcher],
+        tasks=[task],
+        stream=True
+    )
+
+    # 스레드 기반 async 스트리밍 시작
+    streaming = await crew.kickoff_async(inputs={"topic": "AI"})
+
+    # 청크에 대한 비동기 반복
+    async for chunk in streaming:
+        print(chunk.content, end="", flush=True)
+
+    # 최종 결과 접근
+    result = streaming.result
+    print(f"\n\n최종 출력: {result.raw}")
+
+asyncio.run(stream_crew())
+```
+
+<Note>
+고동시성 워크로드의 경우, 태스크 실행, 메모리 작업, 지식 검색에 네이티브 async를 사용하는 `akickoff()`가 권장됩니다. 자세한 내용은 [Crew 비동기 시작](/ko/learn/kickoff-async) 가이드를 참조하세요.
+</Note>
+
+## kickoff_for_each를 사용한 스트리밍
+
+`kickoff_for_each()`로 여러 입력에 대해 crew를 실행할 때, 동기 또는 비동기 여부에 따라 스트리밍이 다르게 작동합니다:
+
+### 동기 kickoff_for_each
+
+동기 `kickoff_for_each()`를 사용하면, 각 입력에 대해 하나씩 `CrewStreamingOutput` 객체의 리스트가 반환됩니다:
+
+```python Code
+crew = Crew(
+    agents=[researcher],
+    tasks=[task],
+    stream=True
+)
+
+inputs_list = [
+    {"topic": "AI in healthcare"},
+    {"topic": "AI in finance"}
+]
+
+# 스트리밍 출력 리스트 반환
+streaming_outputs = crew.kickoff_for_each(inputs=inputs_list)
+
+# 각 스트리밍 출력에 대해 반복
+for i, streaming in enumerate(streaming_outputs):
+    print(f"\n=== 입력 {i + 1} ===")
+    for chunk in streaming:
+        print(chunk.content, end="", flush=True)
+
+    result = streaming.result
+    print(f"\n\n결과 {i + 1}: {result.raw}")
+```
+
+### 비동기 kickoff_for_each_async
+
+비동기 `kickoff_for_each_async()`를 사용하면, 모든 crew의 청크가 동시에 도착하는 대로 반환하는 단일 `CrewStreamingOutput`이 반환됩니다:
+
+```python Code
+import asyncio
+
+async def stream_multiple_crews():
+    crew = Crew(
+        agents=[researcher],
+        tasks=[task],
+        stream=True
+    )
+
+    inputs_list = [
+        {"topic": "AI in healthcare"},
+        {"topic": "AI in finance"}
+    ]
+
+    # 모든 crew에 대한 단일 스트리밍 출력 반환
+    streaming = await crew.kickoff_for_each_async(inputs=inputs_list)
+
+    # 모든 crew의 청크가 생성되는 대로 도착
+    async for chunk in streaming:
+        print(f"[{chunk.task_name}] {chunk.content}", end="", flush=True)
+
+    # 모든 결과 접근
+    results = streaming.results  # CrewOutput 객체 리스트
+    for i, result in enumerate(results):
+        print(f"\n\n결과 {i + 1}: {result.raw}")
+
+asyncio.run(stream_multiple_crews())
+```
+
+## 스트림 청크 타입
+
+청크는 `chunk_type` 필드로 표시되는 다양한 타입을 가질 수 있습니다:
+
+### TEXT 청크
+
+LLM 응답의 표준 텍스트 콘텐츠:
+
+```python Code
+for chunk in streaming:
+    if chunk.chunk_type == StreamChunkType.TEXT:
+        print(chunk.content, end="", flush=True)
+```
+
+### TOOL_CALL 청크
+
+수행 중인 도구 호출에 대한 정보:
+
+```python Code
+for chunk in streaming:
+    if chunk.chunk_type == StreamChunkType.TOOL_CALL:
+        print(f"\n도구 호출: {chunk.tool_call.tool_name}")
+        print(f"인자: {chunk.tool_call.arguments}")
+```
+
+## 실용적인 예시: 스트리밍을 사용한 UI 구축
+
+다음은 스트리밍을 사용한 대화형 애플리케이션을 구축하는 방법을 보여주는 완전한 예시입니다:
+
+```python Code
+import asyncio
+from crewai import Agent, Crew, Task
+from crewai.types.streaming import StreamChunkType
+
+async def interactive_research():
+    # 스트리밍이 활성화된 crew 생성
+    researcher = Agent(
+        role="Research Analyst",
+        goal="Provide detailed analysis on any topic",
+        backstory="You are an expert researcher with broad knowledge.",
+    )
+
+    task = Task(
+        description="Research and analyze: {topic}",
+        expected_output="A comprehensive analysis with key insights",
+        agent=researcher,
+    )
+
+    crew = Crew(
+        agents=[researcher],
+        tasks=[task],
+        stream=True,
+        verbose=False
+    )
+
+    # 사용자 입력 받기
+    topic = input("연구할 주제를 입력하세요: ")
+
+    print(f"\n{'='*60}")
+    print(f"연구 중: {topic}")
+    print(f"{'='*60}\n")
+
+    # 스트리밍 실행 시작
+    streaming = await crew.kickoff_async(inputs={"topic": topic})
+
+    current_task = ""
+    async for chunk in streaming:
+        # 태스크 전환 표시
+        if chunk.task_name != current_task:
+            current_task = chunk.task_name
+            print(f"\n[{chunk.agent_role}] 작업 중: {chunk.task_name}")
+            print("-" * 60)
+
+        # 텍스트 청크 표시
+        if chunk.chunk_type == StreamChunkType.TEXT:
+            print(chunk.content, end="", flush=True)
+
+        # 도구 호출 표시
+        elif chunk.chunk_type == StreamChunkType.TOOL_CALL and chunk.tool_call:
+            print(f"\n🔧 도구 사용: {chunk.tool_call.tool_name}")
+
+    # 최종 결과 표시
+    result = streaming.result
+    print(f"\n\n{'='*60}")
+    print("분석 완료!")
+    print(f"{'='*60}")
+    print(f"\n토큰 사용량: {result.token_usage}")
+
+asyncio.run(interactive_research())
+```
+
+## 사용 사례
+
+스트리밍은 다음과 같은 경우에 특히 유용합니다:
+
+- **대화형 애플리케이션**: 에이전트가 작업하는 동안 사용자에게 실시간 피드백 제공
+- **장시간 실행 태스크**: 연구, 분석 또는 콘텐츠 생성의 진행 상황 표시
+- **디버깅 및 모니터링**: 에이전트 동작과 의사 결정을 실시간으로 관찰
+- **사용자 경험**: 점진적인 결과를 표시하여 체감 지연 시간 감소
+- **라이브 대시보드**: crew 실행 상태를 표시하는 모니터링 인터페이스 구축
+
+## 중요 사항
+
+- 스트리밍은 crew의 모든 에이전트에 대해 자동으로 LLM 스트리밍을 활성화합니다
+- `.result` 속성에 접근하기 전에 모든 청크를 반복해야 합니다
+- 스트리밍을 사용하는 `kickoff_for_each_async()`의 경우, 모든 출력을 가져오려면 `.results`(복수형)를 사용하세요
+- 스트리밍은 최소한의 오버헤드를 추가하며 실제로 체감 성능을 향상시킬 수 있습니다
+- 각 청크는 풍부한 UI를 위한 전체 컨텍스트(태스크, 에이전트, 청크 타입)를 포함합니다
+
+## 오류 처리
+
+스트리밍 실행 중 오류 처리:
+
+```python Code
+streaming = crew.kickoff(inputs={"topic": "AI"})
+
+try:
+    for chunk in streaming:
+        print(chunk.content, end="", flush=True)
+
+    result = streaming.result
+    print(f"\n성공: {result.raw}")
+
+except Exception as e:
+    print(f"\n스트리밍 중 오류 발생: {e}")
+    if streaming.is_completed:
+        print("스트리밍은 완료되었지만 오류가 발생했습니다")
+```
+
+스트리밍을 활용하면 CrewAI로 더 반응성이 좋고 대화형인 애플리케이션을 구축하여 사용자에게 에이전트 실행과 결과에 대한 실시간 가시성을 제공할 수 있습니다.

docs/pt-BR/concepts/crews.mdx

@@ -32,6 +32,8 @@ Uma crew no crewAI representa um grupo colaborativo de agentes trabalhando em co
 | **Prompt File** _(opcional)_          | `prompt_file`           | Caminho para o arquivo JSON de prompt a ser utilizado pela crew.                                                                                                                   |
 | **Planning** *(opcional)*             | `planning`              | Adiciona habilidade de planejamento à Crew. Quando ativado, antes de cada iteração, todos os dados da Crew são enviados a um AgentPlanner que planejará as tasks e este plano será adicionado à descrição de cada task.    |
 | **Planning LLM** *(opcional)*         | `planning_llm`          | O modelo de linguagem usado pelo AgentPlanner em um processo de planejamento.                                                                                                      |
+| **Knowledge Sources** _(opcional)_    | `knowledge_sources`    | Fontes de conhecimento disponíveis no nível da crew, acessíveis a todos os agentes.                                                                                                                                                             |
+| **Stream** _(opcional)_               | `stream`               | Habilita saída em streaming para receber atualizações em tempo real durante a execução da crew. Retorna um objeto `CrewStreamingOutput` que pode ser iterado para chunks. O padrão é `False`.                                                                                    |
 
 <Tip>
 **Crew Max RPM**: O atributo `max_rpm` define o número máximo de requisições por minuto que a crew pode executar para evitar limites de taxa e irá sobrescrever as configurações de `max_rpm` dos agentes individuais se você o definir.
@@ -303,12 +305,27 @@ print(result)
 
 ### Diferentes Formas de Iniciar uma Crew
 
-Assim que sua crew estiver definida, inicie o fluxo de trabalho com o método kickoff apropriado. O CrewAI oferece vários métodos para melhor controle do processo: `kickoff()`, `kickoff_for_each()`, `kickoff_async()` e `kickoff_for_each_async()`.
+Assim que sua crew estiver definida, inicie o fluxo de trabalho com o método kickoff apropriado. O CrewAI oferece vários métodos para melhor controle do processo.
+
+#### Métodos Síncronos
 
 - `kickoff()`: Inicia o processo de execução seguindo o fluxo definido.
 - `kickoff_for_each()`: Executa tasks sequencialmente para cada evento de entrada ou item da coleção fornecida.
-- `kickoff_async()`: Inicia o workflow de forma assíncrona.
-- `kickoff_for_each_async()`: Executa as tasks concorrentemente para cada entrada, aproveitando o processamento assíncrono.
+
+#### Métodos Assíncronos
+
+O CrewAI oferece duas abordagens para execução assíncrona:
+
+| Método | Tipo | Descrição |
+|--------|------|-------------|
+| `akickoff()` | Async nativo | Async/await verdadeiro em toda a cadeia de execução |
+| `akickoff_for_each()` | Async nativo | Execução async nativa para cada entrada em uma lista |
+| `kickoff_async()` | Baseado em thread | Envolve execução síncrona em `asyncio.to_thread` |
+| `kickoff_for_each_async()` | Baseado em thread | Async baseado em thread para cada entrada em uma lista |
+
+<Note>
+Para cargas de trabalho de alta concorrência, `akickoff()` e `akickoff_for_each()` são recomendados pois usam async nativo para execução de tasks, operações de memória e recuperação de conhecimento.
+</Note>
 
 ```python Code
 # Iniciar execução das tasks da crew
@@ -321,19 +338,53 @@ results = my_crew.kickoff_for_each(inputs=inputs_array)
 for result in results:
     print(result)
 
-# Exemplo com kickoff_async
+# Exemplo usando async nativo com akickoff
+inputs = {'topic': 'AI in healthcare'}
+async_result = await my_crew.akickoff(inputs=inputs)
+print(async_result)
+
+# Exemplo usando async nativo com akickoff_for_each
+inputs_array = [{'topic': 'AI in healthcare'}, {'topic': 'AI in finance'}]
+async_results = await my_crew.akickoff_for_each(inputs=inputs_array)
+for async_result in async_results:
+    print(async_result)
+
+# Exemplo usando kickoff_async baseado em thread
 inputs = {'topic': 'AI in healthcare'}
 async_result = await my_crew.kickoff_async(inputs=inputs)
 print(async_result)
 
-# Exemplo com kickoff_for_each_async
+# Exemplo usando kickoff_for_each_async baseado em thread
 inputs_array = [{'topic': 'AI in healthcare'}, {'topic': 'AI in finance'}]
 async_results = await my_crew.kickoff_for_each_async(inputs=inputs_array)
 for async_result in async_results:
     print(async_result)
 ```
 
-Esses métodos fornecem flexibilidade para gerenciar e executar tasks dentro de sua crew, permitindo fluxos de trabalho síncronos e assíncronos de acordo com sua necessidade.
+Esses métodos fornecem flexibilidade para gerenciar e executar tasks dentro de sua crew, permitindo fluxos de trabalho síncronos e assíncronos de acordo com sua necessidade. Para exemplos detalhados de async, consulte o guia [Inicie uma Crew de Forma Assíncrona](/pt-BR/learn/kickoff-async).
+
+### Streaming na Execução da Crew
+
+Para visibilidade em tempo real da execução da crew, você pode habilitar streaming para receber saída conforme é gerada:
+
+```python Code
+# Habilitar streaming
+crew = Crew(
+    agents=[researcher],
+    tasks=[task],
+    stream=True
+)
+
+# Iterar sobre saída em streaming
+streaming = crew.kickoff(inputs={"topic": "AI"})
+for chunk in streaming:
+    print(chunk.content, end="", flush=True)
+
+# Acessar resultado final
+result = streaming.result
+```
+
+Saiba mais sobre streaming no guia [Streaming na Execução da Crew](/pt-BR/learn/streaming-crew-execution).
 
 ### Repetindo Execução a partir de uma Task Específica
 

docs/pt-BR/concepts/memory.mdx

@@ -515,8 +515,7 @@ crew = Crew(
         "provider": "huggingface",
         "config": {
             "api_key": "your-hf-token",  # Opcional para modelos públicos
-            "model": "sentence-transformers/all-MiniLM-L6-v2",
-            "api_url": "https://api-inference.huggingface.co"  # ou seu endpoint customizado
+            "model": "sentence-transformers/all-MiniLM-L6-v2"
         }
     }
 )

docs/pt-BR/learn/create-custom-tools.mdx

@@ -66,5 +66,55 @@ def my_cache_strategy(arguments: dict, result: str) -> bool:
 cached_tool.cache_function = my_cache_strategy
 ```
 
+### Criando Ferramentas Assíncronas
+
+O CrewAI suporta ferramentas assíncronas para operações de I/O não bloqueantes. Isso é útil quando sua ferramenta precisa fazer requisições HTTP, consultas a banco de dados ou outras operações de I/O.
+
+#### Usando o Decorador `@tool` com Funções Assíncronas
+
+A maneira mais simples de criar uma ferramenta assíncrona é usando o decorador `@tool` com uma função async:
+
+```python Code
+import aiohttp
+from crewai.tools import tool
+
+@tool("Async Web Fetcher")
+async def fetch_webpage(url: str) -> str:
+    """Fetch content from a webpage asynchronously."""
+    async with aiohttp.ClientSession() as session:
+        async with session.get(url) as response:
+            return await response.text()
+```
+
+#### Subclassificando `BaseTool` com Suporte Assíncrono
+
+Para maior controle, herde de `BaseTool` e implemente os métodos `_run` (síncrono) e `_arun` (assíncrono):
+
+```python Code
+import requests
+import aiohttp
+from crewai.tools import BaseTool
+from pydantic import BaseModel, Field
+
+class WebFetcherInput(BaseModel):
+    """Input schema for WebFetcher."""
+    url: str = Field(..., description="The URL to fetch")
+
+class WebFetcherTool(BaseTool):
+    name: str = "Web Fetcher"
+    description: str = "Fetches content from a URL"
+    args_schema: type[BaseModel] = WebFetcherInput
+
+    def _run(self, url: str) -> str:
+        """Synchronous implementation."""
+        return requests.get(url).text
+
+    async def _arun(self, url: str) -> str:
+        """Asynchronous implementation for non-blocking I/O."""
+        async with aiohttp.ClientSession() as session:
+            async with session.get(url) as response:
+                return await response.text()
+```
+
 Seguindo essas orientações e incorporando novas funcionalidades e ferramentas de colaboração nos seus processos de criação e gerenciamento de ferramentas,
-você pode aproveitar ao máximo as capacidades do framework CrewAI, aprimorando tanto a experiência de desenvolvimento quanto a eficiência dos seus agentes de IA.
+você pode aproveitar ao máximo as capacidades do framework CrewAI, aprimorando tanto a experiência de desenvolvimento quanto a eficiência dos seus agentes de IA.

docs/pt-BR/learn/kickoff-async.mdx

@@ -7,17 +7,28 @@ mode: "wide"
 
 ## Introdução
 
-A CrewAI oferece a capacidade de iniciar uma crew de forma assíncrona, permitindo que você comece a execução da crew de maneira não bloqueante. 
+A CrewAI oferece a capacidade de iniciar uma crew de forma assíncrona, permitindo que você comece a execução da crew de maneira não bloqueante.
 Esse recurso é especialmente útil quando você deseja executar múltiplas crews simultaneamente ou quando precisa realizar outras tarefas enquanto a crew está em execução.
 
-## Execução Assíncrona de Crew
+O CrewAI oferece duas abordagens para execução assíncrona:
 
-Para iniciar uma crew de forma assíncrona, utilize o método `kickoff_async()`. Este método inicia a execução da crew em uma thread separada, permitindo que a thread principal continue executando outras tarefas.
+| Método | Tipo | Descrição |
+|--------|------|-------------|
+| `akickoff()` | Async nativo | Async/await verdadeiro em toda a cadeia de execução |
+| `kickoff_async()` | Baseado em thread | Envolve execução síncrona em `asyncio.to_thread` |
+
+<Note>
+Para cargas de trabalho de alta concorrência, `akickoff()` é recomendado pois usa async nativo para execução de tasks, operações de memória e recuperação de conhecimento.
+</Note>
+
+## Execução Async Nativa com `akickoff()`
+
+O método `akickoff()` fornece execução async nativa verdadeira, usando async/await em toda a cadeia de execução, incluindo execução de tasks, operações de memória e consultas de conhecimento.
 
 ### Assinatura do Método
 
 ```python Code
-def kickoff_async(self, inputs: dict) -> CrewOutput:
+async def akickoff(self, inputs: dict) -> CrewOutput:
 ```
 
 ### Parâmetros
@@ -28,97 +39,268 @@ def kickoff_async(self, inputs: dict) -> CrewOutput:
 
 - `CrewOutput`: Um objeto que representa o resultado da execução da crew.
 
-## Possíveis Casos de Uso
-
-- **Geração Paralela de Conteúdo**: Inicie múltiplas crews independentes de forma assíncrona, cada uma responsável por gerar conteúdo sobre temas diferentes. Por exemplo, uma crew pode pesquisar e redigir um artigo sobre tendências em IA, enquanto outra gera posts para redes sociais sobre o lançamento de um novo produto. Cada crew atua de forma independente, permitindo a escala eficiente da produção de conteúdo.
-
-- **Tarefas Conjuntas de Pesquisa de Mercado**: Lance múltiplas crews de forma assíncrona para realizar pesquisas de mercado em paralelo. Uma crew pode analisar tendências do setor, outra examinar estratégias de concorrentes e ainda outra avaliar o sentimento do consumidor. Cada crew conclui sua tarefa de forma independente, proporcionando insights mais rápidos e abrangentes.
-
-- **Módulos Independentes de Planejamento de Viagem**: Execute crews separadas para planejar diferentes aspectos de uma viagem de forma independente. Uma crew pode cuidar das opções de voo, outra das acomodações e uma terceira do planejamento das atividades. Cada crew trabalha de maneira assíncrona, permitindo que os vários componentes da viagem sejam planejados ao mesmo tempo e de maneira independente, para resultados mais rápidos.
-
-## Exemplo: Execução Assíncrona de uma Única Crew
-
-Veja um exemplo de como iniciar uma crew de forma assíncrona utilizando asyncio e aguardando o resultado:
+### Exemplo: Execução Async Nativa de Crew
 
 ```python Code
 import asyncio
 from crewai import Crew, Agent, Task
 
-# Create an agent with code execution enabled
+# Criar um agente
 coding_agent = Agent(
-    role="Analista de Dados Python",
-    goal="Analisar dados e fornecer insights usando Python",
-    backstory="Você é um analista de dados experiente com fortes habilidades em Python.",
+    role="Python Data Analyst",
+    goal="Analyze data and provide insights using Python",
+    backstory="You are an experienced data analyst with strong Python skills.",
     allow_code_execution=True
 )
 
-# Create a task that requires code execution
+# Criar uma tarefa
 data_analysis_task = Task(
-    description="Analise o conjunto de dados fornecido e calcule a idade média dos participantes. Idades: {ages}",
+    description="Analyze the given dataset and calculate the average age of participants. Ages: {ages}",
     agent=coding_agent,
-    expected_output="A idade média dos participantes."
+    expected_output="The average age of the participants."
 )
 
-# Create a crew and add the task
+# Criar uma crew
 analysis_crew = Crew(
     agents=[coding_agent],
     tasks=[data_analysis_task]
 )
 
-# Async function to kickoff the crew asynchronously
-async def async_crew_execution():
-    result = await analysis_crew.kickoff_async(inputs={"ages": [25, 30, 35, 40, 45]})
+# Execução async nativa
+async def main():
+    result = await analysis_crew.akickoff(inputs={"ages": [25, 30, 35, 40, 45]})
     print("Crew Result:", result)
 
-# Run the async function
-asyncio.run(async_crew_execution())
+asyncio.run(main())
 ```
 
-## Exemplo: Execução Assíncrona de Múltiplas Crews
+### Exemplo: Múltiplas Crews Async Nativas
 
-Neste exemplo, mostraremos como iniciar múltiplas crews de forma assíncrona e aguardar todas serem concluídas usando `asyncio.gather()`:
+Execute múltiplas crews concorrentemente usando `asyncio.gather()` com async nativo:
 
 ```python Code
 import asyncio
 from crewai import Crew, Agent, Task
 
-# Create an agent with code execution enabled
 coding_agent = Agent(
-    role="Analista de Dados Python",
-    goal="Analisar dados e fornecer insights usando Python",
-    backstory="Você é um analista de dados experiente com fortes habilidades em Python.",
+    role="Python Data Analyst",
+    goal="Analyze data and provide insights using Python",
+    backstory="You are an experienced data analyst with strong Python skills.",
     allow_code_execution=True
 )
 
-# Create tasks that require code execution
 task_1 = Task(
-    description="Analise o primeiro conjunto de dados e calcule a idade média dos participantes. Idades: {ages}",
+    description="Analyze the first dataset and calculate the average age. Ages: {ages}",
     agent=coding_agent,
-    expected_output="A idade média dos participantes."
+    expected_output="The average age of the participants."
 )
 
 task_2 = Task(
-    description="Analise o segundo conjunto de dados e calcule a idade média dos participantes. Idades: {ages}",
+    description="Analyze the second dataset and calculate the average age. Ages: {ages}",
     agent=coding_agent,
-    expected_output="A idade média dos participantes."
+    expected_output="The average age of the participants."
+)
+
+crew_1 = Crew(agents=[coding_agent], tasks=[task_1])
+crew_2 = Crew(agents=[coding_agent], tasks=[task_2])
+
+async def main():
+    results = await asyncio.gather(
+        crew_1.akickoff(inputs={"ages": [25, 30, 35, 40, 45]}),
+        crew_2.akickoff(inputs={"ages": [20, 22, 24, 28, 30]})
+    )
+
+    for i, result in enumerate(results, 1):
+        print(f"Crew {i} Result:", result)
+
+asyncio.run(main())
+```
+
+### Exemplo: Async Nativo para Múltiplas Entradas
+
+Use `akickoff_for_each()` para executar sua crew contra múltiplas entradas concorrentemente com async nativo:
+
+```python Code
+import asyncio
+from crewai import Crew, Agent, Task
+
+coding_agent = Agent(
+    role="Python Data Analyst",
+    goal="Analyze data and provide insights using Python",
+    backstory="You are an experienced data analyst with strong Python skills.",
+    allow_code_execution=True
+)
+
+data_analysis_task = Task(
+    description="Analyze the dataset and calculate the average age. Ages: {ages}",
+    agent=coding_agent,
+    expected_output="The average age of the participants."
+)
+
+analysis_crew = Crew(
+    agents=[coding_agent],
+    tasks=[data_analysis_task]
+)
+
+async def main():
+    datasets = [
+        {"ages": [25, 30, 35, 40, 45]},
+        {"ages": [20, 22, 24, 28, 30]},
+        {"ages": [30, 35, 40, 45, 50]}
+    ]
+
+    results = await analysis_crew.akickoff_for_each(datasets)
+
+    for i, result in enumerate(results, 1):
+        print(f"Dataset {i} Result:", result)
+
+asyncio.run(main())
+```
+
+## Async Baseado em Thread com `kickoff_async()`
+
+O método `kickoff_async()` fornece execução async envolvendo o `kickoff()` síncrono em uma thread. Isso é útil para integração async mais simples ou compatibilidade retroativa.
+
+### Assinatura do Método
+
+```python Code
+async def kickoff_async(self, inputs: dict) -> CrewOutput:
+```
+
+### Parâmetros
+
+- `inputs` (dict): Um dicionário contendo os dados de entrada necessários para as tarefas.
+
+### Retorno
+
+- `CrewOutput`: Um objeto que representa o resultado da execução da crew.
+
+### Exemplo: Execução Async Baseada em Thread
+
+```python Code
+import asyncio
+from crewai import Crew, Agent, Task
+
+coding_agent = Agent(
+    role="Python Data Analyst",
+    goal="Analyze data and provide insights using Python",
+    backstory="You are an experienced data analyst with strong Python skills.",
+    allow_code_execution=True
+)
+
+data_analysis_task = Task(
+    description="Analyze the given dataset and calculate the average age of participants. Ages: {ages}",
+    agent=coding_agent,
+    expected_output="The average age of the participants."
+)
+
+analysis_crew = Crew(
+    agents=[coding_agent],
+    tasks=[data_analysis_task]
+)
+
+async def async_crew_execution():
+    result = await analysis_crew.kickoff_async(inputs={"ages": [25, 30, 35, 40, 45]})
+    print("Crew Result:", result)
+
+asyncio.run(async_crew_execution())
+```
+
+### Exemplo: Múltiplas Crews Async Baseadas em Thread
+
+```python Code
+import asyncio
+from crewai import Crew, Agent, Task
+
+coding_agent = Agent(
+    role="Python Data Analyst",
+    goal="Analyze data and provide insights using Python",
+    backstory="You are an experienced data analyst with strong Python skills.",
+    allow_code_execution=True
+)
+
+task_1 = Task(
+    description="Analyze the first dataset and calculate the average age of participants. Ages: {ages}",
+    agent=coding_agent,
+    expected_output="The average age of the participants."
+)
+
+task_2 = Task(
+    description="Analyze the second dataset and calculate the average age of participants. Ages: {ages}",
+    agent=coding_agent,
+    expected_output="The average age of the participants."
 )
 
-# Create two crews and add tasks
 crew_1 = Crew(agents=[coding_agent], tasks=[task_1])
 crew_2 = Crew(agents=[coding_agent], tasks=[task_2])
 
-# Async function to kickoff multiple crews asynchronously and wait for all to finish
 async def async_multiple_crews():
-    # Create coroutines for concurrent execution
     result_1 = crew_1.kickoff_async(inputs={"ages": [25, 30, 35, 40, 45]})
     result_2 = crew_2.kickoff_async(inputs={"ages": [20, 22, 24, 28, 30]})
 
-    # Wait for both crews to finish
     results = await asyncio.gather(result_1, result_2)
 
     for i, result in enumerate(results, 1):
         print(f"Crew {i} Result:", result)
 
-# Run the async function
 asyncio.run(async_multiple_crews())
-```
+```
+
+## Streaming Assíncrono
+
+Ambos os métodos async suportam streaming quando `stream=True` está definido na crew:
+
+```python Code
+import asyncio
+from crewai import Crew, Agent, Task
+
+agent = Agent(
+    role="Researcher",
+    goal="Research and summarize topics",
+    backstory="You are an expert researcher."
+)
+
+task = Task(
+    description="Research the topic: {topic}",
+    agent=agent,
+    expected_output="A comprehensive summary of the topic."
+)
+
+crew = Crew(
+    agents=[agent],
+    tasks=[task],
+    stream=True  # Habilitar streaming
+)
+
+async def main():
+    streaming_output = await crew.akickoff(inputs={"topic": "AI trends in 2024"})
+
+    # Iteração async sobre chunks de streaming
+    async for chunk in streaming_output:
+        print(f"Chunk: {chunk.content}")
+
+    # Acessar resultado final após streaming completar
+    result = streaming_output.result
+    print(f"Final result: {result.raw}")
+
+asyncio.run(main())
+```
+
+## Possíveis Casos de Uso
+
+- **Geração Paralela de Conteúdo**: Inicie múltiplas crews independentes de forma assíncrona, cada uma responsável por gerar conteúdo sobre temas diferentes. Por exemplo, uma crew pode pesquisar e redigir um artigo sobre tendências em IA, enquanto outra gera posts para redes sociais sobre o lançamento de um novo produto.
+
+- **Tarefas Conjuntas de Pesquisa de Mercado**: Lance múltiplas crews de forma assíncrona para realizar pesquisas de mercado em paralelo. Uma crew pode analisar tendências do setor, outra examinar estratégias de concorrentes e ainda outra avaliar o sentimento do consumidor.
+
+- **Módulos Independentes de Planejamento de Viagem**: Execute crews separadas para planejar diferentes aspectos de uma viagem de forma independente. Uma crew pode cuidar das opções de voo, outra das acomodações e uma terceira do planejamento das atividades.
+
+## Escolhendo entre `akickoff()` e `kickoff_async()`
+
+| Recurso | `akickoff()` | `kickoff_async()` |
+|---------|--------------|-------------------|
+| Modelo de execução | Async/await nativo | Wrapper baseado em thread |
+| Execução de tasks | Async com `aexecute_sync()` | Síncrono em thread pool |
+| Operações de memória | Async | Síncrono em thread pool |
+| Recuperação de conhecimento | Async | Síncrono em thread pool |
+| Melhor para | Alta concorrência, cargas I/O-bound | Integração async simples |
+| Suporte a streaming | Sim | Sim |

docs/pt-BR/learn/streaming-crew-execution.mdx

@@ -0,0 +1,356 @@
+---
+title: Streaming na Execução da Crew
+description: Transmita saída em tempo real da execução da sua crew no CrewAI
+icon: wave-pulse
+mode: "wide"
+---
+
+## Introdução
+
+O CrewAI fornece a capacidade de transmitir saída em tempo real durante a execução da crew, permitindo que você exiba resultados conforme são gerados, em vez de esperar que todo o processo seja concluído. Este recurso é particularmente útil para construir aplicações interativas, fornecer feedback ao usuário e monitorar processos de longa duração.
+
+## Como o Streaming Funciona
+
+Quando o streaming está ativado, o CrewAI captura respostas do LLM e chamadas de ferramentas conforme acontecem, empacotando-as em chunks estruturados que incluem contexto sobre qual task e agent está executando. Você pode iterar sobre esses chunks em tempo real e acessar o resultado final quando a execução for concluída.
+
+## Ativando o Streaming
+
+Para ativar o streaming, defina o parâmetro `stream` como `True` ao criar sua crew:
+
+```python Code
+from crewai import Agent, Crew, Task
+
+# Crie seus agentes e tasks
+researcher = Agent(
+    role="Research Analyst",
+    goal="Gather comprehensive information on topics",
+    backstory="You are an experienced researcher with excellent analytical skills.",
+)
+
+task = Task(
+    description="Research the latest developments in AI",
+    expected_output="A detailed report on recent AI advancements",
+    agent=researcher,
+)
+
+# Ativar streaming
+crew = Crew(
+    agents=[researcher],
+    tasks=[task],
+    stream=True  # Ativar saída em streaming
+)
+```
+
+## Streaming Síncrono
+
+Quando você chama `kickoff()` em uma crew com streaming ativado, ele retorna um objeto `CrewStreamingOutput` que você pode iterar para receber chunks conforme chegam:
+
+```python Code
+# Iniciar execução com streaming
+streaming = crew.kickoff(inputs={"topic": "artificial intelligence"})
+
+# Iterar sobre chunks conforme chegam
+for chunk in streaming:
+    print(chunk.content, end="", flush=True)
+
+# Acessar o resultado final após o streaming completar
+result = streaming.result
+print(f"\n\nSaída final: {result.raw}")
+```
+
+### Informações do Chunk de Stream
+
+Cada chunk fornece contexto rico sobre a execução:
+
+```python Code
+streaming = crew.kickoff(inputs={"topic": "AI"})
+
+for chunk in streaming:
+    print(f"Task: {chunk.task_name} (índice {chunk.task_index})")
+    print(f"Agent: {chunk.agent_role}")
+    print(f"Content: {chunk.content}")
+    print(f"Type: {chunk.chunk_type}")  # TEXT ou TOOL_CALL
+    if chunk.tool_call:
+        print(f"Tool: {chunk.tool_call.tool_name}")
+        print(f"Arguments: {chunk.tool_call.arguments}")
+```
+
+### Acessando Resultados do Streaming
+
+O objeto `CrewStreamingOutput` fornece várias propriedades úteis:
+
+```python Code
+streaming = crew.kickoff(inputs={"topic": "AI"})
+
+# Iterar e coletar chunks
+for chunk in streaming:
+    print(chunk.content, end="", flush=True)
+
+# Após a iteração completar
+print(f"\nCompletado: {streaming.is_completed}")
+print(f"Texto completo: {streaming.get_full_text()}")
+print(f"Todos os chunks: {len(streaming.chunks)}")
+print(f"Resultado final: {streaming.result.raw}")
+```
+
+## Streaming Assíncrono
+
+Para aplicações assíncronas, você pode usar `akickoff()` (async nativo) ou `kickoff_async()` (baseado em threads) com iteração assíncrona:
+
+### Async Nativo com `akickoff()`
+
+O método `akickoff()` fornece execução async nativa verdadeira em toda a cadeia:
+
+```python Code
+import asyncio
+
+async def stream_crew():
+    crew = Crew(
+        agents=[researcher],
+        tasks=[task],
+        stream=True
+    )
+
+    # Iniciar streaming async nativo
+    streaming = await crew.akickoff(inputs={"topic": "AI"})
+
+    # Iteração assíncrona sobre chunks
+    async for chunk in streaming:
+        print(chunk.content, end="", flush=True)
+
+    # Acessar resultado final
+    result = streaming.result
+    print(f"\n\nSaída final: {result.raw}")
+
+asyncio.run(stream_crew())
+```
+
+### Async Baseado em Threads com `kickoff_async()`
+
+Para integração async mais simples ou compatibilidade retroativa:
+
+```python Code
+import asyncio
+
+async def stream_crew():
+    crew = Crew(
+        agents=[researcher],
+        tasks=[task],
+        stream=True
+    )
+
+    # Iniciar streaming async baseado em threads
+    streaming = await crew.kickoff_async(inputs={"topic": "AI"})
+
+    # Iteração assíncrona sobre chunks
+    async for chunk in streaming:
+        print(chunk.content, end="", flush=True)
+
+    # Acessar resultado final
+    result = streaming.result
+    print(f"\n\nSaída final: {result.raw}")
+
+asyncio.run(stream_crew())
+```
+
+<Note>
+Para cargas de trabalho de alta concorrência, `akickoff()` é recomendado pois usa async nativo para execução de tasks, operações de memória e recuperação de conhecimento. Consulte o guia [Iniciar Crew de Forma Assíncrona](/pt-BR/learn/kickoff-async) para mais detalhes.
+</Note>
+
+## Streaming com kickoff_for_each
+
+Ao executar uma crew para múltiplas entradas com `kickoff_for_each()`, o streaming funciona de forma diferente dependendo se você usa síncrono ou assíncrono:
+
+### kickoff_for_each Síncrono
+
+Com `kickoff_for_each()` síncrono, você obtém uma lista de objetos `CrewStreamingOutput`, um para cada entrada:
+
+```python Code
+crew = Crew(
+    agents=[researcher],
+    tasks=[task],
+    stream=True
+)
+
+inputs_list = [
+    {"topic": "AI in healthcare"},
+    {"topic": "AI in finance"}
+]
+
+# Retorna lista de saídas de streaming
+streaming_outputs = crew.kickoff_for_each(inputs=inputs_list)
+
+# Iterar sobre cada saída de streaming
+for i, streaming in enumerate(streaming_outputs):
+    print(f"\n=== Entrada {i + 1} ===")
+    for chunk in streaming:
+        print(chunk.content, end="", flush=True)
+
+    result = streaming.result
+    print(f"\n\nResultado {i + 1}: {result.raw}")
+```
+
+### kickoff_for_each_async Assíncrono
+
+Com `kickoff_for_each_async()` assíncrono, você obtém um único `CrewStreamingOutput` que produz chunks de todas as crews conforme chegam concorrentemente:
+
+```python Code
+import asyncio
+
+async def stream_multiple_crews():
+    crew = Crew(
+        agents=[researcher],
+        tasks=[task],
+        stream=True
+    )
+
+    inputs_list = [
+        {"topic": "AI in healthcare"},
+        {"topic": "AI in finance"}
+    ]
+
+    # Retorna saída de streaming única para todas as crews
+    streaming = await crew.kickoff_for_each_async(inputs=inputs_list)
+
+    # Chunks de todas as crews chegam conforme são gerados
+    async for chunk in streaming:
+        print(f"[{chunk.task_name}] {chunk.content}", end="", flush=True)
+
+    # Acessar todos os resultados
+    results = streaming.results  # Lista de objetos CrewOutput
+    for i, result in enumerate(results):
+        print(f"\n\nResultado {i + 1}: {result.raw}")
+
+asyncio.run(stream_multiple_crews())
+```
+
+## Tipos de Chunk de Stream
+
+Chunks podem ser de diferentes tipos, indicados pelo campo `chunk_type`:
+
+### Chunks TEXT
+
+Conteúdo de texto padrão de respostas do LLM:
+
+```python Code
+for chunk in streaming:
+    if chunk.chunk_type == StreamChunkType.TEXT:
+        print(chunk.content, end="", flush=True)
+```
+
+### Chunks TOOL_CALL
+
+Informações sobre chamadas de ferramentas sendo feitas:
+
+```python Code
+for chunk in streaming:
+    if chunk.chunk_type == StreamChunkType.TOOL_CALL:
+        print(f"\nChamando ferramenta: {chunk.tool_call.tool_name}")
+        print(f"Argumentos: {chunk.tool_call.arguments}")
+```
+
+## Exemplo Prático: Construindo uma UI com Streaming
+
+Aqui está um exemplo completo mostrando como construir uma aplicação interativa com streaming:
+
+```python Code
+import asyncio
+from crewai import Agent, Crew, Task
+from crewai.types.streaming import StreamChunkType
+
+async def interactive_research():
+    # Criar crew com streaming ativado
+    researcher = Agent(
+        role="Research Analyst",
+        goal="Provide detailed analysis on any topic",
+        backstory="You are an expert researcher with broad knowledge.",
+    )
+
+    task = Task(
+        description="Research and analyze: {topic}",
+        expected_output="A comprehensive analysis with key insights",
+        agent=researcher,
+    )
+
+    crew = Crew(
+        agents=[researcher],
+        tasks=[task],
+        stream=True,
+        verbose=False
+    )
+
+    # Obter entrada do usuário
+    topic = input("Digite um tópico para pesquisar: ")
+
+    print(f"\n{'='*60}")
+    print(f"Pesquisando: {topic}")
+    print(f"{'='*60}\n")
+
+    # Iniciar execução com streaming
+    streaming = await crew.kickoff_async(inputs={"topic": topic})
+
+    current_task = ""
+    async for chunk in streaming:
+        # Mostrar transições de task
+        if chunk.task_name != current_task:
+            current_task = chunk.task_name
+            print(f"\n[{chunk.agent_role}] Trabalhando em: {chunk.task_name}")
+            print("-" * 60)
+
+        # Exibir chunks de texto
+        if chunk.chunk_type == StreamChunkType.TEXT:
+            print(chunk.content, end="", flush=True)
+
+        # Exibir chamadas de ferramentas
+        elif chunk.chunk_type == StreamChunkType.TOOL_CALL and chunk.tool_call:
+            print(f"\n🔧 Usando ferramenta: {chunk.tool_call.tool_name}")
+
+    # Mostrar resultado final
+    result = streaming.result
+    print(f"\n\n{'='*60}")
+    print("Análise Completa!")
+    print(f"{'='*60}")
+    print(f"\nUso de Tokens: {result.token_usage}")
+
+asyncio.run(interactive_research())
+```
+
+## Casos de Uso
+
+O streaming é particularmente valioso para:
+
+- **Aplicações Interativas**: Fornecer feedback em tempo real aos usuários enquanto os agentes trabalham
+- **Tasks de Longa Duração**: Mostrar progresso para pesquisa, análise ou geração de conteúdo
+- **Depuração e Monitoramento**: Observar comportamento e tomada de decisão dos agentes em tempo real
+- **Experiência do Usuário**: Reduzir latência percebida mostrando resultados incrementais
+- **Dashboards ao Vivo**: Construir interfaces de monitoramento que exibem status de execução da crew
+
+## Notas Importantes
+
+- O streaming ativa automaticamente o streaming do LLM para todos os agentes na crew
+- Você deve iterar através de todos os chunks antes de acessar a propriedade `.result`
+- Para `kickoff_for_each_async()` com streaming, use `.results` (plural) para obter todas as saídas
+- O streaming adiciona overhead mínimo e pode realmente melhorar a performance percebida
+- Cada chunk inclui contexto completo (task, agente, tipo de chunk) para UIs ricas
+
+## Tratamento de Erros
+
+Trate erros durante a execução com streaming:
+
+```python Code
+streaming = crew.kickoff(inputs={"topic": "AI"})
+
+try:
+    for chunk in streaming:
+        print(chunk.content, end="", flush=True)
+
+    result = streaming.result
+    print(f"\nSucesso: {result.raw}")
+
+except Exception as e:
+    print(f"\nErro durante o streaming: {e}")
+    if streaming.is_completed:
+        print("O streaming foi completado mas ocorreu um erro")
+```
+
+Ao aproveitar o streaming, você pode construir aplicações mais responsivas e interativas com o CrewAI, fornecendo aos usuários visibilidade em tempo real da execução dos agentes e resultados.

lib/crewai-tools/BUILDING_TOOLS.md

@@ -218,7 +218,7 @@ Update the root `README.md` only if the tool introduces a new category or notabl
 
 ## Discovery and specs
 
-Our internal tooling discovers classes whose names end with `Tool`. Keep your class exported from the module path under `crewai_tools/tools/...` to be picked up by scripts like `generate_tool_specs.py`.
+Our internal tooling discovers classes whose names end with `Tool`. Keep your class exported from the module path under `crewai_tools/tools/...` to be picked up by scripts like `crewai_tools.generate_tool_specs.py`.
 
 ---
 

lib/crewai-tools/pyproject.toml

@@ -8,17 +8,17 @@ authors = [
 ]
 requires-python = ">=3.10, <3.14"
 dependencies = [
-    "lancedb>=0.5.4",
-    "pytube>=15.0.0",
-    "requests>=2.32.5",
-    "docker>=7.1.0",
-    "crewai==1.6.1",
-    "lancedb>=0.5.4",
-    "tiktoken>=0.8.0",
-    "beautifulsoup4>=4.13.4",
-    "python-docx>=1.2.0",
-    "youtube-transcript-api>=1.2.2",
-    "pymupdf>=1.26.6",
+    "lancedb~=0.5.4",
+    "pytube~=15.0.0",
+    "requests~=2.32.5",
+    "docker~=7.1.0",
+    "crewai==1.7.0",
+    "lancedb~=0.5.4",
+    "tiktoken~=0.8.0",
+    "beautifulsoup4~=4.13.4",
+    "python-docx~=1.2.0",
+    "youtube-transcript-api~=1.2.2",
+    "pymupdf~=1.26.6",
 ]
 
 

lib/crewai-tools/src/crewai_tools/__init__.py

@@ -291,4 +291,4 @@ __all__ = [
     "ZapierActionTools",
 ]
 
-__version__ = "1.6.1"
+__version__ = "1.7.0"

lib/crewai-tools/src/crewai_tools/generate_tool_specs.py

@@ -4,17 +4,20 @@ from collections.abc import Mapping
 import inspect
 import json
 from pathlib import Path
-from typing import Any, cast
+from typing import Any
 
 from crewai.tools.base_tool import BaseTool, EnvVar
-from crewai_tools import tools
 from pydantic import BaseModel
 from pydantic.json_schema import GenerateJsonSchema
 from pydantic_core import PydanticOmit
 
+from crewai_tools import tools
+
 
 class SchemaGenerator(GenerateJsonSchema):
-    def handle_invalid_for_json_schema(self, schema, error_info):
+    def handle_invalid_for_json_schema(
+        self, schema: Any, error_info: Any
+    ) -> dict[str, Any]:
         raise PydanticOmit
 
 
@@ -73,7 +76,7 @@ class ToolSpecExtractor:
 
     @staticmethod
     def _extract_field_default(
-        field: dict | None, fallback: str | list[Any] = ""
+        field: dict[str, Any] | None, fallback: str | list[Any] = ""
     ) -> str | list[Any] | int:
         if not field:
             return fallback
@@ -83,7 +86,7 @@ class ToolSpecExtractor:
         return default if isinstance(default, (list, str, int)) else fallback
 
     @staticmethod
-    def _extract_params(args_schema_field: dict | None) -> dict[str, Any]:
+    def _extract_params(args_schema_field: dict[str, Any] | None) -> dict[str, Any]:
         if not args_schema_field:
             return {}
 
@@ -94,15 +97,15 @@ class ToolSpecExtractor:
         ):
             return {}
 
-        # Cast to type[BaseModel] after runtime check
-        schema_class = cast(type[BaseModel], args_schema_class)
         try:
-            return schema_class.model_json_schema(schema_generator=SchemaGenerator)
+            return args_schema_class.model_json_schema(schema_generator=SchemaGenerator)
         except Exception:
             return {}
 
     @staticmethod
-    def _extract_env_vars(env_vars_field: dict | None) -> list[dict[str, Any]]:
+    def _extract_env_vars(
+        env_vars_field: dict[str, Any] | None,
+    ) -> list[dict[str, Any]]:
         if not env_vars_field:
             return []
 

lib/crewai-tools/tests/it/tools/conftest.py

@@ -1,21 +0,0 @@
-import pytest
-
-
-def pytest_configure(config):
-    """Register custom markers."""
-    config.addinivalue_line("markers", "integration: mark test as an integration test")
-    config.addinivalue_line("markers", "asyncio: mark test as an async test")
-
-    # Set the asyncio loop scope through ini configuration
-    config.inicfg["asyncio_mode"] = "auto"
-
-
-@pytest.fixture(scope="function")
-def event_loop():
-    """Create an instance of the default event loop for each test case."""
-    import asyncio
-
-    loop = asyncio.new_event_loop()
-    asyncio.set_event_loop(loop)
-    yield loop
-    loop.close()

lib/crewai-tools/tests/test_generate_tool_specs.py

@@ -2,7 +2,7 @@ import json
 from unittest import mock
 
 from crewai.tools.base_tool import BaseTool, EnvVar
-from generate_tool_specs import ToolSpecExtractor
+from crewai_tools.generate_tool_specs import ToolSpecExtractor
 from pydantic import BaseModel, Field
 import pytest
 
@@ -61,8 +61,8 @@ def test_unwrap_schema(extractor):
 @pytest.fixture
 def mock_tool_extractor(extractor):
     with (
-        mock.patch("generate_tool_specs.dir", return_value=["MockTool"]),
-        mock.patch("generate_tool_specs.getattr", return_value=MockTool),
+        mock.patch("crewai_tools.generate_tool_specs.dir", return_value=["MockTool"]),
+        mock.patch("crewai_tools.generate_tool_specs.getattr", return_value=MockTool),
     ):
         extractor.extract_all_tools()
         assert len(extractor.tools_spec) == 1

lib/crewai-tools/tests/tools/firecrawl_crawl_website_tool_test.py

@@ -4,7 +4,7 @@ from crewai_tools.tools.firecrawl_crawl_website_tool.firecrawl_crawl_website_too
     FirecrawlCrawlWebsiteTool,
 )
 
-@pytest.mark.vcr(filter_headers=["authorization"])
+@pytest.mark.vcr()
 def test_firecrawl_crawl_tool_integration():
     tool = FirecrawlCrawlWebsiteTool(config={
         "limit": 2,

lib/crewai-tools/tests/tools/firecrawl_scrape_website_tool_test.py

@@ -4,7 +4,7 @@ from crewai_tools.tools.firecrawl_scrape_website_tool.firecrawl_scrape_website_t
     FirecrawlScrapeWebsiteTool,
 )
 
-@pytest.mark.vcr(filter_headers=["authorization"])
+@pytest.mark.vcr()
 def test_firecrawl_scrape_tool_integration():
     tool = FirecrawlScrapeWebsiteTool()
     result = tool.run(url="https://firecrawl.dev")

lib/crewai-tools/tests/tools/firecrawl_search_tool_test.py

@@ -3,7 +3,7 @@ import pytest
 from crewai_tools.tools.firecrawl_search_tool.firecrawl_search_tool import FirecrawlSearchTool
 
 
-@pytest.mark.vcr(filter_headers=["authorization"])
+@pytest.mark.vcr()
 def test_firecrawl_search_tool_integration():
     tool = FirecrawlSearchTool()
     result = tool.run(query="firecrawl")

lib/crewai-tools/tests/tools/test_search_tools.py

@@ -23,15 +23,13 @@ from crewai_tools.tools.rag.rag_tool import Adapter
 import pytest
 
 
-pytestmark = [pytest.mark.vcr(filter_headers=["authorization"])]
-
-
 @pytest.fixture
 def mock_adapter():
     mock_adapter = MagicMock(spec=Adapter)
     return mock_adapter
 
 
+@pytest.mark.vcr()
 def test_directory_search_tool():
     with tempfile.TemporaryDirectory() as temp_dir:
         test_file = Path(temp_dir) / "test.txt"
@@ -65,6 +63,7 @@ def test_pdf_search_tool(mock_adapter):
     )
 
 
+@pytest.mark.vcr()
 def test_txt_search_tool():
     with tempfile.NamedTemporaryFile(suffix=".txt", delete=False) as temp_file:
         temp_file.write(b"This is a test file for txt search")
@@ -102,6 +101,7 @@ def test_docx_search_tool(mock_adapter):
     )
 
 
+@pytest.mark.vcr()
 def test_json_search_tool():
     with tempfile.NamedTemporaryFile(suffix=".json", delete=False) as temp_file:
         temp_file.write(b'{"test": "This is a test JSON file"}')
@@ -127,6 +127,7 @@ def test_xml_search_tool(mock_adapter):
     )
 
 
+@pytest.mark.vcr()
 def test_csv_search_tool():
     with tempfile.NamedTemporaryFile(suffix=".csv", delete=False) as temp_file:
         temp_file.write(b"name,description\ntest,This is a test CSV file")
@@ -141,6 +142,7 @@ def test_csv_search_tool():
         os.unlink(temp_file_path)
 
 
+@pytest.mark.vcr()
 def test_mdx_search_tool():
     with tempfile.NamedTemporaryFile(suffix=".mdx", delete=False) as temp_file:
         temp_file.write(b"# Test MDX\nThis is a test MDX file")

lib/crewai/pyproject.toml

@@ -9,35 +9,36 @@ authors = [
 requires-python = ">=3.10, <3.14"
 dependencies = [
     # Core Dependencies
-    "pydantic>=2.11.9",
-    "openai>=1.13.3",
+    "pydantic~=2.11.9",
+    "openai~=1.83.0",
     "instructor>=1.3.3",
     # Text Processing
-    "pdfplumber>=0.11.4",
-    "regex>=2024.9.11",
+    "pdfplumber~=0.11.4",
+    "regex~=2024.9.11",
     # Telemetry and Monitoring
-    "opentelemetry-api>=1.30.0",
-    "opentelemetry-sdk>=1.30.0",
-    "opentelemetry-exporter-otlp-proto-http>=1.30.0",
+    "opentelemetry-api~=1.34.0",
+    "opentelemetry-sdk~=1.34.0",
+    "opentelemetry-exporter-otlp-proto-http~=1.34.0",
     # Data Handling
     "chromadb~=1.1.0",
-    "tokenizers>=0.20.3",
-    "openpyxl>=3.1.5",
+    "tokenizers~=0.20.3",
+    "openpyxl~=3.1.5",
     # Authentication and Security
-    "python-dotenv>=1.1.1",
-    "pyjwt>=2.9.0",
+    "python-dotenv~=1.1.1",
+    "pyjwt~=2.9.0",
     # Configuration and Utils
-    "click>=8.1.7",
-    "appdirs>=1.4.4",
-    "jsonref>=1.1.0",
-    "json-repair==0.25.2",
-    "uv>=0.4.25",
-    "tomli-w>=1.1.0",
-    "tomli>=2.0.2",
-    "json5>=0.10.0",
-    "portalocker==2.7.0",
-    "pydantic-settings>=2.10.1",
-    "mcp>=1.16.0",
+    "click~=8.1.7",
+    "appdirs~=1.4.4",
+    "jsonref~=1.1.0",
+    "json-repair~=0.25.2",
+    "tomli-w~=1.1.0",
+    "tomli~=2.0.2",
+    "json5~=0.10.0",
+    "portalocker~=2.7.0",
+    "pydantic-settings~=2.10.1",
+    "mcp~=1.16.0",
+    "uv~=0.9.13",
+    "aiosqlite~=0.21.0",
 ]
 
 [project.urls]
@@ -48,55 +49,54 @@ Repository = "https://github.com/crewAIInc/crewAI"
 
 [project.optional-dependencies]
 tools = [
-    "crewai-tools==1.6.1",
+    "crewai-tools==1.7.0",
 ]
 embeddings = [
     "tiktoken~=0.8.0"
 ]
-pdfplumber = [
-    "pdfplumber>=0.11.4",
-]
 pandas = [
-    "pandas>=2.2.3",
+    "pandas~=2.2.3",
 ]
 openpyxl = [
-    "openpyxl>=3.1.5",
+    "openpyxl~=3.1.5",
 ]
-mem0 = ["mem0ai>=0.1.94"]
+mem0 = ["mem0ai~=0.1.94"]
 docling = [
-    "docling>=2.12.0",
+    "docling~=2.63.0",
 ]
 qdrant = [
-    "qdrant-client[fastembed]>=1.14.3",
+    "qdrant-client[fastembed]~=1.14.3",
 ]
 aws = [
-    "boto3>=1.40.38",
+    "boto3~=1.40.38",
+    "aiobotocore~=2.25.2",
 ]
 watson = [
-    "ibm-watsonx-ai>=1.3.39",
+    "ibm-watsonx-ai~=1.3.39",
 ]
 voyageai = [
-    "voyageai>=0.3.5",
+    "voyageai~=0.3.5",
 ]
 litellm = [
-    "litellm>=1.74.9",
+    "litellm~=1.74.9",
 ]
 bedrock = [
-    "boto3>=1.40.45",
+    "boto3~=1.40.45",
 ]
 google-genai = [
-    "google-genai>=1.2.0",
+    "google-genai~=1.49.0",
 ]
 azure-ai-inference = [
-    "azure-ai-inference>=1.0.0b9",
+    "azure-ai-inference~=1.0.0b9",
 ]
 anthropic = [
-    "anthropic>=0.69.0",
+    "anthropic~=0.71.0",
 ]
- a2a = [
+a2a = [
      "a2a-sdk~=0.3.10",
-     "httpx-auth>=0.23.1",
-     "httpx-sse>=0.4.0",
+     "httpx-auth~=0.23.1",
+     "httpx-sse~=0.4.0",
+     "aiocache[redis,memcached]~=0.12.3",
 ]
 
 

lib/crewai/src/crewai/__init__.py

@@ -40,7 +40,7 @@ def _suppress_pydantic_deprecation_warnings() -> None:
 
 _suppress_pydantic_deprecation_warnings()
 
-__version__ = "1.6.1"
+__version__ = "1.7.0"
 _telemetry_submitted = False
 
 

lib/crewai/src/crewai/a2a/extensions/__init__.py

@@ -0,0 +1,4 @@
+"""A2A Protocol Extensions for CrewAI.
+
+This module contains extensions to the A2A (Agent-to-Agent) protocol.
+"""

lib/crewai/src/crewai/a2a/extensions/base.py

@@ -0,0 +1,193 @@
+"""Base extension interface for A2A wrapper integrations.
+
+This module defines the protocol for extending A2A wrapper functionality
+with custom logic for conversation processing, prompt augmentation, and
+agent response handling.
+"""
+
+from __future__ import annotations
+
+from collections.abc import Sequence
+from typing import TYPE_CHECKING, Any, Protocol
+
+
+if TYPE_CHECKING:
+    from a2a.types import Message
+
+    from crewai.agent.core import Agent
+
+
+class ConversationState(Protocol):
+    """Protocol for extension-specific conversation state.
+
+    Extensions can define their own state classes that implement this protocol
+    to track conversation-specific data extracted from message history.
+    """
+
+    def is_ready(self) -> bool:
+        """Check if the state indicates readiness for some action.
+
+        Returns:
+            True if the state is ready, False otherwise.
+        """
+        ...
+
+
+class A2AExtension(Protocol):
+    """Protocol for A2A wrapper extensions.
+
+    Extensions can implement this protocol to inject custom logic into
+    the A2A conversation flow at various integration points.
+    """
+
+    def inject_tools(self, agent: Agent) -> None:
+        """Inject extension-specific tools into the agent.
+
+        Called when an agent is wrapped with A2A capabilities. Extensions
+        can add tools that enable extension-specific functionality.
+
+        Args:
+            agent: The agent instance to inject tools into.
+        """
+        ...
+
+    def extract_state_from_history(
+        self, conversation_history: Sequence[Message]
+    ) -> ConversationState | None:
+        """Extract extension-specific state from conversation history.
+
+        Called during prompt augmentation to allow extensions to analyze
+        the conversation history and extract relevant state information.
+
+        Args:
+            conversation_history: The sequence of A2A messages exchanged.
+
+        Returns:
+            Extension-specific conversation state, or None if no relevant state.
+        """
+        ...
+
+    def augment_prompt(
+        self,
+        base_prompt: str,
+        conversation_state: ConversationState | None,
+    ) -> str:
+        """Augment the task prompt with extension-specific instructions.
+
+        Called during prompt augmentation to allow extensions to add
+        custom instructions based on conversation state.
+
+        Args:
+            base_prompt: The base prompt to augment.
+            conversation_state: Extension-specific state from extract_state_from_history.
+
+        Returns:
+            The augmented prompt with extension-specific instructions.
+        """
+        ...
+
+    def process_response(
+        self,
+        agent_response: Any,
+        conversation_state: ConversationState | None,
+    ) -> Any:
+        """Process and potentially modify the agent response.
+
+        Called after parsing the agent's response, allowing extensions to
+        enhance or modify the response based on conversation state.
+
+        Args:
+            agent_response: The parsed agent response.
+            conversation_state: Extension-specific state from extract_state_from_history.
+
+        Returns:
+            The processed agent response (may be modified or original).
+        """
+        ...
+
+
+class ExtensionRegistry:
+    """Registry for managing A2A extensions.
+
+    Maintains a collection of extensions and provides methods to invoke
+    their hooks at various integration points.
+    """
+
+    def __init__(self) -> None:
+        """Initialize the extension registry."""
+        self._extensions: list[A2AExtension] = []
+
+    def register(self, extension: A2AExtension) -> None:
+        """Register an extension.
+
+        Args:
+            extension: The extension to register.
+        """
+        self._extensions.append(extension)
+
+    def inject_all_tools(self, agent: Agent) -> None:
+        """Inject tools from all registered extensions.
+
+        Args:
+            agent: The agent instance to inject tools into.
+        """
+        for extension in self._extensions:
+            extension.inject_tools(agent)
+
+    def extract_all_states(
+        self, conversation_history: Sequence[Message]
+    ) -> dict[type[A2AExtension], ConversationState]:
+        """Extract conversation states from all registered extensions.
+
+        Args:
+            conversation_history: The sequence of A2A messages exchanged.
+
+        Returns:
+            Mapping of extension types to their conversation states.
+        """
+        states: dict[type[A2AExtension], ConversationState] = {}
+        for extension in self._extensions:
+            state = extension.extract_state_from_history(conversation_history)
+            if state is not None:
+                states[type(extension)] = state
+        return states
+
+    def augment_prompt_with_all(
+        self,
+        base_prompt: str,
+        extension_states: dict[type[A2AExtension], ConversationState],
+    ) -> str:
+        """Augment prompt with instructions from all registered extensions.
+
+        Args:
+            base_prompt: The base prompt to augment.
+            extension_states: Mapping of extension types to conversation states.
+
+        Returns:
+            The fully augmented prompt.
+        """
+        augmented = base_prompt
+        for extension in self._extensions:
+            state = extension_states.get(type(extension))
+            augmented = extension.augment_prompt(augmented, state)
+        return augmented
+
+    def process_response_with_all(
+        self,
+        agent_response: Any,
+        extension_states: dict[type[A2AExtension], ConversationState],
+    ) -> Any:
+        """Process response through all registered extensions.
+
+        Args:
+            agent_response: The parsed agent response.
+            extension_states: Mapping of extension types to conversation states.
+
+        Returns:
+            The processed agent response.
+        """
+        processed = agent_response
+        for extension in self._extensions:
+            state = extension_states.get(type(extension))
+            processed = extension.process_response(processed, state)
+        return processed

lib/crewai/src/crewai/a2a/extensions/registry.py

@@ -0,0 +1,34 @@
+"""Extension registry factory for A2A configurations.
+
+This module provides utilities for creating extension registries from A2A configurations.
+"""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING
+
+from crewai.a2a.extensions.base import ExtensionRegistry
+
+
+if TYPE_CHECKING:
+    from crewai.a2a.config import A2AConfig
+
+
+def create_extension_registry_from_config(
+    a2a_config: list[A2AConfig] | A2AConfig,
+) -> ExtensionRegistry:
+    """Create an extension registry from A2A configuration.
+
+    Args:
+        a2a_config: A2A configuration (single or list)
+
+    Returns:
+        Configured extension registry with all applicable extensions
+    """
+    registry = ExtensionRegistry()
+    configs = a2a_config if isinstance(a2a_config, list) else [a2a_config]
+
+    for _ in configs:
+        pass
+
+    return registry

lib/crewai/src/crewai/a2a/utils.py

@@ -23,6 +23,8 @@ from a2a.types import (
     TextPart,
     TransportProtocol,
 )
+from aiocache import cached  # type: ignore[import-untyped]
+from aiocache.serializers import PickleSerializer  # type: ignore[import-untyped]
 import httpx
 from pydantic import BaseModel, Field, create_model
 
@@ -65,7 +67,7 @@ def _fetch_agent_card_cached(
         endpoint: A2A agent endpoint URL
         auth_hash: Hash of the auth object
         timeout: Request timeout
-        _ttl_hash: Time-based hash for cache invalidation (unused in body)
+        _ttl_hash: Time-based hash for cache invalidation
 
     Returns:
         Cached AgentCard
@@ -106,7 +108,18 @@ def fetch_agent_card(
         A2AClientHTTPError: If authentication fails
     """
     if use_cache:
-        auth_hash = hash((type(auth).__name__, id(auth))) if auth else 0
+        if auth:
+            auth_data = auth.model_dump_json(
+                exclude={
+                    "_access_token",
+                    "_token_expires_at",
+                    "_refresh_token",
+                    "_authorization_callback",
+                }
+            )
+            auth_hash = hash((type(auth).__name__, auth_data))
+        else:
+            auth_hash = 0
         _auth_store[auth_hash] = auth
         ttl_hash = int(time.time() // cache_ttl)
         return _fetch_agent_card_cached(endpoint, auth_hash, timeout, ttl_hash)
@@ -121,6 +134,26 @@ def fetch_agent_card(
         loop.close()
 
 
+@cached(ttl=300, serializer=PickleSerializer())  # type: ignore[untyped-decorator]
+async def _fetch_agent_card_async_cached(
+    endpoint: str,
+    auth_hash: int,
+    timeout: int,
+) -> AgentCard:
+    """Cached async implementation of AgentCard fetching.
+
+    Args:
+        endpoint: A2A agent endpoint URL
+        auth_hash: Hash of the auth object
+        timeout: Request timeout in seconds
+
+    Returns:
+        Cached AgentCard object
+    """
+    auth = _auth_store.get(auth_hash)
+    return await _fetch_agent_card_async(endpoint=endpoint, auth=auth, timeout=timeout)
+
+
 async def _fetch_agent_card_async(
     endpoint: str,
     auth: AuthScheme | None,
@@ -339,7 +372,22 @@ async def _execute_a2a_delegation_async(
     Returns:
         Dictionary with status, result/error, and new history
     """
-    agent_card = await _fetch_agent_card_async(endpoint, auth, timeout)
+    if auth:
+        auth_data = auth.model_dump_json(
+            exclude={
+                "_access_token",
+                "_token_expires_at",
+                "_refresh_token",
+                "_authorization_callback",
+            }
+        )
+        auth_hash = hash((type(auth).__name__, auth_data))
+    else:
+        auth_hash = 0
+    _auth_store[auth_hash] = auth
+    agent_card = await _fetch_agent_card_async_cached(
+        endpoint=endpoint, auth_hash=auth_hash, timeout=timeout
+    )
 
     validate_auth_against_agent_card(agent_card, auth)
 
@@ -556,6 +604,34 @@ async def _execute_a2a_delegation_async(
                         }
                         break
         except Exception as e:
+            if isinstance(e, A2AClientHTTPError):
+                error_msg = f"HTTP Error {e.status_code}: {e!s}"
+
+                error_message = Message(
+                    role=Role.agent,
+                    message_id=str(uuid.uuid4()),
+                    parts=[Part(root=TextPart(text=error_msg))],
+                    context_id=context_id,
+                    task_id=task_id,
+                )
+                new_messages.append(error_message)
+
+                crewai_event_bus.emit(
+                    None,
+                    A2AResponseReceivedEvent(
+                        response=error_msg,
+                        turn_number=turn_number,
+                        is_multiturn=is_multiturn,
+                        status="failed",
+                        agent_role=agent_role,
+                    ),
+                )
+                return {
+                    "status": "failed",
+                    "error": error_msg,
+                    "history": new_messages,
+                }
+
             current_exception: Exception | BaseException | None = e
             while current_exception:
                 if hasattr(current_exception, "response"):
@@ -752,4 +828,5 @@ def get_a2a_agents_and_response_model(
         Tuple of A2A agent IDs and response model
     """
     a2a_agents, agent_ids = extract_a2a_agent_ids_from_config(a2a_config=a2a_config)
+
     return a2a_agents, create_agent_response_model(agent_ids)

lib/crewai/src/crewai/a2a/wrapper.py

@@ -15,6 +15,7 @@ from a2a.types import Role
 from pydantic import BaseModel, ValidationError
 
 from crewai.a2a.config import A2AConfig
+from crewai.a2a.extensions.base import ExtensionRegistry
 from crewai.a2a.templates import (
     AVAILABLE_AGENTS_TEMPLATE,
     CONVERSATION_TURN_INFO_TEMPLATE,
@@ -42,7 +43,9 @@ if TYPE_CHECKING:
     from crewai.tools.base_tool import BaseTool
 
 
-def wrap_agent_with_a2a_instance(agent: Agent) -> None:
+def wrap_agent_with_a2a_instance(
+    agent: Agent, extension_registry: ExtensionRegistry | None = None
+) -> None:
     """Wrap an agent instance's execute_task method with A2A support.
 
     This function modifies the agent instance by wrapping its execute_task
@@ -51,7 +54,13 @@ def wrap_agent_with_a2a_instance(agent: Agent) -> None:
 
     Args:
         agent: The agent instance to wrap
+        extension_registry: Optional registry of A2A extensions for injecting tools and custom logic
     """
+    if extension_registry is None:
+        extension_registry = ExtensionRegistry()
+
+    extension_registry.inject_all_tools(agent)
+
     original_execute_task = agent.execute_task.__func__  # type: ignore[attr-defined]
 
     @wraps(original_execute_task)
@@ -85,6 +94,7 @@ def wrap_agent_with_a2a_instance(agent: Agent) -> None:
             agent_response_model=agent_response_model,
             context=context,
             tools=tools,
+            extension_registry=extension_registry,
         )
 
     object.__setattr__(agent, "execute_task", MethodType(execute_task_with_a2a, agent))
@@ -154,6 +164,7 @@ def _execute_task_with_a2a(
     agent_response_model: type[BaseModel],
     context: str | None,
     tools: list[BaseTool] | None,
+    extension_registry: ExtensionRegistry,
 ) -> str:
     """Wrap execute_task with A2A delegation logic.
 
@@ -165,6 +176,7 @@ def _execute_task_with_a2a(
         context: Optional context for task execution
         tools: Optional tools available to the agent
         agent_response_model: Optional agent response model
+        extension_registry: Registry of A2A extensions
 
     Returns:
         Task execution result (either from LLM or A2A agent)
@@ -190,11 +202,12 @@ def _execute_task_with_a2a(
         finally:
             task.description = original_description
 
-    task.description = _augment_prompt_with_a2a(
+    task.description, _ = _augment_prompt_with_a2a(
         a2a_agents=a2a_agents,
         task_description=original_description,
         agent_cards=agent_cards,
         failed_agents=failed_agents,
+        extension_registry=extension_registry,
     )
     task.response_model = agent_response_model
 
@@ -204,6 +217,11 @@ def _execute_task_with_a2a(
             raw_result=raw_result, agent_response_model=agent_response_model
         )
 
+        if extension_registry and isinstance(agent_response, BaseModel):
+            agent_response = extension_registry.process_response_with_all(
+                agent_response, {}
+            )
+
         if isinstance(agent_response, BaseModel) and isinstance(
             agent_response, AgentResponseProtocol
         ):
@@ -217,6 +235,7 @@ def _execute_task_with_a2a(
                     tools=tools,
                     agent_cards=agent_cards,
                     original_task_description=original_description,
+                    extension_registry=extension_registry,
                 )
             return str(agent_response.message)
 
@@ -235,7 +254,8 @@ def _augment_prompt_with_a2a(
     turn_num: int = 0,
     max_turns: int | None = None,
     failed_agents: dict[str, str] | None = None,
-) -> str:
+    extension_registry: ExtensionRegistry | None = None,
+) -> tuple[str, bool]:
     """Add A2A delegation instructions to prompt.
 
     Args:
@@ -246,13 +266,14 @@ def _augment_prompt_with_a2a(
         turn_num: Current turn number (0-indexed)
         max_turns: Maximum allowed turns (from config)
         failed_agents: Dictionary mapping failed agent endpoints to error messages
+        extension_registry: Optional registry of A2A extensions
 
     Returns:
-        Augmented task description with A2A instructions
+        Tuple of (augmented prompt, disable_structured_output flag)
     """
 
     if not agent_cards:
-        return task_description
+        return task_description, False
 
     agents_text = ""
 
@@ -270,6 +291,7 @@ def _augment_prompt_with_a2a(
     agents_text = AVAILABLE_AGENTS_TEMPLATE.substitute(available_a2a_agents=agents_text)
 
     history_text = ""
+
     if conversation_history:
         for msg in conversation_history:
             history_text += f"\n{msg.model_dump_json(indent=2, exclude_none=True, exclude={'message_id'})}\n"
@@ -277,6 +299,15 @@ def _augment_prompt_with_a2a(
     history_text = PREVIOUS_A2A_CONVERSATION_TEMPLATE.substitute(
         previous_a2a_conversation=history_text
     )
+
+    extension_states = {}
+    disable_structured_output = False
+    if extension_registry and conversation_history:
+        extension_states = extension_registry.extract_all_states(conversation_history)
+        for state in extension_states.values():
+            if state.is_ready():
+                disable_structured_output = True
+                break
     turn_info = ""
 
     if max_turns is not None and conversation_history:
@@ -296,16 +327,22 @@ def _augment_prompt_with_a2a(
             warning=warning,
         )
 
-    return f"""{task_description}
+    augmented_prompt = f"""{task_description}
 
 IMPORTANT: You have the ability to delegate this task to remote A2A agents.
-
 {agents_text}
 {history_text}{turn_info}
 
 
 """
 
+    if extension_registry:
+        augmented_prompt = extension_registry.augment_prompt_with_all(
+            augmented_prompt, extension_states
+        )
+
+    return augmented_prompt, disable_structured_output
+
 
 def _parse_agent_response(
     raw_result: str | dict[str, Any], agent_response_model: type[BaseModel]
@@ -373,7 +410,7 @@ def _handle_agent_response_and_continue(
     if "agent_card" in a2a_result and agent_id not in agent_cards_dict:
         agent_cards_dict[agent_id] = a2a_result["agent_card"]
 
-    task.description = _augment_prompt_with_a2a(
+    task.description, disable_structured_output = _augment_prompt_with_a2a(
         a2a_agents=a2a_agents,
         task_description=original_task_description,
         conversation_history=conversation_history,
@@ -382,7 +419,38 @@ def _handle_agent_response_and_continue(
         agent_cards=agent_cards_dict,
     )
 
+    original_response_model = task.response_model
+    if disable_structured_output:
+        task.response_model = None
+
     raw_result = original_fn(self, task, context, tools)
+
+    if disable_structured_output:
+        task.response_model = original_response_model
+
+    if disable_structured_output:
+        final_turn_number = turn_num + 1
+        result_text = str(raw_result)
+        crewai_event_bus.emit(
+            None,
+            A2AMessageSentEvent(
+                message=result_text,
+                turn_number=final_turn_number,
+                is_multiturn=True,
+                agent_role=self.role,
+            ),
+        )
+        crewai_event_bus.emit(
+            None,
+            A2AConversationCompletedEvent(
+                status="completed",
+                final_result=result_text,
+                error=None,
+                total_turns=final_turn_number,
+            ),
+        )
+        return result_text, None
+
     llm_response = _parse_agent_response(
         raw_result=raw_result, agent_response_model=agent_response_model
     )
@@ -425,6 +493,7 @@ def _delegate_to_a2a(
     tools: list[BaseTool] | None,
     agent_cards: dict[str, AgentCard] | None = None,
     original_task_description: str | None = None,
+    extension_registry: ExtensionRegistry | None = None,
 ) -> str:
     """Delegate to A2A agent with multi-turn conversation support.
 
@@ -437,6 +506,7 @@ def _delegate_to_a2a(
         tools: Optional tools available to the agent
         agent_cards: Pre-fetched agent cards from _execute_task_with_a2a
         original_task_description: The original task description before A2A augmentation
+        extension_registry: Optional registry of A2A extensions
 
     Returns:
         Result from A2A agent
@@ -447,9 +517,13 @@ def _delegate_to_a2a(
     a2a_agents, agent_response_model = get_a2a_agents_and_response_model(self.a2a)
     agent_ids = tuple(config.endpoint for config in a2a_agents)
     current_request = str(agent_response.message)
-    agent_id = agent_response.a2a_ids[0]
 
-    if agent_id not in agent_ids:
+    if hasattr(agent_response, "a2a_ids") and agent_response.a2a_ids:
+        agent_id = agent_response.a2a_ids[0]
+    else:
+        agent_id = agent_ids[0] if agent_ids else ""
+
+    if agent_id and agent_id not in agent_ids:
         raise ValueError(
             f"Unknown A2A agent ID(s): {agent_response.a2a_ids} not in {agent_ids}"
         )
@@ -458,10 +532,11 @@ def _delegate_to_a2a(
     task_config = task.config or {}
     context_id = task_config.get("context_id")
     task_id_config = task_config.get("task_id")
-    reference_task_ids = task_config.get("reference_task_ids")
     metadata = task_config.get("metadata")
     extensions = task_config.get("extensions")
 
+    reference_task_ids = task_config.get("reference_task_ids", [])
+
     if original_task_description is None:
         original_task_description = task.description
 
@@ -497,11 +572,27 @@ def _delegate_to_a2a(
 
             conversation_history = a2a_result.get("history", [])
 
+            if conversation_history:
+                latest_message = conversation_history[-1]
+                if latest_message.task_id is not None:
+                    task_id_config = latest_message.task_id
+                if latest_message.context_id is not None:
+                    context_id = latest_message.context_id
+
             if a2a_result["status"] in ["completed", "input_required"]:
                 if (
                     a2a_result["status"] == "completed"
                     and agent_config.trust_remote_completion_status
                 ):
+                    if (
+                        task_id_config is not None
+                        and task_id_config not in reference_task_ids
+                    ):
+                        reference_task_ids.append(task_id_config)
+                        if task.config is None:
+                            task.config = {}
+                        task.config["reference_task_ids"] = reference_task_ids
+
                     result_text = a2a_result.get("result", "")
                     final_turn_number = turn_num + 1
                     crewai_event_bus.emit(
@@ -513,7 +604,7 @@ def _delegate_to_a2a(
                             total_turns=final_turn_number,
                         ),
                     )
-                    return result_text  # type: ignore[no-any-return]
+                    return cast(str, result_text)
 
                 final_result, next_request = _handle_agent_response_and_continue(
                     self=self,
@@ -541,6 +632,31 @@ def _delegate_to_a2a(
                 continue
 
             error_msg = a2a_result.get("error", "Unknown error")
+
+            final_result, next_request = _handle_agent_response_and_continue(
+                self=self,
+                a2a_result=a2a_result,
+                agent_id=agent_id,
+                agent_cards=agent_cards,
+                a2a_agents=a2a_agents,
+                original_task_description=original_task_description,
+                conversation_history=conversation_history,
+                turn_num=turn_num,
+                max_turns=max_turns,
+                task=task,
+                original_fn=original_fn,
+                context=context,
+                tools=tools,
+                agent_response_model=agent_response_model,
+            )
+
+            if final_result is not None:
+                return final_result
+
+            if next_request is not None:
+                current_request = next_request
+                continue
+
             crewai_event_bus.emit(
                 None,
                 A2AConversationCompletedEvent(
@@ -550,7 +666,7 @@ def _delegate_to_a2a(
                     total_turns=turn_num + 1,
                 ),
             )
-            raise Exception(f"A2A delegation failed: {error_msg}")
+            return f"A2A delegation failed: {error_msg}"
 
         if conversation_history:
             for msg in reversed(conversation_history):

lib/crewai/src/crewai/agent/core.py

@@ -2,7 +2,6 @@ from __future__ import annotations
 
 import asyncio
 from collections.abc import Sequence
-import json
 import shutil
 import subprocess
 import time
@@ -19,6 +18,19 @@ from pydantic import BaseModel, Field, InstanceOf, PrivateAttr, model_validator
 from typing_extensions import Self
 
 from crewai.a2a.config import A2AConfig
+from crewai.agent.utils import (
+    ahandle_knowledge_retrieval,
+    apply_training_data,
+    build_task_prompt_with_schema,
+    format_task_with_context,
+    get_knowledge_config,
+    handle_knowledge_retrieval,
+    handle_reasoning,
+    prepare_tools,
+    process_tool_results,
+    save_last_messages,
+    validate_max_execution_time,
+)
 from crewai.agents.agent_builder.base_agent import BaseAgent
 from crewai.agents.cache.cache_handler import CacheHandler
 from crewai.agents.crew_agent_executor import CrewAgentExecutor
@@ -27,9 +39,6 @@ from crewai.events.types.knowledge_events import (
     KnowledgeQueryCompletedEvent,
     KnowledgeQueryFailedEvent,
     KnowledgeQueryStartedEvent,
-    KnowledgeRetrievalCompletedEvent,
-    KnowledgeRetrievalStartedEvent,
-    KnowledgeSearchQueryFailedEvent,
 )
 from crewai.events.types.memory_events import (
     MemoryRetrievalCompletedEvent,
@@ -37,7 +46,6 @@ from crewai.events.types.memory_events import (
 )
 from crewai.knowledge.knowledge import Knowledge
 from crewai.knowledge.source.base_knowledge_source import BaseKnowledgeSource
-from crewai.knowledge.utils.knowledge_utils import extract_knowledge_context
 from crewai.lite_agent import LiteAgent
 from crewai.llms.base_llm import BaseLLM
 from crewai.mcp import (
@@ -61,7 +69,7 @@ from crewai.utilities.agent_utils import (
     render_text_description_and_args,
 )
 from crewai.utilities.constants import TRAINED_AGENTS_DATA_FILE, TRAINING_DATA_FILE
-from crewai.utilities.converter import Converter, generate_model_description
+from crewai.utilities.converter import Converter
 from crewai.utilities.guardrail_types import GuardrailType
 from crewai.utilities.llm_utils import create_llm
 from crewai.utilities.prompts import Prompts
@@ -295,53 +303,15 @@ class Agent(BaseAgent):
             ValueError: If the max execution time is not a positive integer.
             RuntimeError: If the agent execution fails for other reasons.
         """
-        if self.reasoning:
-            try:
-                from crewai.utilities.reasoning_handler import (
-                    AgentReasoning,
-                    AgentReasoningOutput,
-                )
-
-                reasoning_handler = AgentReasoning(task=task, agent=self)
-                reasoning_output: AgentReasoningOutput = (
-                    reasoning_handler.handle_agent_reasoning()
-                )
-
-                # Add the reasoning plan to the task description
-                task.description += f"\n\nReasoning Plan:\n{reasoning_output.plan.plan}"
-            except Exception as e:
-                self._logger.log("error", f"Error during reasoning process: {e!s}")
+        handle_reasoning(self, task)
         self._inject_date_to_task(task)
 
         if self.tools_handler:
             self.tools_handler.last_used_tool = None
 
         task_prompt = task.prompt()
-
-        # If the task requires output in JSON or Pydantic format,
-        # append specific instructions to the task prompt to ensure
-        # that the final answer does not include any code block markers
-        # Skip this if task.response_model is set, as native structured outputs handle schema automatically
-        if (task.output_json or task.output_pydantic) and not task.response_model:
-            # Generate the schema based on the output format
-            if task.output_json:
-                schema_dict = generate_model_description(task.output_json)
-                schema = json.dumps(schema_dict["json_schema"]["schema"], indent=2)
-                task_prompt += "\n" + self.i18n.slice(
-                    "formatted_task_instructions"
-                ).format(output_format=schema)
-
-            elif task.output_pydantic:
-                schema_dict = generate_model_description(task.output_pydantic)
-                schema = json.dumps(schema_dict["json_schema"]["schema"], indent=2)
-                task_prompt += "\n" + self.i18n.slice(
-                    "formatted_task_instructions"
-                ).format(output_format=schema)
-
-        if context:
-            task_prompt = self.i18n.slice("task_with_context").format(
-                task=task_prompt, context=context
-            )
+        task_prompt = build_task_prompt_with_schema(task, task_prompt, self.i18n)
+        task_prompt = format_task_with_context(task_prompt, context, self.i18n)
 
         if self._is_any_available_memory():
             crewai_event_bus.emit(
@@ -379,84 +349,20 @@ class Agent(BaseAgent):
                     from_task=task,
                 ),
             )
-        knowledge_config = (
-            self.knowledge_config.model_dump() if self.knowledge_config else {}
+
+        knowledge_config = get_knowledge_config(self)
+        task_prompt = handle_knowledge_retrieval(
+            self,
+            task,
+            task_prompt,
+            knowledge_config,
+            self.knowledge.query if self.knowledge else lambda *a, **k: None,
+            self.crew.query_knowledge if self.crew else lambda *a, **k: None,
         )
 
-        if self.knowledge or (self.crew and self.crew.knowledge):
-            crewai_event_bus.emit(
-                self,
-                event=KnowledgeRetrievalStartedEvent(
-                    from_task=task,
-                    from_agent=self,
-                ),
-            )
-            try:
-                self.knowledge_search_query = self._get_knowledge_search_query(
-                    task_prompt, task
-                )
-                if self.knowledge_search_query:
-                    # Quering agent specific knowledge
-                    if self.knowledge:
-                        agent_knowledge_snippets = self.knowledge.query(
-                            [self.knowledge_search_query], **knowledge_config
-                        )
-                        if agent_knowledge_snippets:
-                            self.agent_knowledge_context = extract_knowledge_context(
-                                agent_knowledge_snippets
-                            )
-                            if self.agent_knowledge_context:
-                                task_prompt += self.agent_knowledge_context
+        prepare_tools(self, tools, task)
+        task_prompt = apply_training_data(self, task_prompt)
 
-                    # Quering crew specific knowledge
-                    knowledge_snippets = self.crew.query_knowledge(
-                        [self.knowledge_search_query], **knowledge_config
-                    )
-                    if knowledge_snippets:
-                        self.crew_knowledge_context = extract_knowledge_context(
-                            knowledge_snippets
-                        )
-                        if self.crew_knowledge_context:
-                            task_prompt += self.crew_knowledge_context
-
-                    crewai_event_bus.emit(
-                        self,
-                        event=KnowledgeRetrievalCompletedEvent(
-                            query=self.knowledge_search_query,
-                            from_task=task,
-                            from_agent=self,
-                            retrieved_knowledge=(
-                                (self.agent_knowledge_context or "")
-                                + (
-                                    "\n"
-                                    if self.agent_knowledge_context
-                                    and self.crew_knowledge_context
-                                    else ""
-                                )
-                                + (self.crew_knowledge_context or "")
-                            ),
-                        ),
-                    )
-            except Exception as e:
-                crewai_event_bus.emit(
-                    self,
-                    event=KnowledgeSearchQueryFailedEvent(
-                        query=self.knowledge_search_query or "",
-                        error=str(e),
-                        from_task=task,
-                        from_agent=self,
-                    ),
-                )
-
-        tools = tools or self.tools or []
-        self.create_agent_executor(tools=tools, task=task)
-
-        if self.crew and self.crew._train:
-            task_prompt = self._training_handler(task_prompt=task_prompt)
-        else:
-            task_prompt = self._use_trained_data(task_prompt=task_prompt)
-
-        # Import agent events locally to avoid circular imports
         from crewai.events.types.agent_events import (
             AgentExecutionCompletedEvent,
             AgentExecutionErrorEvent,
@@ -474,15 +380,8 @@ class Agent(BaseAgent):
                 ),
             )
 
-            # Determine execution method based on timeout setting
+            validate_max_execution_time(self.max_execution_time)
             if self.max_execution_time is not None:
-                if (
-                    not isinstance(self.max_execution_time, int)
-                    or self.max_execution_time <= 0
-                ):
-                    raise ValueError(
-                        "Max Execution time must be a positive integer greater than zero"
-                    )
                 result = self._execute_with_timeout(
                     task_prompt, task, self.max_execution_time
                 )
@@ -490,7 +389,6 @@ class Agent(BaseAgent):
                 result = self._execute_without_timeout(task_prompt, task)
 
         except TimeoutError as e:
-            # Propagate TimeoutError without retry
             crewai_event_bus.emit(
                 self,
                 event=AgentExecutionErrorEvent(
@@ -502,7 +400,6 @@ class Agent(BaseAgent):
             raise e
         except Exception as e:
             if e.__class__.__module__.startswith("litellm"):
-                # Do not retry on litellm errors
                 crewai_event_bus.emit(
                     self,
                     event=AgentExecutionErrorEvent(
@@ -528,23 +425,13 @@ class Agent(BaseAgent):
         if self.max_rpm and self._rpm_controller:
             self._rpm_controller.stop_rpm_counter()
 
-        # If there was any tool in self.tools_results that had result_as_answer
-        # set to True, return the results of the last tool that had
-        # result_as_answer set to True
-        for tool_result in self.tools_results:
-            if tool_result.get("result_as_answer", False):
-                result = tool_result["result"]
+        result = process_tool_results(self, result)
         crewai_event_bus.emit(
             self,
             event=AgentExecutionCompletedEvent(agent=self, task=task, output=result),
         )
 
-        self._last_messages = (
-            self.agent_executor.messages.copy()
-            if self.agent_executor and hasattr(self.agent_executor, "messages")
-            else []
-        )
-
+        save_last_messages(self)
         self._cleanup_mcp_clients()
 
         return result
@@ -604,6 +491,208 @@ class Agent(BaseAgent):
             }
         )["output"]
 
+    async def aexecute_task(
+        self,
+        task: Task,
+        context: str | None = None,
+        tools: list[BaseTool] | None = None,
+    ) -> Any:
+        """Execute a task with the agent asynchronously.
+
+        Args:
+            task: Task to execute.
+            context: Context to execute the task in.
+            tools: Tools to use for the task.
+
+        Returns:
+            Output of the agent.
+
+        Raises:
+            TimeoutError: If execution exceeds the maximum execution time.
+            ValueError: If the max execution time is not a positive integer.
+            RuntimeError: If the agent execution fails for other reasons.
+        """
+        handle_reasoning(self, task)
+        self._inject_date_to_task(task)
+
+        if self.tools_handler:
+            self.tools_handler.last_used_tool = None
+
+        task_prompt = task.prompt()
+        task_prompt = build_task_prompt_with_schema(task, task_prompt, self.i18n)
+        task_prompt = format_task_with_context(task_prompt, context, self.i18n)
+
+        if self._is_any_available_memory():
+            crewai_event_bus.emit(
+                self,
+                event=MemoryRetrievalStartedEvent(
+                    task_id=str(task.id) if task else None,
+                    source_type="agent",
+                    from_agent=self,
+                    from_task=task,
+                ),
+            )
+
+            start_time = time.time()
+
+            contextual_memory = ContextualMemory(
+                self.crew._short_term_memory,
+                self.crew._long_term_memory,
+                self.crew._entity_memory,
+                self.crew._external_memory,
+                agent=self,
+                task=task,
+            )
+            memory = await contextual_memory.abuild_context_for_task(
+                task, context or ""
+            )
+            if memory.strip() != "":
+                task_prompt += self.i18n.slice("memory").format(memory=memory)
+
+            crewai_event_bus.emit(
+                self,
+                event=MemoryRetrievalCompletedEvent(
+                    task_id=str(task.id) if task else None,
+                    memory_content=memory,
+                    retrieval_time_ms=(time.time() - start_time) * 1000,
+                    source_type="agent",
+                    from_agent=self,
+                    from_task=task,
+                ),
+            )
+
+        knowledge_config = get_knowledge_config(self)
+        task_prompt = await ahandle_knowledge_retrieval(
+            self, task, task_prompt, knowledge_config
+        )
+
+        prepare_tools(self, tools, task)
+        task_prompt = apply_training_data(self, task_prompt)
+
+        from crewai.events.types.agent_events import (
+            AgentExecutionCompletedEvent,
+            AgentExecutionErrorEvent,
+            AgentExecutionStartedEvent,
+        )
+
+        try:
+            crewai_event_bus.emit(
+                self,
+                event=AgentExecutionStartedEvent(
+                    agent=self,
+                    tools=self.tools,
+                    task_prompt=task_prompt,
+                    task=task,
+                ),
+            )
+
+            validate_max_execution_time(self.max_execution_time)
+            if self.max_execution_time is not None:
+                result = await self._aexecute_with_timeout(
+                    task_prompt, task, self.max_execution_time
+                )
+            else:
+                result = await self._aexecute_without_timeout(task_prompt, task)
+
+        except TimeoutError as e:
+            crewai_event_bus.emit(
+                self,
+                event=AgentExecutionErrorEvent(
+                    agent=self,
+                    task=task,
+                    error=str(e),
+                ),
+            )
+            raise e
+        except Exception as e:
+            if e.__class__.__module__.startswith("litellm"):
+                crewai_event_bus.emit(
+                    self,
+                    event=AgentExecutionErrorEvent(
+                        agent=self,
+                        task=task,
+                        error=str(e),
+                    ),
+                )
+                raise e
+            self._times_executed += 1
+            if self._times_executed > self.max_retry_limit:
+                crewai_event_bus.emit(
+                    self,
+                    event=AgentExecutionErrorEvent(
+                        agent=self,
+                        task=task,
+                        error=str(e),
+                    ),
+                )
+                raise e
+            result = await self.aexecute_task(task, context, tools)
+
+        if self.max_rpm and self._rpm_controller:
+            self._rpm_controller.stop_rpm_counter()
+
+        result = process_tool_results(self, result)
+        crewai_event_bus.emit(
+            self,
+            event=AgentExecutionCompletedEvent(agent=self, task=task, output=result),
+        )
+
+        save_last_messages(self)
+        self._cleanup_mcp_clients()
+
+        return result
+
+    async def _aexecute_with_timeout(
+        self, task_prompt: str, task: Task, timeout: int
+    ) -> Any:
+        """Execute a task with a timeout asynchronously.
+
+        Args:
+            task_prompt: The prompt to send to the agent.
+            task: The task being executed.
+            timeout: Maximum execution time in seconds.
+
+        Returns:
+            The output of the agent.
+
+        Raises:
+            TimeoutError: If execution exceeds the timeout.
+            RuntimeError: If execution fails for other reasons.
+        """
+        try:
+            return await asyncio.wait_for(
+                self._aexecute_without_timeout(task_prompt, task),
+                timeout=timeout,
+            )
+        except asyncio.TimeoutError as e:
+            raise TimeoutError(
+                f"Task '{task.description}' execution timed out after {timeout} seconds. "
+                "Consider increasing max_execution_time or optimizing the task."
+            ) from e
+
+    async def _aexecute_without_timeout(self, task_prompt: str, task: Task) -> Any:
+        """Execute a task without a timeout asynchronously.
+
+        Args:
+            task_prompt: The prompt to send to the agent.
+            task: The task being executed.
+
+        Returns:
+            The output of the agent.
+        """
+        if not self.agent_executor:
+            raise RuntimeError("Agent executor is not initialized.")
+
+        result = await self.agent_executor.ainvoke(
+            {
+                "input": task_prompt,
+                "tool_names": self.agent_executor.tools_names,
+                "tools": self.agent_executor.tools_description,
+                "ask_for_human_input": task.human_input,
+            }
+        )
+        return result["output"]
+
     def create_agent_executor(
         self, tools: list[BaseTool] | None = None, task: Task | None = None
     ) -> None:
@@ -633,7 +722,7 @@ class Agent(BaseAgent):
             )
 
         self.agent_executor = CrewAgentExecutor(
-            llm=self.llm,
+            llm=self.llm,  # type: ignore[arg-type]
             task=task,  # type: ignore[arg-type]
             agent=self,
             crew=self.crew,
@@ -810,6 +899,7 @@ class Agent(BaseAgent):
         from crewai.tools.base_tool import BaseTool
         from crewai.tools.mcp_native_tool import MCPNativeTool
 
+        transport: StdioTransport | HTTPTransport | SSETransport
         if isinstance(mcp_config, MCPServerStdio):
             transport = StdioTransport(
                 command=mcp_config.command,
@@ -903,10 +993,10 @@ class Agent(BaseAgent):
                                 server_name=server_name,
                                 run_context=None,
                             )
-                            if mcp_config.tool_filter(context, tool):
+                            if mcp_config.tool_filter(context, tool):  # type: ignore[call-arg, arg-type]
                                 filtered_tools.append(tool)
                         except (TypeError, AttributeError):
-                            if mcp_config.tool_filter(tool):
+                            if mcp_config.tool_filter(tool):  # type: ignore[call-arg, arg-type]
                                 filtered_tools.append(tool)
                     else:
                         # Not callable - include tool
@@ -981,7 +1071,9 @@ class Agent(BaseAgent):
         path = parsed.path.replace("/", "_").strip("_")
         return f"{domain}_{path}" if path else domain
 
-    def _get_mcp_tool_schemas(self, server_params: dict) -> dict[str, dict]:
+    def _get_mcp_tool_schemas(
+        self, server_params: dict[str, Any]
+    ) -> dict[str, dict[str, Any]]:
         """Get tool schemas from MCP server for wrapper creation with caching."""
         server_url = server_params["url"]
 
@@ -995,7 +1087,7 @@ class Agent(BaseAgent):
                 self._logger.log(
                     "debug", f"Using cached MCP tool schemas for {server_url}"
                 )
-                return cached_data
+                return cached_data  # type: ignore[no-any-return]
 
         try:
             schemas = asyncio.run(self._get_mcp_tool_schemas_async(server_params))
@@ -1013,7 +1105,7 @@ class Agent(BaseAgent):
 
     async def _get_mcp_tool_schemas_async(
         self, server_params: dict[str, Any]
-    ) -> dict[str, dict]:
+    ) -> dict[str, dict[str, Any]]:
         """Async implementation of MCP tool schema retrieval with timeouts and retries."""
         server_url = server_params["url"]
         return await self._retry_mcp_discovery(
@@ -1021,7 +1113,7 @@ class Agent(BaseAgent):
         )
 
     async def _retry_mcp_discovery(
-        self, operation_func, server_url: str
+        self, operation_func: Any, server_url: str
     ) -> dict[str, dict[str, Any]]:
         """Retry MCP discovery operation with exponential backoff, avoiding try-except in loop."""
         last_error = None
@@ -1052,7 +1144,7 @@ class Agent(BaseAgent):
 
     @staticmethod
     async def _attempt_mcp_discovery(
-        operation_func, server_url: str
+        operation_func: Any, server_url: str
     ) -> tuple[dict[str, dict[str, Any]] | None, str, bool]:
         """Attempt single MCP discovery operation and return (result, error_message, should_retry)."""
         try:
@@ -1142,7 +1234,7 @@ class Agent(BaseAgent):
         properties = json_schema.get("properties", {})
         required_fields = json_schema.get("required", [])
 
-        field_definitions = {}
+        field_definitions: dict[str, Any] = {}
 
         for field_name, field_schema in properties.items():
             field_type = self._json_type_to_python(field_schema)
@@ -1162,7 +1254,7 @@ class Agent(BaseAgent):
                 )
 
         model_name = f"{tool_name.replace('-', '_').replace(' ', '_')}Schema"
-        return create_model(model_name, **field_definitions)
+        return create_model(model_name, **field_definitions)  # type: ignore[no-any-return]
 
     def _json_type_to_python(self, field_schema: dict[str, Any]) -> type:
         """Convert JSON Schema type to Python type.
@@ -1177,7 +1269,7 @@ class Agent(BaseAgent):
         json_type = field_schema.get("type")
 
         if "anyOf" in field_schema:
-            types = []
+            types: list[type] = []
             for option in field_schema["anyOf"]:
                 if "const" in option:
                     types.append(str)
@@ -1185,13 +1277,13 @@ class Agent(BaseAgent):
                     types.append(self._json_type_to_python(option))
             unique_types = list(set(types))
             if len(unique_types) > 1:
-                result = unique_types[0]
+                result: Any = unique_types[0]
                 for t in unique_types[1:]:
                     result = result | t
-                return result
+                return result  # type: ignore[no-any-return]
             return unique_types[0]
 
-        type_mapping = {
+        type_mapping: dict[str | None, type] = {
             "string": str,
             "number": float,
             "integer": int,
@@ -1203,7 +1295,7 @@ class Agent(BaseAgent):
         return type_mapping.get(json_type, Any)
 
     @staticmethod
-    def _fetch_amp_mcp_servers(mcp_name: str) -> list[dict]:
+    def _fetch_amp_mcp_servers(mcp_name: str) -> list[dict[str, Any]]:
         """Fetch MCP server configurations from CrewAI AOP API."""
         # TODO: Implement AMP API call to "integrations/mcps" endpoint
         # Should return list of server configs with URLs
@@ -1438,11 +1530,11 @@ class Agent(BaseAgent):
         """
         if self.apps:
             platform_tools = self.get_platform_tools(self.apps)
-            if platform_tools:
+            if platform_tools and self.tools is not None:
                 self.tools.extend(platform_tools)
         if self.mcps:
             mcps = self.get_mcp_tools(self.mcps)
-            if mcps:
+            if mcps and self.tools is not None:
                 self.tools.extend(mcps)
 
         lite_agent = LiteAgent(

lib/crewai/src/crewai/agent/internal/meta.py

@@ -4,9 +4,8 @@ This metaclass enables extension capabilities for agents by detecting
 extension fields in class annotations and applying appropriate wrappers.
 """
 
-import warnings
-from functools import wraps
 from typing import Any
+import warnings
 
 from pydantic import model_validator
 from pydantic._internal._model_construction import ModelMetaclass
@@ -59,9 +58,15 @@ class AgentMeta(ModelMetaclass):
 
                 a2a_value = getattr(self, "a2a", None)
                 if a2a_value is not None:
+                    from crewai.a2a.extensions.registry import (
+                        create_extension_registry_from_config,
+                    )
                     from crewai.a2a.wrapper import wrap_agent_with_a2a_instance
 
-                    wrap_agent_with_a2a_instance(self)
+                    extension_registry = create_extension_registry_from_config(
+                        a2a_value
+                    )
+                    wrap_agent_with_a2a_instance(self, extension_registry)
 
                 return result
 

lib/crewai/src/crewai/agent/utils.py

@@ -0,0 +1,355 @@
+"""Utility functions for agent task execution.
+
+This module contains shared logic extracted from the Agent's execute_task
+and aexecute_task methods to reduce code duplication.
+"""
+
+from __future__ import annotations
+
+import json
+from typing import TYPE_CHECKING, Any
+
+from crewai.events.event_bus import crewai_event_bus
+from crewai.events.types.knowledge_events import (
+    KnowledgeRetrievalCompletedEvent,
+    KnowledgeRetrievalStartedEvent,
+    KnowledgeSearchQueryFailedEvent,
+)
+from crewai.knowledge.utils.knowledge_utils import extract_knowledge_context
+from crewai.utilities.pydantic_schema_utils import generate_model_description
+
+
+if TYPE_CHECKING:
+    from crewai.agent.core import Agent
+    from crewai.task import Task
+    from crewai.tools.base_tool import BaseTool
+    from crewai.utilities.i18n import I18N
+
+
+def handle_reasoning(agent: Agent, task: Task) -> None:
+    """Handle the reasoning process for an agent before task execution.
+
+    Args:
+        agent: The agent performing the task.
+        task: The task to execute.
+    """
+    if not agent.reasoning:
+        return
+
+    try:
+        from crewai.utilities.reasoning_handler import (
+            AgentReasoning,
+            AgentReasoningOutput,
+        )
+
+        reasoning_handler = AgentReasoning(task=task, agent=agent)
+        reasoning_output: AgentReasoningOutput = (
+            reasoning_handler.handle_agent_reasoning()
+        )
+        task.description += f"\n\nReasoning Plan:\n{reasoning_output.plan.plan}"
+    except Exception as e:
+        agent._logger.log("error", f"Error during reasoning process: {e!s}")
+
+
+def build_task_prompt_with_schema(task: Task, task_prompt: str, i18n: I18N) -> str:
+    """Build task prompt with JSON/Pydantic schema instructions if applicable.
+
+    Args:
+        task: The task being executed.
+        task_prompt: The initial task prompt.
+        i18n: Internationalization instance.
+
+    Returns:
+        The task prompt potentially augmented with schema instructions.
+    """
+    if (task.output_json or task.output_pydantic) and not task.response_model:
+        if task.output_json:
+            schema_dict = generate_model_description(task.output_json)
+            schema = json.dumps(schema_dict["json_schema"]["schema"], indent=2)
+            task_prompt += "\n" + i18n.slice("formatted_task_instructions").format(
+                output_format=schema
+            )
+        elif task.output_pydantic:
+            schema_dict = generate_model_description(task.output_pydantic)
+            schema = json.dumps(schema_dict["json_schema"]["schema"], indent=2)
+            task_prompt += "\n" + i18n.slice("formatted_task_instructions").format(
+                output_format=schema
+            )
+    return task_prompt
+
+
+def format_task_with_context(task_prompt: str, context: str | None, i18n: I18N) -> str:
+    """Format task prompt with context if provided.
+
+    Args:
+        task_prompt: The task prompt.
+        context: Optional context string.
+        i18n: Internationalization instance.
+
+    Returns:
+        The task prompt formatted with context if provided.
+    """
+    if context:
+        return i18n.slice("task_with_context").format(task=task_prompt, context=context)
+    return task_prompt
+
+
+def get_knowledge_config(agent: Agent) -> dict[str, Any]:
+    """Get knowledge configuration from agent.
+
+    Args:
+        agent: The agent instance.
+
+    Returns:
+        Dictionary of knowledge configuration.
+    """
+    return agent.knowledge_config.model_dump() if agent.knowledge_config else {}
+
+
+def handle_knowledge_retrieval(
+    agent: Agent,
+    task: Task,
+    task_prompt: str,
+    knowledge_config: dict[str, Any],
+    query_func: Any,
+    crew_query_func: Any,
+) -> str:
+    """Handle knowledge retrieval for task execution.
+
+    This function handles both agent-specific and crew-specific knowledge queries.
+
+    Args:
+        agent: The agent performing the task.
+        task: The task being executed.
+        task_prompt: The current task prompt.
+        knowledge_config: Knowledge configuration dictionary.
+        query_func: Function to query agent knowledge (sync or async).
+        crew_query_func: Function to query crew knowledge (sync or async).
+
+    Returns:
+        The task prompt potentially augmented with knowledge context.
+    """
+    if not (agent.knowledge or (agent.crew and agent.crew.knowledge)):
+        return task_prompt
+
+    crewai_event_bus.emit(
+        agent,
+        event=KnowledgeRetrievalStartedEvent(
+            from_task=task,
+            from_agent=agent,
+        ),
+    )
+    try:
+        agent.knowledge_search_query = agent._get_knowledge_search_query(
+            task_prompt, task
+        )
+        if agent.knowledge_search_query:
+            if agent.knowledge:
+                agent_knowledge_snippets = query_func(
+                    [agent.knowledge_search_query], **knowledge_config
+                )
+                if agent_knowledge_snippets:
+                    agent.agent_knowledge_context = extract_knowledge_context(
+                        agent_knowledge_snippets
+                    )
+                    if agent.agent_knowledge_context:
+                        task_prompt += agent.agent_knowledge_context
+
+            knowledge_snippets = crew_query_func(
+                [agent.knowledge_search_query], **knowledge_config
+            )
+            if knowledge_snippets:
+                agent.crew_knowledge_context = extract_knowledge_context(
+                    knowledge_snippets
+                )
+                if agent.crew_knowledge_context:
+                    task_prompt += agent.crew_knowledge_context
+
+            crewai_event_bus.emit(
+                agent,
+                event=KnowledgeRetrievalCompletedEvent(
+                    query=agent.knowledge_search_query,
+                    from_task=task,
+                    from_agent=agent,
+                    retrieved_knowledge=_combine_knowledge_context(agent),
+                ),
+            )
+    except Exception as e:
+        crewai_event_bus.emit(
+            agent,
+            event=KnowledgeSearchQueryFailedEvent(
+                query=agent.knowledge_search_query or "",
+                error=str(e),
+                from_task=task,
+                from_agent=agent,
+            ),
+        )
+    return task_prompt
+
+
+def _combine_knowledge_context(agent: Agent) -> str:
+    """Combine agent and crew knowledge contexts into a single string.
+
+    Args:
+        agent: The agent with knowledge contexts.
+
+    Returns:
+        Combined knowledge context string.
+    """
+    agent_ctx = agent.agent_knowledge_context or ""
+    crew_ctx = agent.crew_knowledge_context or ""
+    separator = "\n" if agent_ctx and crew_ctx else ""
+    return agent_ctx + separator + crew_ctx
+
+
+def apply_training_data(agent: Agent, task_prompt: str) -> str:
+    """Apply training data to the task prompt.
+
+    Args:
+        agent: The agent performing the task.
+        task_prompt: The task prompt.
+
+    Returns:
+        The task prompt with training data applied.
+    """
+    if agent.crew and agent.crew._train:
+        return agent._training_handler(task_prompt=task_prompt)
+    return agent._use_trained_data(task_prompt=task_prompt)
+
+
+def process_tool_results(agent: Agent, result: Any) -> Any:
+    """Process tool results, returning result_as_answer if applicable.
+
+    Args:
+        agent: The agent with tool results.
+        result: The current result.
+
+    Returns:
+        The final result, potentially overridden by tool result_as_answer.
+    """
+    for tool_result in agent.tools_results:
+        if tool_result.get("result_as_answer", False):
+            result = tool_result["result"]
+    return result
+
+
+def save_last_messages(agent: Agent) -> None:
+    """Save the last messages from agent executor.
+
+    Args:
+        agent: The agent instance.
+    """
+    agent._last_messages = (
+        agent.agent_executor.messages.copy()
+        if agent.agent_executor and hasattr(agent.agent_executor, "messages")
+        else []
+    )
+
+
+def prepare_tools(
+    agent: Agent, tools: list[BaseTool] | None, task: Task
+) -> list[BaseTool]:
+    """Prepare tools for task execution and create agent executor.
+
+    Args:
+        agent: The agent instance.
+        tools: Optional list of tools.
+        task: The task being executed.
+
+    Returns:
+        The list of tools to use.
+    """
+    final_tools = tools or agent.tools or []
+    agent.create_agent_executor(tools=final_tools, task=task)
+    return final_tools
+
+
+def validate_max_execution_time(max_execution_time: int | None) -> None:
+    """Validate max_execution_time parameter.
+
+    Args:
+        max_execution_time: The maximum execution time to validate.
+
+    Raises:
+        ValueError: If max_execution_time is not a positive integer.
+    """
+    if max_execution_time is not None:
+        if not isinstance(max_execution_time, int) or max_execution_time <= 0:
+            raise ValueError(
+                "Max Execution time must be a positive integer greater than zero"
+            )
+
+
+async def ahandle_knowledge_retrieval(
+    agent: Agent,
+    task: Task,
+    task_prompt: str,
+    knowledge_config: dict[str, Any],
+) -> str:
+    """Handle async knowledge retrieval for task execution.
+
+    Args:
+        agent: The agent performing the task.
+        task: The task being executed.
+        task_prompt: The current task prompt.
+        knowledge_config: Knowledge configuration dictionary.
+
+    Returns:
+        The task prompt potentially augmented with knowledge context.
+    """
+    if not (agent.knowledge or (agent.crew and agent.crew.knowledge)):
+        return task_prompt
+
+    crewai_event_bus.emit(
+        agent,
+        event=KnowledgeRetrievalStartedEvent(
+            from_task=task,
+            from_agent=agent,
+        ),
+    )
+    try:
+        agent.knowledge_search_query = agent._get_knowledge_search_query(
+            task_prompt, task
+        )
+        if agent.knowledge_search_query:
+            if agent.knowledge:
+                agent_knowledge_snippets = await agent.knowledge.aquery(
+                    [agent.knowledge_search_query], **knowledge_config
+                )
+                if agent_knowledge_snippets:
+                    agent.agent_knowledge_context = extract_knowledge_context(
+                        agent_knowledge_snippets
+                    )
+                    if agent.agent_knowledge_context:
+                        task_prompt += agent.agent_knowledge_context
+
+            knowledge_snippets = await agent.crew.aquery_knowledge(
+                [agent.knowledge_search_query], **knowledge_config
+            )
+            if knowledge_snippets:
+                agent.crew_knowledge_context = extract_knowledge_context(
+                    knowledge_snippets
+                )
+                if agent.crew_knowledge_context:
+                    task_prompt += agent.crew_knowledge_context
+
+            crewai_event_bus.emit(
+                agent,
+                event=KnowledgeRetrievalCompletedEvent(
+                    query=agent.knowledge_search_query,
+                    from_task=task,
+                    from_agent=agent,
+                    retrieved_knowledge=_combine_knowledge_context(agent),
+                ),
+            )
+    except Exception as e:
+        crewai_event_bus.emit(
+            agent,
+            event=KnowledgeSearchQueryFailedEvent(
+                query=agent.knowledge_search_query or "",
+                error=str(e),
+                from_task=task,
+                from_agent=agent,
+            ),
+        )
+    return task_prompt

lib/crewai/src/crewai/agents/agent_adapters/base_converter_adapter.py

@@ -5,10 +5,9 @@ from __future__ import annotations
 from abc import ABC, abstractmethod
 import json
 import re
-from typing import TYPE_CHECKING, Final, Literal
-
-from crewai.utilities.converter import generate_model_description
+from typing import TYPE_CHECKING, Any, Final, Literal
 
+from crewai.utilities.pydantic_schema_utils import generate_model_description
 
 
 if TYPE_CHECKING:
@@ -42,7 +41,7 @@ class BaseConverterAdapter(ABC):
         """
         self.agent_adapter = agent_adapter
         self._output_format: Literal["json", "pydantic"] | None = None
-        self._schema: str | None = None
+        self._schema: dict[str, Any] | None = None
 
     @abstractmethod
     def configure_structured_output(self, task: Task) -> None:
@@ -129,7 +128,7 @@ class BaseConverterAdapter(ABC):
     @staticmethod
     def _configure_format_from_task(
         task: Task,
-    ) -> tuple[Literal["json", "pydantic"] | None, str | None]:
+    ) -> tuple[Literal["json", "pydantic"] | None, dict[str, Any] | None]:
         """Determine output format and schema from task requirements.
 
         This is a helper method that examines the task's output requirements

lib/crewai/src/crewai/agents/agent_adapters/openai_agents/structured_output_converter.py

@@ -4,6 +4,7 @@ This module contains the OpenAIConverterAdapter class that handles structured
 output conversion for OpenAI agents, supporting JSON and Pydantic model formats.
 """
 
+import json
 from typing import Any
 
 from crewai.agents.agent_adapters.base_converter_adapter import BaseConverterAdapter
@@ -61,7 +62,7 @@ class OpenAIConverterAdapter(BaseConverterAdapter):
         output_schema: str = (
             get_i18n()
             .slice("formatted_task_instructions")
-            .format(output_format=self._schema)
+            .format(output_format=json.dumps(self._schema, indent=2))
         )
 
         return f"{base_prompt}\n\n{output_schema}"

lib/crewai/src/crewai/agents/agent_builder/base_agent.py

@@ -265,7 +265,7 @@ class BaseAgent(BaseModel, ABC, metaclass=AgentMeta):
         if not mcps:
             return mcps
 
-        validated_mcps = []
+        validated_mcps: list[str | MCPServerConfig] = []
         for mcp in mcps:
             if isinstance(mcp, str):
                 if mcp.startswith(("https://", "crewai-amp:")):
@@ -347,6 +347,15 @@ class BaseAgent(BaseModel, ABC, metaclass=AgentMeta):
     ) -> str:
         pass
 
+    @abstractmethod
+    async def aexecute_task(
+        self,
+        task: Any,
+        context: str | None = None,
+        tools: list[BaseTool] | None = None,
+    ) -> str:
+        """Execute a task asynchronously."""
+
     @abstractmethod
     def create_agent_executor(self, tools: list[BaseTool] | None = None) -> None:
         pass

lib/crewai/src/crewai/agents/crew_agent_executor.py

@@ -28,6 +28,7 @@ from crewai.hooks.llm_hooks import (
     get_before_llm_call_hooks,
 )
 from crewai.utilities.agent_utils import (
+    aget_llm_response,
     enforce_rpm_limit,
     format_message_for_llm,
     get_llm_response,
@@ -43,7 +44,10 @@ from crewai.utilities.agent_utils import (
 from crewai.utilities.constants import TRAINING_DATA_FILE
 from crewai.utilities.i18n import I18N, get_i18n
 from crewai.utilities.printer import Printer
-from crewai.utilities.tool_utils import execute_tool_and_check_finality
+from crewai.utilities.tool_utils import (
+    aexecute_tool_and_check_finality,
+    execute_tool_and_check_finality,
+)
 from crewai.utilities.training_handler import CrewTrainingHandler
 
 
@@ -134,8 +138,8 @@ class CrewAgentExecutor(CrewAgentExecutorMixin):
         self.messages: list[LLMMessage] = []
         self.iterations = 0
         self.log_error_after = 3
-        self.before_llm_call_hooks: list[Callable] = []
-        self.after_llm_call_hooks: list[Callable] = []
+        self.before_llm_call_hooks: list[Callable[..., Any]] = []
+        self.after_llm_call_hooks: list[Callable[..., Any]] = []
         self.before_llm_call_hooks.extend(get_before_llm_call_hooks())
         self.after_llm_call_hooks.extend(get_after_llm_call_hooks())
         if self.llm:
@@ -312,6 +316,154 @@ class CrewAgentExecutor(CrewAgentExecutorMixin):
         self._show_logs(formatted_answer)
         return formatted_answer
 
+    async def ainvoke(self, inputs: dict[str, Any]) -> dict[str, Any]:
+        """Execute the agent asynchronously with given inputs.
+
+        Args:
+            inputs: Input dictionary containing prompt variables.
+
+        Returns:
+            Dictionary with agent output.
+        """
+        if "system" in self.prompt:
+            system_prompt = self._format_prompt(
+                cast(str, self.prompt.get("system", "")), inputs
+            )
+            user_prompt = self._format_prompt(
+                cast(str, self.prompt.get("user", "")), inputs
+            )
+            self.messages.append(format_message_for_llm(system_prompt, role="system"))
+            self.messages.append(format_message_for_llm(user_prompt))
+        else:
+            user_prompt = self._format_prompt(self.prompt.get("prompt", ""), inputs)
+            self.messages.append(format_message_for_llm(user_prompt))
+
+        self._show_start_logs()
+
+        self.ask_for_human_input = bool(inputs.get("ask_for_human_input", False))
+
+        try:
+            formatted_answer = await self._ainvoke_loop()
+        except AssertionError:
+            self._printer.print(
+                content="Agent failed to reach a final answer. This is likely a bug - please report it.",
+                color="red",
+            )
+            raise
+        except Exception as e:
+            handle_unknown_error(self._printer, e)
+            raise
+
+        if self.ask_for_human_input:
+            formatted_answer = self._handle_human_feedback(formatted_answer)
+
+        self._create_short_term_memory(formatted_answer)
+        self._create_long_term_memory(formatted_answer)
+        self._create_external_memory(formatted_answer)
+        return {"output": formatted_answer.output}
+
+    async def _ainvoke_loop(self) -> AgentFinish:
+        """Execute agent loop asynchronously until completion.
+
+        Returns:
+            Final answer from the agent.
+        """
+        formatted_answer = None
+        while not isinstance(formatted_answer, AgentFinish):
+            try:
+                if has_reached_max_iterations(self.iterations, self.max_iter):
+                    formatted_answer = handle_max_iterations_exceeded(
+                        formatted_answer,
+                        printer=self._printer,
+                        i18n=self._i18n,
+                        messages=self.messages,
+                        llm=self.llm,
+                        callbacks=self.callbacks,
+                    )
+                    break
+
+                enforce_rpm_limit(self.request_within_rpm_limit)
+
+                answer = await aget_llm_response(
+                    llm=self.llm,
+                    messages=self.messages,
+                    callbacks=self.callbacks,
+                    printer=self._printer,
+                    from_task=self.task,
+                    from_agent=self.agent,
+                    response_model=self.response_model,
+                    executor_context=self,
+                )
+                formatted_answer = process_llm_response(answer, self.use_stop_words)  # type: ignore[assignment]
+
+                if isinstance(formatted_answer, AgentAction):
+                    fingerprint_context = {}
+                    if (
+                        self.agent
+                        and hasattr(self.agent, "security_config")
+                        and hasattr(self.agent.security_config, "fingerprint")
+                    ):
+                        fingerprint_context = {
+                            "agent_fingerprint": str(
+                                self.agent.security_config.fingerprint
+                            )
+                        }
+
+                    tool_result = await aexecute_tool_and_check_finality(
+                        agent_action=formatted_answer,
+                        fingerprint_context=fingerprint_context,
+                        tools=self.tools,
+                        i18n=self._i18n,
+                        agent_key=self.agent.key if self.agent else None,
+                        agent_role=self.agent.role if self.agent else None,
+                        tools_handler=self.tools_handler,
+                        task=self.task,
+                        agent=self.agent,
+                        function_calling_llm=self.function_calling_llm,
+                        crew=self.crew,
+                    )
+                    formatted_answer = self._handle_agent_action(
+                        formatted_answer, tool_result
+                    )
+
+                self._invoke_step_callback(formatted_answer)  # type: ignore[arg-type]
+                self._append_message(formatted_answer.text)  # type: ignore[union-attr,attr-defined]
+
+            except OutputParserError as e:
+                formatted_answer = handle_output_parser_exception(  # type: ignore[assignment]
+                    e=e,
+                    messages=self.messages,
+                    iterations=self.iterations,
+                    log_error_after=self.log_error_after,
+                    printer=self._printer,
+                )
+
+            except Exception as e:
+                if e.__class__.__module__.startswith("litellm"):
+                    raise e
+                if is_context_length_exceeded(e):
+                    handle_context_length(
+                        respect_context_window=self.respect_context_window,
+                        printer=self._printer,
+                        messages=self.messages,
+                        llm=self.llm,
+                        callbacks=self.callbacks,
+                        i18n=self._i18n,
+                    )
+                    continue
+                handle_unknown_error(self._printer, e)
+                raise e
+            finally:
+                self.iterations += 1
+
+        if not isinstance(formatted_answer, AgentFinish):
+            raise RuntimeError(
+                "Agent execution ended without reaching a final answer. "
+                f"Got {type(formatted_answer).__name__} instead of AgentFinish."
+            )
+        self._show_logs(formatted_answer)
+        return formatted_answer
+
     def _handle_agent_action(
         self, formatted_answer: AgentAction, tool_result: ToolResult
     ) -> AgentAction | AgentFinish:

lib/crewai/src/crewai/cli/crew_chat.py

@@ -14,7 +14,8 @@ import tomli
 from crewai.cli.utils import read_toml
 from crewai.cli.version import get_crewai_version
 from crewai.crew import Crew
-from crewai.llm import LLM, BaseLLM
+from crewai.llm import LLM
+from crewai.llms.base_llm import BaseLLM
 from crewai.types.crew_chat import ChatInputField, ChatInputs
 from crewai.utilities.llm_utils import create_llm
 from crewai.utilities.printer import Printer
@@ -27,7 +28,7 @@ MIN_REQUIRED_VERSION: Final[Literal["0.98.0"]] = "0.98.0"
 
 
 def check_conversational_crews_version(
-    crewai_version: str, pyproject_data: dict
+    crewai_version: str, pyproject_data: dict[str, Any]
 ) -> bool:
     """
     Check if the installed crewAI version supports conversational crews.
@@ -53,7 +54,7 @@ def check_conversational_crews_version(
     return True
 
 
-def run_chat():
+def run_chat() -> None:
     """
     Runs an interactive chat loop using the Crew's chat LLM with function calling.
     Incorporates crew_name, crew_description, and input fields to build a tool schema.
@@ -101,7 +102,7 @@ def run_chat():
 
     click.secho(f"Assistant: {introductory_message}\n", fg="green")
 
-    messages = [
+    messages: list[LLMMessage] = [
         {"role": "system", "content": system_message},
         {"role": "assistant", "content": introductory_message},
     ]
@@ -113,7 +114,7 @@ def run_chat():
     chat_loop(chat_llm, messages, crew_tool_schema, available_functions)
 
 
-def show_loading(event: threading.Event):
+def show_loading(event: threading.Event) -> None:
     """Display animated loading dots while processing."""
     while not event.is_set():
         _printer.print(".", end="")
@@ -162,23 +163,23 @@ def build_system_message(crew_chat_inputs: ChatInputs) -> str:
     )
 
 
-def create_tool_function(crew: Crew, messages: list[dict[str, str]]) -> Any:
+def create_tool_function(crew: Crew, messages: list[LLMMessage]) -> Any:
     """Creates a wrapper function for running the crew tool with messages."""
 
-    def run_crew_tool_with_messages(**kwargs):
+    def run_crew_tool_with_messages(**kwargs: Any) -> str:
         return run_crew_tool(crew, messages, **kwargs)
 
     return run_crew_tool_with_messages
 
 
-def flush_input():
+def flush_input() -> None:
     """Flush any pending input from the user."""
     if platform.system() == "Windows":
         # Windows platform
         import msvcrt
 
-        while msvcrt.kbhit():
-            msvcrt.getch()
+        while msvcrt.kbhit():  # type: ignore[attr-defined]
+            msvcrt.getch()  # type: ignore[attr-defined]
     else:
         # Unix-like platforms (Linux, macOS)
         import termios
@@ -186,7 +187,12 @@ def flush_input():
         termios.tcflush(sys.stdin, termios.TCIFLUSH)
 
 
-def chat_loop(chat_llm, messages, crew_tool_schema, available_functions):
+def chat_loop(
+    chat_llm: LLM | BaseLLM,
+    messages: list[LLMMessage],
+    crew_tool_schema: dict[str, Any],
+    available_functions: dict[str, Any],
+) -> None:
     """Main chat loop for interacting with the user."""
     while True:
         try:
@@ -225,7 +231,7 @@ def get_user_input() -> str:
 
 def handle_user_input(
     user_input: str,
-    chat_llm: LLM,
+    chat_llm: LLM | BaseLLM,
     messages: list[LLMMessage],
     crew_tool_schema: dict[str, Any],
     available_functions: dict[str, Any],
@@ -255,7 +261,7 @@ def handle_user_input(
     click.secho(f"\nAssistant: {final_response}\n", fg="green")
 
 
-def generate_crew_tool_schema(crew_inputs: ChatInputs) -> dict:
+def generate_crew_tool_schema(crew_inputs: ChatInputs) -> dict[str, Any]:
     """
     Dynamically build a Littellm 'function' schema for the given crew.
 
@@ -286,7 +292,7 @@ def generate_crew_tool_schema(crew_inputs: ChatInputs) -> dict:
     }
 
 
-def run_crew_tool(crew: Crew, messages: list[dict[str, str]], **kwargs):
+def run_crew_tool(crew: Crew, messages: list[LLMMessage], **kwargs: Any) -> str:
     """
     Runs the crew using crew.kickoff(inputs=kwargs) and returns the output.
 
@@ -372,7 +378,9 @@ def load_crew_and_name() -> tuple[Crew, str]:
     return crew_instance, crew_class_name
 
 
-def generate_crew_chat_inputs(crew: Crew, crew_name: str, chat_llm) -> ChatInputs:
+def generate_crew_chat_inputs(
+    crew: Crew, crew_name: str, chat_llm: LLM | BaseLLM
+) -> ChatInputs:
     """
     Generates the ChatInputs required for the crew by analyzing the tasks and agents.
 
@@ -410,23 +418,12 @@ def fetch_required_inputs(crew: Crew) -> set[str]:
     Returns:
         Set[str]: A set of placeholder names.
     """
-    placeholder_pattern = re.compile(r"\{(.+?)}")
-    required_inputs: set[str] = set()
-
-    # Scan tasks
-    for task in crew.tasks:
-        text = f"{task.description or ''} {task.expected_output or ''}"
-        required_inputs.update(placeholder_pattern.findall(text))
-
-    # Scan agents
-    for agent in crew.agents:
-        text = f"{agent.role or ''} {agent.goal or ''} {agent.backstory or ''}"
-        required_inputs.update(placeholder_pattern.findall(text))
-
-    return required_inputs
+    return crew.fetch_inputs()
 
 
-def generate_input_description_with_ai(input_name: str, crew: Crew, chat_llm) -> str:
+def generate_input_description_with_ai(
+    input_name: str, crew: Crew, chat_llm: LLM | BaseLLM
+) -> str:
     """
     Generates an input description using AI based on the context of the crew.
 
@@ -484,10 +481,10 @@ def generate_input_description_with_ai(input_name: str, crew: Crew, chat_llm) ->
         f"{context}"
     )
     response = chat_llm.call(messages=[{"role": "user", "content": prompt}])
-    return response.strip()
+    return str(response).strip()
 
 
-def generate_crew_description_with_ai(crew: Crew, chat_llm) -> str:
+def generate_crew_description_with_ai(crew: Crew, chat_llm: LLM | BaseLLM) -> str:
     """
     Generates a brief description of the crew using AI.
 
@@ -534,4 +531,4 @@ def generate_crew_description_with_ai(crew: Crew, chat_llm) -> str:
         f"{context}"
     )
     response = chat_llm.call(messages=[{"role": "user", "content": prompt}])
-    return response.strip()
+    return str(response).strip()

lib/crewai/src/crewai/cli/shared/token_manager.py

@@ -3,103 +3,56 @@ import json
 import os
 from pathlib import Path
 import sys
-from typing import BinaryIO, cast
+import tempfile
+from typing import Final, Literal, cast
 
 from cryptography.fernet import Fernet
 
 
-if sys.platform == "win32":
-    import msvcrt
-else:
-    import fcntl
+_FERNET_KEY_LENGTH: Final[Literal[44]] = 44
 
 
 class TokenManager:
-    def __init__(self, file_path: str = "tokens.enc") -> None:
-        """
-        Initialize the TokenManager class.
+    """Manages encrypted token storage."""
 
-        :param file_path: The file path to store the encrypted tokens. Default is "tokens.enc".
+    def __init__(self, file_path: str = "tokens.enc") -> None:
+        """Initialize the TokenManager.
+
+        Args:
+            file_path: The file path to store encrypted tokens.
         """
         self.file_path = file_path
         self.key = self._get_or_create_key()
         self.fernet = Fernet(self.key)
 
-    @staticmethod
-    def _acquire_lock(file_handle: BinaryIO) -> None:
-        """
-        Acquire an exclusive lock on a file handle.
-
-        Args:
-            file_handle: Open file handle to lock.
-        """
-        if sys.platform == "win32":
-            msvcrt.locking(file_handle.fileno(), msvcrt.LK_LOCK, 1)
-        else:
-            fcntl.flock(file_handle.fileno(), fcntl.LOCK_EX)
-
-    @staticmethod
-    def _release_lock(file_handle: BinaryIO) -> None:
-        """
-        Release the lock on a file handle.
-
-        Args:
-            file_handle: Open file handle to unlock.
-        """
-        if sys.platform == "win32":
-            msvcrt.locking(file_handle.fileno(), msvcrt.LK_UNLCK, 1)
-        else:
-            fcntl.flock(file_handle.fileno(), fcntl.LOCK_UN)
-
     def _get_or_create_key(self) -> bytes:
-        """
-        Get or create the encryption key with file locking to prevent race conditions.
+        """Get or create the encryption key.
 
         Returns:
-            The encryption key.
+            The encryption key as bytes.
         """
-        key_filename = "secret.key"
-        storage_path = self.get_secure_storage_path()
+        key_filename: str = "secret.key"
 
-        key = self.read_secure_file(key_filename)
-        if key is not None and len(key) == 44:
+        key = self._read_secure_file(key_filename)
+        if key is not None and len(key) == _FERNET_KEY_LENGTH:
             return key
 
-        lock_file_path = storage_path / f"{key_filename}.lock"
-
-        try:
-            lock_file_path.touch()
-
-            with open(lock_file_path, "r+b") as lock_file:
-                self._acquire_lock(lock_file)
-                try:
-                    key = self.read_secure_file(key_filename)
-                    if key is not None and len(key) == 44:
-                        return key
-
-                    new_key = Fernet.generate_key()
-                    self.save_secure_file(key_filename, new_key)
-                    return new_key
-                finally:
-                    try:
-                        self._release_lock(lock_file)
-                    except OSError:
-                        pass
-        except OSError:
-            key = self.read_secure_file(key_filename)
-            if key is not None and len(key) == 44:
-                return key
-
-            new_key = Fernet.generate_key()
-            self.save_secure_file(key_filename, new_key)
+        new_key = Fernet.generate_key()
+        if self._atomic_create_secure_file(key_filename, new_key):
             return new_key
 
-    def save_tokens(self, access_token: str, expires_at: int) -> None:
-        """
-        Save the access token and its expiration time.
+        key = self._read_secure_file(key_filename)
+        if key is not None and len(key) == _FERNET_KEY_LENGTH:
+            return key
 
-        :param access_token: The access token to save.
-        :param expires_at: The UNIX timestamp of the expiration time.
+        raise RuntimeError("Failed to create or read encryption key")
+
+    def save_tokens(self, access_token: str, expires_at: int) -> None:
+        """Save the access token and its expiration time.
+
+        Args:
+            access_token: The access token to save.
+            expires_at: The UNIX timestamp of the expiration time.
         """
         expiration_time = datetime.fromtimestamp(expires_at)
         data = {
@@ -107,15 +60,15 @@ class TokenManager:
             "expiration": expiration_time.isoformat(),
         }
         encrypted_data = self.fernet.encrypt(json.dumps(data).encode())
-        self.save_secure_file(self.file_path, encrypted_data)
+        self._atomic_write_secure_file(self.file_path, encrypted_data)
 
     def get_token(self) -> str | None:
-        """
-        Get the access token if it is valid and not expired.
+        """Get the access token if it is valid and not expired.
 
-        :return: The access token if valid and not expired, otherwise None.
+        Returns:
+            The access token if valid and not expired, otherwise None.
         """
-        encrypted_data = self.read_secure_file(self.file_path)
+        encrypted_data = self._read_secure_file(self.file_path)
         if encrypted_data is None:
             return None
 
@@ -126,20 +79,18 @@ class TokenManager:
         if expiration <= datetime.now():
             return None
 
-        return cast(str | None, data["access_token"])
+        return cast(str | None, data.get("access_token"))
 
     def clear_tokens(self) -> None:
-        """
-        Clear the tokens.
-        """
-        self.delete_secure_file(self.file_path)
+        """Clear the stored tokens."""
+        self._delete_secure_file(self.file_path)
 
     @staticmethod
-    def get_secure_storage_path() -> Path:
-        """
-        Get the secure storage path based on the operating system.
+    def _get_secure_storage_path() -> Path:
+        """Get the secure storage path based on the operating system.
 
-        :return: The secure storage path.
+        Returns:
+            The secure storage path.
         """
         if sys.platform == "win32":
             base_path = os.environ.get("LOCALAPPDATA")
@@ -155,44 +106,81 @@ class TokenManager:
 
         return storage_path
 
-    def save_secure_file(self, filename: str, content: bytes) -> None:
-        """
-        Save the content to a secure file.
+    def _atomic_create_secure_file(self, filename: str, content: bytes) -> bool:
+        """Create a file only if it doesn't exist.
 
-        :param filename: The name of the file.
-        :param content: The content to save.
+        Args:
+            filename: The name of the file.
+            content: The content to write.
+
+        Returns:
+            True if file was created, False if it already exists.
         """
-        storage_path = self.get_secure_storage_path()
+        storage_path = self._get_secure_storage_path()
         file_path = storage_path / filename
 
-        with open(file_path, "wb") as f:
-            f.write(content)
+        try:
+            fd = os.open(file_path, os.O_CREAT | os.O_EXCL | os.O_WRONLY, 0o600)
+            try:
+                os.write(fd, content)
+            finally:
+                os.close(fd)
+            return True
+        except FileExistsError:
+            return False
 
-        os.chmod(file_path, 0o600)
+    def _atomic_write_secure_file(self, filename: str, content: bytes) -> None:
+        """Write content to a secure file.
 
-    def read_secure_file(self, filename: str) -> bytes | None:
+        Args:
+            filename: The name of the file.
+            content: The content to write.
         """
-        Read the content of a secure file.
-
-        :param filename: The name of the file.
-        :return: The content of the file if it exists, otherwise None.
-        """
-        storage_path = self.get_secure_storage_path()
+        storage_path = self._get_secure_storage_path()
         file_path = storage_path / filename
 
-        if not file_path.exists():
+        fd, temp_path = tempfile.mkstemp(dir=storage_path, prefix=f".{filename}.")
+        fd_closed = False
+        try:
+            os.write(fd, content)
+            os.close(fd)
+            fd_closed = True
+            os.chmod(temp_path, 0o600)
+            os.replace(temp_path, file_path)
+        except Exception:
+            if not fd_closed:
+                os.close(fd)
+            if os.path.exists(temp_path):
+                os.unlink(temp_path)
+            raise
+
+    def _read_secure_file(self, filename: str) -> bytes | None:
+        """Read the content of a secure file.
+
+        Args:
+            filename: The name of the file.
+
+        Returns:
+            The content of the file if it exists, otherwise None.
+        """
+        storage_path = self._get_secure_storage_path()
+        file_path = storage_path / filename
+
+        try:
+            with open(file_path, "rb") as f:
+                return f.read()
+        except FileNotFoundError:
             return None
 
-        with open(file_path, "rb") as f:
-            return f.read()
+    def _delete_secure_file(self, filename: str) -> None:
+        """Delete a secure file.
 
-    def delete_secure_file(self, filename: str) -> None:
+        Args:
+            filename: The name of the file.
         """
-        Delete the secure file.
-
-        :param filename: The name of the file.
-        """
-        storage_path = self.get_secure_storage_path()
+        storage_path = self._get_secure_storage_path()
         file_path = storage_path / filename
-        if file_path.exists():
-            file_path.unlink(missing_ok=True)
+        try:
+            file_path.unlink()
+        except FileNotFoundError:
+            pass

lib/crewai/src/crewai/cli/templates/crew/pyproject.toml

@@ -5,7 +5,7 @@ description = "{{name}} using crewAI"
 authors = [{ name = "Your Name", email = "you@example.com" }]
 requires-python = ">=3.10,<3.14"
 dependencies = [
-    "crewai[tools]==1.6.1"
+    "crewai[tools]==1.7.0"
 ]
 
 [project.scripts]

lib/crewai/src/crewai/cli/templates/flow/pyproject.toml

@@ -5,7 +5,7 @@ description = "{{name}} using crewAI"
 authors = [{ name = "Your Name", email = "you@example.com" }]
 requires-python = ">=3.10,<3.14"
 dependencies = [
-    "crewai[tools]==1.6.1"
+    "crewai[tools]==1.7.0"
 ]
 
 [project.scripts]

lib/crewai/src/crewai/cli/tools/main.py

@@ -1,4 +1,5 @@
 import base64
+from json import JSONDecodeError
 import os
 from pathlib import Path
 import subprocess
@@ -162,9 +163,19 @@ class ToolCommand(BaseCommand, PlusAPIMixin):
 
         if login_response.status_code != 200:
             console.print(
-                "Authentication failed. Verify if the currently active organization access to the tool repository, and run 'crewai login' again. ",
+                "Authentication failed. Verify if the currently active organization can access the tool repository, and run 'crewai login' again.",
                 style="bold red",
             )
+            try:
+                console.print(
+                    f"[{login_response.status_code} error - {login_response.json().get('message', 'Unknown error')}]",
+                    style="bold red italic",
+                )
+            except JSONDecodeError:
+                console.print(
+                    f"[{login_response.status_code} error - Unknown error - Invalid JSON response]",
+                    style="bold red italic",
+                )
             raise SystemExit
 
         login_response_json = login_response.json()

lib/crewai/src/crewai/crew.py

@@ -35,6 +35,14 @@ from crewai.agent import Agent
 from crewai.agents.agent_builder.base_agent import BaseAgent
 from crewai.agents.cache.cache_handler import CacheHandler
 from crewai.crews.crew_output import CrewOutput
+from crewai.crews.utils import (
+    StreamingContext,
+    check_conditional_skip,
+    enable_agent_streaming,
+    prepare_kickoff,
+    prepare_task_execution,
+    run_for_each_async,
+)
 from crewai.events.event_bus import crewai_event_bus
 from crewai.events.event_listener import EventListener
 from crewai.events.listeners.tracing.trace_listener import (
@@ -47,7 +55,6 @@ from crewai.events.listeners.tracing.utils import (
 from crewai.events.types.crew_events import (
     CrewKickoffCompletedEvent,
     CrewKickoffFailedEvent,
-    CrewKickoffStartedEvent,
     CrewTestCompletedEvent,
     CrewTestFailedEvent,
     CrewTestStartedEvent,
@@ -74,7 +81,7 @@ from crewai.tasks.conditional_task import ConditionalTask
 from crewai.tasks.task_output import TaskOutput
 from crewai.tools.agent_tools.agent_tools import AgentTools
 from crewai.tools.base_tool import BaseTool
-from crewai.types.streaming import CrewStreamingOutput, FlowStreamingOutput
+from crewai.types.streaming import CrewStreamingOutput
 from crewai.types.usage_metrics import UsageMetrics
 from crewai.utilities.constants import NOT_SPECIFIED, TRAINING_DATA_FILE
 from crewai.utilities.crew.models import CrewContext
@@ -92,10 +99,8 @@ from crewai.utilities.planning_handler import CrewPlanner
 from crewai.utilities.printer import PrinterColor
 from crewai.utilities.rpm_controller import RPMController
 from crewai.utilities.streaming import (
-    TaskInfo,
     create_async_chunk_generator,
     create_chunk_generator,
-    create_streaming_state,
     signal_end,
     signal_error,
 )
@@ -268,7 +273,7 @@ class Crew(FlowTrackable, BaseModel):
         description="list of file paths for task execution JSON files.",
     )
     execution_logs: list[dict[str, Any]] = Field(
-        default=[],
+        default_factory=list,
         description="list of execution logs for tasks",
     )
     knowledge_sources: list[BaseKnowledgeSource] | None = Field(
@@ -327,7 +332,7 @@ class Crew(FlowTrackable, BaseModel):
     def set_private_attrs(self) -> Crew:
         """set private attributes."""
         self._cache_handler = CacheHandler()
-        event_listener = EventListener()  # type: ignore[no-untyped-call]
+        event_listener = EventListener()
 
         # Determine and set tracing state once for this execution
         tracing_enabled = should_enable_tracing(override=self.tracing)
@@ -348,12 +353,12 @@ class Crew(FlowTrackable, BaseModel):
         return self
 
     def _initialize_default_memories(self) -> None:
-        self._long_term_memory = self._long_term_memory or LongTermMemory()  # type: ignore[no-untyped-call]
-        self._short_term_memory = self._short_term_memory or ShortTermMemory(  # type: ignore[no-untyped-call]
+        self._long_term_memory = self._long_term_memory or LongTermMemory()
+        self._short_term_memory = self._short_term_memory or ShortTermMemory(
             crew=self,
             embedder_config=self.embedder,
         )
-        self._entity_memory = self.entity_memory or EntityMemory(  # type: ignore[no-untyped-call]
+        self._entity_memory = self.entity_memory or EntityMemory(
             crew=self, embedder_config=self.embedder
         )
 
@@ -404,8 +409,7 @@ class Crew(FlowTrackable, BaseModel):
                 raise PydanticCustomError(
                     "missing_manager_llm_or_manager_agent",
                     (
-                        "Attribute `manager_llm` or `manager_agent` is required "
-                        "when using hierarchical process."
+                        "Attribute `manager_llm` or `manager_agent` is required when using hierarchical process."
                     ),
                     {},
                 )
@@ -511,10 +515,9 @@ class Crew(FlowTrackable, BaseModel):
                 raise PydanticCustomError(
                     "invalid_async_conditional_task",
                     (
-                        f"Conditional Task: {task.description}, "
-                        f"cannot be executed asynchronously."
+                        "Conditional Task: {description}, cannot be executed asynchronously."
                     ),
-                    {},
+                    {"description": task.description},
                 )
         return self
 
@@ -675,21 +678,8 @@ class Crew(FlowTrackable, BaseModel):
         inputs: dict[str, Any] | None = None,
     ) -> CrewOutput | CrewStreamingOutput:
         if self.stream:
-            for agent in self.agents:
-                if agent.llm is not None:
-                    agent.llm.stream = True
-
-            result_holder: list[CrewOutput] = []
-            current_task_info: TaskInfo = {
-                "index": 0,
-                "name": "",
-                "id": "",
-                "agent_role": "",
-                "agent_id": "",
-            }
-
-            state = create_streaming_state(current_task_info, result_holder)
-            output_holder: list[CrewStreamingOutput | FlowStreamingOutput] = []
+            enable_agent_streaming(self.agents)
+            ctx = StreamingContext()
 
             def run_crew() -> None:
                 """Execute the crew and capture the result."""
@@ -697,59 +687,28 @@ class Crew(FlowTrackable, BaseModel):
                     self.stream = False
                     crew_result = self.kickoff(inputs=inputs)
                     if isinstance(crew_result, CrewOutput):
-                        result_holder.append(crew_result)
+                        ctx.result_holder.append(crew_result)
                 except Exception as exc:
-                    signal_error(state, exc)
+                    signal_error(ctx.state, exc)
                 finally:
                     self.stream = True
-                    signal_end(state)
+                    signal_end(ctx.state)
 
             streaming_output = CrewStreamingOutput(
-                sync_iterator=create_chunk_generator(state, run_crew, output_holder)
+                sync_iterator=create_chunk_generator(
+                    ctx.state, run_crew, ctx.output_holder
+                )
             )
-            output_holder.append(streaming_output)
+            ctx.output_holder.append(streaming_output)
             return streaming_output
 
-        ctx = baggage.set_baggage(
+        baggage_ctx = baggage.set_baggage(
             "crew_context", CrewContext(id=str(self.id), key=self.key)
         )
-        token = attach(ctx)
+        token = attach(baggage_ctx)
 
         try:
-            for before_callback in self.before_kickoff_callbacks:
-                if inputs is None:
-                    inputs = {}
-                inputs = before_callback(inputs)
-
-            crewai_event_bus.emit(
-                self,
-                CrewKickoffStartedEvent(crew_name=self.name, inputs=inputs),
-            )
-
-            # Starts the crew to work on its assigned tasks.
-            self._task_output_handler.reset()
-            self._logging_color = "bold_purple"
-
-            if inputs is not None:
-                self._inputs = inputs
-                self._interpolate_inputs(inputs)
-            self._set_tasks_callbacks()
-            self._set_allow_crewai_trigger_context_for_first_task()
-
-            for agent in self.agents:
-                agent.crew = self
-                agent.set_knowledge(crew_embedder=self.embedder)
-                # TODO: Create an AgentFunctionCalling protocol for future refactoring
-                if not agent.function_calling_llm:  # type: ignore # "BaseAgent" has no attribute "function_calling_llm"
-                    agent.function_calling_llm = self.function_calling_llm  # type: ignore # "BaseAgent" has no attribute "function_calling_llm"
-
-                if not agent.step_callback:  # type: ignore # "BaseAgent" has no attribute "step_callback"
-                    agent.step_callback = self.step_callback  # type: ignore # "BaseAgent" has no attribute "step_callback"
-
-                agent.create_agent_executor()
-
-            if self.planning:
-                self._handle_crew_planning()
+            inputs = prepare_kickoff(self, inputs)
 
             if self.process == Process.sequential:
                 result = self._run_sequential_process()
@@ -814,42 +773,27 @@ class Crew(FlowTrackable, BaseModel):
         inputs = inputs or {}
 
         if self.stream:
-            for agent in self.agents:
-                if agent.llm is not None:
-                    agent.llm.stream = True
-
-            result_holder: list[CrewOutput] = []
-            current_task_info: TaskInfo = {
-                "index": 0,
-                "name": "",
-                "id": "",
-                "agent_role": "",
-                "agent_id": "",
-            }
-
-            state = create_streaming_state(
-                current_task_info, result_holder, use_async=True
-            )
-            output_holder: list[CrewStreamingOutput | FlowStreamingOutput] = []
+            enable_agent_streaming(self.agents)
+            ctx = StreamingContext(use_async=True)
 
             async def run_crew() -> None:
                 try:
                     self.stream = False
                     result = await asyncio.to_thread(self.kickoff, inputs)
                     if isinstance(result, CrewOutput):
-                        result_holder.append(result)
+                        ctx.result_holder.append(result)
                 except Exception as e:
-                    signal_error(state, e, is_async=True)
+                    signal_error(ctx.state, e, is_async=True)
                 finally:
                     self.stream = True
-                    signal_end(state, is_async=True)
+                    signal_end(ctx.state, is_async=True)
 
             streaming_output = CrewStreamingOutput(
                 async_iterator=create_async_chunk_generator(
-                    state, run_crew, output_holder
+                    ctx.state, run_crew, ctx.output_holder
                 )
             )
-            output_holder.append(streaming_output)
+            ctx.output_holder.append(streaming_output)
 
             return streaming_output
 
@@ -864,89 +808,207 @@ class Crew(FlowTrackable, BaseModel):
         from all crews as they arrive. After iteration, access results via .results
         (list of CrewOutput).
         """
-        crew_copies = [self.copy() for _ in inputs]
 
+        async def kickoff_fn(
+            crew: Crew, input_data: dict[str, Any]
+        ) -> CrewOutput | CrewStreamingOutput:
+            return await crew.kickoff_async(inputs=input_data)
+
+        return await run_for_each_async(self, inputs, kickoff_fn)
+
+    async def akickoff(
+        self, inputs: dict[str, Any] | None = None
+    ) -> CrewOutput | CrewStreamingOutput:
+        """Native async kickoff method using async task execution throughout.
+
+        Unlike kickoff_async which wraps sync kickoff in a thread, this method
+        uses native async/await for all operations including task execution,
+        memory operations, and knowledge queries.
+        """
         if self.stream:
-            result_holder: list[list[CrewOutput]] = [[]]
-            current_task_info: TaskInfo = {
-                "index": 0,
-                "name": "",
-                "id": "",
-                "agent_role": "",
-                "agent_id": "",
-            }
+            enable_agent_streaming(self.agents)
+            ctx = StreamingContext(use_async=True)
 
-            state = create_streaming_state(
-                current_task_info, result_holder, use_async=True
-            )
-            output_holder: list[CrewStreamingOutput | FlowStreamingOutput] = []
-
-            async def run_all_crews() -> None:
-                """Run all crew copies and aggregate their streaming outputs."""
+            async def run_crew() -> None:
                 try:
-                    streaming_outputs: list[CrewStreamingOutput] = []
-                    for i, crew in enumerate(crew_copies):
-                        streaming = await crew.kickoff_async(inputs=inputs[i])
-                        if isinstance(streaming, CrewStreamingOutput):
-                            streaming_outputs.append(streaming)
-
-                    async def consume_stream(
-                        stream_output: CrewStreamingOutput,
-                    ) -> CrewOutput:
-                        """Consume stream chunks and forward to parent queue.
-
-                        Args:
-                            stream_output: The streaming output to consume.
-
-                        Returns:
-                            The final CrewOutput result.
-                        """
-                        async for chunk in stream_output:
-                            if state.async_queue is not None and state.loop is not None:
-                                state.loop.call_soon_threadsafe(
-                                    state.async_queue.put_nowait, chunk
-                                )
-                        return stream_output.result
-
-                    crew_results = await asyncio.gather(
-                        *[consume_stream(s) for s in streaming_outputs]
-                    )
-                    result_holder[0] = list(crew_results)
-                except Exception as e:
-                    signal_error(state, e, is_async=True)
+                    self.stream = False
+                    inner_result = await self.akickoff(inputs)
+                    if isinstance(inner_result, CrewOutput):
+                        ctx.result_holder.append(inner_result)
+                except Exception as exc:
+                    signal_error(ctx.state, exc, is_async=True)
                 finally:
-                    signal_end(state, is_async=True)
+                    self.stream = True
+                    signal_end(ctx.state, is_async=True)
 
             streaming_output = CrewStreamingOutput(
                 async_iterator=create_async_chunk_generator(
-                    state, run_all_crews, output_holder
+                    ctx.state, run_crew, ctx.output_holder
                 )
             )
-
-            def set_results_wrapper(result: Any) -> None:
-                """Wrap _set_results to match _set_result signature."""
-                streaming_output._set_results(result)
-
-            streaming_output._set_result = set_results_wrapper  # type: ignore[method-assign]
-            output_holder.append(streaming_output)
+            ctx.output_holder.append(streaming_output)
 
             return streaming_output
 
-        tasks = [
-            asyncio.create_task(crew_copy.kickoff_async(inputs=input_data))
-            for crew_copy, input_data in zip(crew_copies, inputs, strict=True)
-        ]
+        baggage_ctx = baggage.set_baggage(
+            "crew_context", CrewContext(id=str(self.id), key=self.key)
+        )
+        token = attach(baggage_ctx)
 
-        results = await asyncio.gather(*tasks)
+        try:
+            inputs = prepare_kickoff(self, inputs)
 
-        total_usage_metrics = UsageMetrics()
-        for crew_copy in crew_copies:
-            if crew_copy.usage_metrics:
-                total_usage_metrics.add_usage_metrics(crew_copy.usage_metrics)
-        self.usage_metrics = total_usage_metrics
+            if self.process == Process.sequential:
+                result = await self._arun_sequential_process()
+            elif self.process == Process.hierarchical:
+                result = await self._arun_hierarchical_process()
+            else:
+                raise NotImplementedError(
+                    f"The process '{self.process}' is not implemented yet."
+                )
 
-        self._task_output_handler.reset()
-        return list(results)
+            for after_callback in self.after_kickoff_callbacks:
+                result = after_callback(result)
+
+            self.usage_metrics = self.calculate_usage_metrics()
+
+            return result
+        except Exception as e:
+            crewai_event_bus.emit(
+                self,
+                CrewKickoffFailedEvent(error=str(e), crew_name=self.name),
+            )
+            raise
+        finally:
+            detach(token)
+
+    async def akickoff_for_each(
+        self, inputs: list[dict[str, Any]]
+    ) -> list[CrewOutput | CrewStreamingOutput] | CrewStreamingOutput:
+        """Native async execution of the Crew's workflow for each input.
+
+        Uses native async throughout rather than thread-based async.
+        If stream=True, returns a single CrewStreamingOutput that yields chunks
+        from all crews as they arrive.
+        """
+
+        async def kickoff_fn(
+            crew: Crew, input_data: dict[str, Any]
+        ) -> CrewOutput | CrewStreamingOutput:
+            return await crew.akickoff(inputs=input_data)
+
+        return await run_for_each_async(self, inputs, kickoff_fn)
+
+    async def _arun_sequential_process(self) -> CrewOutput:
+        """Executes tasks sequentially using native async and returns the final output."""
+        return await self._aexecute_tasks(self.tasks)
+
+    async def _arun_hierarchical_process(self) -> CrewOutput:
+        """Creates and assigns a manager agent to complete the tasks using native async."""
+        self._create_manager_agent()
+        return await self._aexecute_tasks(self.tasks)
+
+    async def _aexecute_tasks(
+        self,
+        tasks: list[Task],
+        start_index: int | None = 0,
+        was_replayed: bool = False,
+    ) -> CrewOutput:
+        """Executes tasks using native async and returns the final output.
+
+        Args:
+            tasks: List of tasks to execute
+            start_index: Index to start execution from (for replay)
+            was_replayed: Whether this is a replayed execution
+
+        Returns:
+            CrewOutput: Final output of the crew
+        """
+        task_outputs: list[TaskOutput] = []
+        pending_tasks: list[tuple[Task, asyncio.Task[TaskOutput], int]] = []
+        last_sync_output: TaskOutput | None = None
+
+        for task_index, task in enumerate(tasks):
+            exec_data, task_outputs, last_sync_output = prepare_task_execution(
+                self, task, task_index, start_index, task_outputs, last_sync_output
+            )
+            if exec_data.should_skip:
+                continue
+
+            if isinstance(task, ConditionalTask):
+                skipped_task_output = await self._ahandle_conditional_task(
+                    task, task_outputs, pending_tasks, task_index, was_replayed
+                )
+                if skipped_task_output:
+                    task_outputs.append(skipped_task_output)
+                    continue
+
+            if task.async_execution:
+                context = self._get_context(
+                    task, [last_sync_output] if last_sync_output else []
+                )
+                async_task = asyncio.create_task(
+                    task.aexecute_sync(
+                        agent=exec_data.agent,
+                        context=context,
+                        tools=exec_data.tools,
+                    )
+                )
+                pending_tasks.append((task, async_task, task_index))
+            else:
+                if pending_tasks:
+                    task_outputs = await self._aprocess_async_tasks(
+                        pending_tasks, was_replayed
+                    )
+                    pending_tasks.clear()
+
+                context = self._get_context(task, task_outputs)
+                task_output = await task.aexecute_sync(
+                    agent=exec_data.agent,
+                    context=context,
+                    tools=exec_data.tools,
+                )
+                task_outputs.append(task_output)
+                self._process_task_result(task, task_output)
+                self._store_execution_log(task, task_output, task_index, was_replayed)
+
+        if pending_tasks:
+            task_outputs = await self._aprocess_async_tasks(pending_tasks, was_replayed)
+
+        return self._create_crew_output(task_outputs)
+
+    async def _ahandle_conditional_task(
+        self,
+        task: ConditionalTask,
+        task_outputs: list[TaskOutput],
+        pending_tasks: list[tuple[Task, asyncio.Task[TaskOutput], int]],
+        task_index: int,
+        was_replayed: bool,
+    ) -> TaskOutput | None:
+        """Handle conditional task evaluation using native async."""
+        if pending_tasks:
+            task_outputs = await self._aprocess_async_tasks(pending_tasks, was_replayed)
+            pending_tasks.clear()
+
+        return check_conditional_skip(
+            self, task, task_outputs, task_index, was_replayed
+        )
+
+    async def _aprocess_async_tasks(
+        self,
+        pending_tasks: list[tuple[Task, asyncio.Task[TaskOutput], int]],
+        was_replayed: bool = False,
+    ) -> list[TaskOutput]:
+        """Process pending async tasks and return their outputs."""
+        task_outputs: list[TaskOutput] = []
+        for future_task, async_task, task_index in pending_tasks:
+            task_output = await async_task
+            task_outputs.append(task_output)
+            self._process_task_result(future_task, task_output)
+            self._store_execution_log(
+                future_task, task_output, task_index, was_replayed
+            )
+        return task_outputs
 
     def _handle_crew_planning(self) -> None:
         """Handles the Crew planning."""
@@ -955,10 +1017,26 @@ class Crew(FlowTrackable, BaseModel):
             tasks=self.tasks, planning_agent_llm=self.planning_llm
         )._handle_crew_planning()
 
-        for task, step_plan in zip(
-            self.tasks, result.list_of_plans_per_task, strict=False
-        ):
-            task.description += step_plan.plan
+        plan_map: dict[int, str] = {}
+        for step_plan in result.list_of_plans_per_task:
+            if step_plan.task_number in plan_map:
+                self._logger.log(
+                    "warning",
+                    f"Duplicate plan for Task Number {step_plan.task_number}, "
+                    "using the first plan",
+                )
+            else:
+                plan_map[step_plan.task_number] = step_plan.plan
+
+        for idx, task in enumerate(self.tasks):
+            task_number = idx + 1
+            if task_number in plan_map:
+                task.description += plan_map[task_number]
+            else:
+                self._logger.log(
+                    "warning",
+                    f"No plan found for Task Number {task_number}",
+                )
 
     def _store_execution_log(
         self,
@@ -1048,33 +1126,11 @@ class Crew(FlowTrackable, BaseModel):
         last_sync_output: TaskOutput | None = None
 
         for task_index, task in enumerate(tasks):
-            if start_index is not None and task_index < start_index:
-                if task.output:
-                    if task.async_execution:
-                        task_outputs.append(task.output)
-                    else:
-                        task_outputs = [task.output]
-                        last_sync_output = task.output
-                continue
-
-            agent_to_use = self._get_agent_to_use(task)
-            if agent_to_use is None:
-                raise ValueError(
-                    f"No agent available for task: {task.description}. "
-                    f"Ensure that either the task has an assigned agent "
-                    f"or a manager agent is provided."
-                )
-
-            # Determine which tools to use - task tools take precedence over agent tools
-            tools_for_task = task.tools or agent_to_use.tools or []
-            # Prepare tools and ensure they're compatible with task execution
-            tools_for_task = self._prepare_tools(
-                agent_to_use,
-                task,
-                tools_for_task,
+            exec_data, task_outputs, last_sync_output = prepare_task_execution(
+                self, task, task_index, start_index, task_outputs, last_sync_output
             )
-
-            self._log_task_start(task, agent_to_use.role)
+            if exec_data.should_skip:
+                continue
 
             if isinstance(task, ConditionalTask):
                 skipped_task_output = self._handle_conditional_task(
@@ -1089,9 +1145,9 @@ class Crew(FlowTrackable, BaseModel):
                     task, [last_sync_output] if last_sync_output else []
                 )
                 future = task.execute_async(
-                    agent=agent_to_use,
+                    agent=exec_data.agent,
                     context=context,
-                    tools=tools_for_task,
+                    tools=exec_data.tools,
                 )
                 futures.append((task, future, task_index))
             else:
@@ -1101,9 +1157,9 @@ class Crew(FlowTrackable, BaseModel):
 
                 context = self._get_context(task, task_outputs)
                 task_output = task.execute_sync(
-                    agent=agent_to_use,
+                    agent=exec_data.agent,
                     context=context,
-                    tools=tools_for_task,
+                    tools=exec_data.tools,
                 )
                 task_outputs.append(task_output)
                 self._process_task_result(task, task_output)
@@ -1126,19 +1182,9 @@ class Crew(FlowTrackable, BaseModel):
             task_outputs = self._process_async_tasks(futures, was_replayed)
             futures.clear()
 
-        previous_output = task_outputs[-1] if task_outputs else None
-        if previous_output is not None and not task.should_execute(previous_output):
-            self._logger.log(
-                "debug",
-                f"Skipping conditional task: {task.description}",
-                color="yellow",
-            )
-            skipped_task_output = task.get_skipped_task_output()
-
-            if not was_replayed:
-                self._store_execution_log(task, skipped_task_output, task_index)
-            return skipped_task_output
-        return None
+        return check_conditional_skip(
+            self, task, task_outputs, task_index, was_replayed
+        )
 
     def _prepare_tools(
         self, agent: BaseAgent, task: Task, tools: list[BaseTool]
@@ -1302,7 +1348,8 @@ class Crew(FlowTrackable, BaseModel):
                 )
         return tools
 
-    def _get_context(self, task: Task, task_outputs: list[TaskOutput]) -> str:
+    @staticmethod
+    def _get_context(task: Task, task_outputs: list[TaskOutput]) -> str:
         if not task.context:
             return ""
 
@@ -1371,7 +1418,8 @@ class Crew(FlowTrackable, BaseModel):
             )
         return task_outputs
 
-    def _find_task_index(self, task_id: str, stored_outputs: list[Any]) -> int | None:
+    @staticmethod
+    def _find_task_index(task_id: str, stored_outputs: list[Any]) -> int | None:
         return next(
             (
                 index
@@ -1431,6 +1479,16 @@ class Crew(FlowTrackable, BaseModel):
             )
         return None
 
+    async def aquery_knowledge(
+        self, query: list[str], results_limit: int = 3, score_threshold: float = 0.35
+    ) -> list[SearchResult] | None:
+        """Query the crew's knowledge base for relevant information asynchronously."""
+        if self.knowledge:
+            return await self.knowledge.aquery(
+                query, results_limit=results_limit, score_threshold=score_threshold
+            )
+        return None
+
     def fetch_inputs(self) -> set[str]:
         """
         Gathers placeholders (e.g., {something}) referenced in tasks or agents.
@@ -1439,7 +1497,7 @@ class Crew(FlowTrackable, BaseModel):
 
         Returns a set of all discovered placeholder names.
         """
-        placeholder_pattern = re.compile(r"\{(.+?)\}")
+        placeholder_pattern = re.compile(r"\{(.+?)}")
         required_inputs: set[str] = set()
 
         # Scan tasks for inputs
@@ -1687,6 +1745,32 @@ class Crew(FlowTrackable, BaseModel):
             self._logger.log("error", error_msg)
             raise RuntimeError(error_msg) from e
 
+    def _reset_memory_system(
+        self, system: Any, name: str, reset_fn: Callable[[Any], Any]
+    ) -> None:
+        """Reset a single memory system.
+
+        Args:
+            system: The memory system instance to reset.
+            name: Display name of the memory system for logging.
+            reset_fn: Function to call to reset the system.
+
+        Raises:
+            RuntimeError: If the reset operation fails.
+        """
+        try:
+            reset_fn(system)
+            self._logger.log(
+                "info",
+                f"[Crew ({self.name if self.name else self.id})] "
+                f"{name} memory has been reset",
+            )
+        except Exception as e:
+            raise RuntimeError(
+                f"[Crew ({self.name if self.name else self.id})] "
+                f"Failed to reset {name} memory: {e!s}"
+            ) from e
+
     def _reset_all_memories(self) -> None:
         """Reset all available memory systems."""
         memory_systems = self._get_memory_systems()
@@ -1694,21 +1778,10 @@ class Crew(FlowTrackable, BaseModel):
         for config in memory_systems.values():
             if (system := config.get("system")) is not None:
                 name = config.get("name")
-                try:
-                    reset_fn: Callable[[Any], Any] = cast(
-                        Callable[[Any], Any], config.get("reset")
-                    )
-                    reset_fn(system)
-                    self._logger.log(
-                        "info",
-                        f"[Crew ({self.name if self.name else self.id})] "
-                        f"{name} memory has been reset",
-                    )
-                except Exception as e:
-                    raise RuntimeError(
-                        f"[Crew ({self.name if self.name else self.id})] "
-                        f"Failed to reset {name} memory: {e!s}"
-                    ) from e
+                reset_fn: Callable[[Any], Any] = cast(
+                    Callable[[Any], Any], config.get("reset")
+                )
+                self._reset_memory_system(system, name, reset_fn)
 
     def _reset_specific_memory(self, memory_type: str) -> None:
         """Reset a specific memory system.
@@ -1727,21 +1800,8 @@ class Crew(FlowTrackable, BaseModel):
         if system is None:
             raise RuntimeError(f"{name} memory system is not initialized")
 
-        try:
-            reset_fn: Callable[[Any], Any] = cast(
-                Callable[[Any], Any], config.get("reset")
-            )
-            reset_fn(system)
-            self._logger.log(
-                "info",
-                f"[Crew ({self.name if self.name else self.id})] "
-                f"{name} memory has been reset",
-            )
-        except Exception as e:
-            raise RuntimeError(
-                f"[Crew ({self.name if self.name else self.id})] "
-                f"Failed to reset {name} memory: {e!s}"
-            ) from e
+        reset_fn: Callable[[Any], Any] = cast(Callable[[Any], Any], config.get("reset"))
+        self._reset_memory_system(system, name, reset_fn)
 
     def _get_memory_systems(self) -> dict[str, Any]:
         """Get all available memory systems with their configuration.
@@ -1829,7 +1889,8 @@ class Crew(FlowTrackable, BaseModel):
         ):
             self.tasks[0].allow_crewai_trigger_context = True
 
-    def _show_tracing_disabled_message(self) -> None:
+    @staticmethod
+    def _show_tracing_disabled_message() -> None:
         """Show a message when tracing is disabled."""
         from crewai.events.listeners.tracing.utils import has_user_declined_tracing
 

lib/crewai/src/crewai/crews/utils.py

@@ -0,0 +1,363 @@
+"""Utility functions for crew operations."""
+
+from __future__ import annotations
+
+import asyncio
+from collections.abc import Callable, Coroutine, Iterable
+from typing import TYPE_CHECKING, Any
+
+from crewai.agents.agent_builder.base_agent import BaseAgent
+from crewai.crews.crew_output import CrewOutput
+from crewai.rag.embeddings.types import EmbedderConfig
+from crewai.types.streaming import CrewStreamingOutput, FlowStreamingOutput
+from crewai.utilities.streaming import (
+    StreamingState,
+    TaskInfo,
+    create_streaming_state,
+)
+
+
+if TYPE_CHECKING:
+    from crewai.crew import Crew
+
+
+def enable_agent_streaming(agents: Iterable[BaseAgent]) -> None:
+    """Enable streaming on all agents that have an LLM configured.
+
+    Args:
+        agents: Iterable of agents to enable streaming on.
+    """
+    for agent in agents:
+        if agent.llm is not None:
+            agent.llm.stream = True
+
+
+def setup_agents(
+    crew: Crew,
+    agents: Iterable[BaseAgent],
+    embedder: EmbedderConfig | None,
+    function_calling_llm: Any,
+    step_callback: Callable[..., Any] | None,
+) -> None:
+    """Set up agents for crew execution.
+
+    Args:
+        crew: The crew instance agents belong to.
+        agents: Iterable of agents to set up.
+        embedder: Embedder configuration for knowledge.
+        function_calling_llm: Default function calling LLM for agents.
+        step_callback: Default step callback for agents.
+    """
+    for agent in agents:
+        agent.crew = crew
+        agent.set_knowledge(crew_embedder=embedder)
+        if not agent.function_calling_llm:  # type: ignore[attr-defined]
+            agent.function_calling_llm = function_calling_llm  # type: ignore[attr-defined]
+        if not agent.step_callback:  # type: ignore[attr-defined]
+            agent.step_callback = step_callback  # type: ignore[attr-defined]
+        agent.create_agent_executor()
+
+
+class TaskExecutionData:
+    """Data container for prepared task execution information."""
+
+    def __init__(
+        self,
+        agent: BaseAgent | None,
+        tools: list[Any],
+        should_skip: bool = False,
+    ) -> None:
+        """Initialize task execution data.
+
+        Args:
+            agent: The agent to use for task execution (None if skipped).
+            tools: Prepared tools for the task.
+            should_skip: Whether the task should be skipped (replay).
+        """
+        self.agent = agent
+        self.tools = tools
+        self.should_skip = should_skip
+
+
+def prepare_task_execution(
+    crew: Crew,
+    task: Any,
+    task_index: int,
+    start_index: int | None,
+    task_outputs: list[Any],
+    last_sync_output: Any | None,
+) -> tuple[TaskExecutionData, list[Any], Any | None]:
+    """Prepare a task for execution, handling replay skip logic and agent/tool setup.
+
+    Args:
+        crew: The crew instance.
+        task: The task to prepare.
+        task_index: Index of the current task.
+        start_index: Index to start execution from (for replay).
+        task_outputs: Current list of task outputs.
+        last_sync_output: Last synchronous task output.
+
+    Returns:
+        A tuple of (TaskExecutionData or None if skipped, updated task_outputs, updated last_sync_output).
+        If the task should be skipped, TaskExecutionData will have should_skip=True.
+
+    Raises:
+        ValueError: If no agent is available for the task.
+    """
+    # Handle replay skip
+    if start_index is not None and task_index < start_index:
+        if task.output:
+            if task.async_execution:
+                task_outputs.append(task.output)
+            else:
+                task_outputs = [task.output]
+                last_sync_output = task.output
+        return (
+            TaskExecutionData(agent=None, tools=[], should_skip=True),
+            task_outputs,
+            last_sync_output,
+        )
+
+    agent_to_use = crew._get_agent_to_use(task)
+    if agent_to_use is None:
+        raise ValueError(
+            f"No agent available for task: {task.description}. "
+            f"Ensure that either the task has an assigned agent "
+            f"or a manager agent is provided."
+        )
+
+    tools_for_task = task.tools or agent_to_use.tools or []
+    tools_for_task = crew._prepare_tools(
+        agent_to_use,
+        task,
+        tools_for_task,
+    )
+
+    crew._log_task_start(task, agent_to_use.role)
+
+    return (
+        TaskExecutionData(agent=agent_to_use, tools=tools_for_task),
+        task_outputs,
+        last_sync_output,
+    )
+
+
+def check_conditional_skip(
+    crew: Crew,
+    task: Any,
+    task_outputs: list[Any],
+    task_index: int,
+    was_replayed: bool,
+) -> Any | None:
+    """Check if a conditional task should be skipped.
+
+    Args:
+        crew: The crew instance.
+        task: The conditional task to check.
+        task_outputs: List of previous task outputs.
+        task_index: Index of the current task.
+        was_replayed: Whether this is a replayed execution.
+
+    Returns:
+        The skipped task output if the task should be skipped, None otherwise.
+    """
+    previous_output = task_outputs[-1] if task_outputs else None
+    if previous_output is not None and not task.should_execute(previous_output):
+        crew._logger.log(
+            "debug",
+            f"Skipping conditional task: {task.description}",
+            color="yellow",
+        )
+        skipped_task_output = task.get_skipped_task_output()
+
+        if not was_replayed:
+            crew._store_execution_log(task, skipped_task_output, task_index)
+        return skipped_task_output
+    return None
+
+
+def prepare_kickoff(crew: Crew, inputs: dict[str, Any] | None) -> dict[str, Any] | None:
+    """Prepare crew for kickoff execution.
+
+    Handles before callbacks, event emission, task handler reset, input
+    interpolation, task callbacks, agent setup, and planning.
+
+    Args:
+        crew: The crew instance to prepare.
+        inputs: Optional input dictionary to pass to the crew.
+
+    Returns:
+        The potentially modified inputs dictionary after before callbacks.
+    """
+    from crewai.events.event_bus import crewai_event_bus
+    from crewai.events.types.crew_events import CrewKickoffStartedEvent
+
+    for before_callback in crew.before_kickoff_callbacks:
+        if inputs is None:
+            inputs = {}
+        inputs = before_callback(inputs)
+
+    future = crewai_event_bus.emit(
+        crew,
+        CrewKickoffStartedEvent(crew_name=crew.name, inputs=inputs),
+    )
+    if future is not None:
+        try:
+            future.result()
+        except Exception:  # noqa: S110
+            pass
+
+    crew._task_output_handler.reset()
+    crew._logging_color = "bold_purple"
+
+    if inputs is not None:
+        crew._inputs = inputs
+        crew._interpolate_inputs(inputs)
+    crew._set_tasks_callbacks()
+    crew._set_allow_crewai_trigger_context_for_first_task()
+
+    setup_agents(
+        crew,
+        crew.agents,
+        crew.embedder,
+        crew.function_calling_llm,
+        crew.step_callback,
+    )
+
+    if crew.planning:
+        crew._handle_crew_planning()
+
+    return inputs
+
+
+class StreamingContext:
+    """Container for streaming state and holders used during crew execution."""
+
+    def __init__(self, use_async: bool = False) -> None:
+        """Initialize streaming context.
+
+        Args:
+            use_async: Whether to use async streaming mode.
+        """
+        self.result_holder: list[CrewOutput] = []
+        self.current_task_info: TaskInfo = {
+            "index": 0,
+            "name": "",
+            "id": "",
+            "agent_role": "",
+            "agent_id": "",
+        }
+        self.state: StreamingState = create_streaming_state(
+            self.current_task_info, self.result_holder, use_async=use_async
+        )
+        self.output_holder: list[CrewStreamingOutput | FlowStreamingOutput] = []
+
+
+class ForEachStreamingContext:
+    """Container for streaming state used in for_each crew execution methods."""
+
+    def __init__(self) -> None:
+        """Initialize for_each streaming context."""
+        self.result_holder: list[list[CrewOutput]] = [[]]
+        self.current_task_info: TaskInfo = {
+            "index": 0,
+            "name": "",
+            "id": "",
+            "agent_role": "",
+            "agent_id": "",
+        }
+        self.state: StreamingState = create_streaming_state(
+            self.current_task_info, self.result_holder, use_async=True
+        )
+        self.output_holder: list[CrewStreamingOutput | FlowStreamingOutput] = []
+
+
+async def run_for_each_async(
+    crew: Crew,
+    inputs: list[dict[str, Any]],
+    kickoff_fn: Callable[
+        [Crew, dict[str, Any]], Coroutine[Any, Any, CrewOutput | CrewStreamingOutput]
+    ],
+) -> list[CrewOutput | CrewStreamingOutput] | CrewStreamingOutput:
+    """Execute crew workflow for each input asynchronously.
+
+    Args:
+        crew: The crew instance to execute.
+        inputs: List of input dictionaries for each execution.
+        kickoff_fn: Async function to call for each crew copy (kickoff_async or akickoff).
+
+    Returns:
+        If streaming, a single CrewStreamingOutput that yields chunks from all crews.
+        Otherwise, a list of CrewOutput results.
+    """
+    from crewai.types.usage_metrics import UsageMetrics
+    from crewai.utilities.streaming import (
+        create_async_chunk_generator,
+        signal_end,
+        signal_error,
+    )
+
+    crew_copies = [crew.copy() for _ in inputs]
+
+    if crew.stream:
+        ctx = ForEachStreamingContext()
+
+        async def run_all_crews() -> None:
+            try:
+                streaming_outputs: list[CrewStreamingOutput] = []
+                for i, crew_copy in enumerate(crew_copies):
+                    streaming = await kickoff_fn(crew_copy, inputs[i])
+                    if isinstance(streaming, CrewStreamingOutput):
+                        streaming_outputs.append(streaming)
+
+                async def consume_stream(
+                    stream_output: CrewStreamingOutput,
+                ) -> CrewOutput:
+                    async for chunk in stream_output:
+                        if (
+                            ctx.state.async_queue is not None
+                            and ctx.state.loop is not None
+                        ):
+                            ctx.state.loop.call_soon_threadsafe(
+                                ctx.state.async_queue.put_nowait, chunk
+                            )
+                    return stream_output.result
+
+                crew_results = await asyncio.gather(
+                    *[consume_stream(s) for s in streaming_outputs]
+                )
+                ctx.result_holder[0] = list(crew_results)
+            except Exception as e:
+                signal_error(ctx.state, e, is_async=True)
+            finally:
+                signal_end(ctx.state, is_async=True)
+
+        streaming_output = CrewStreamingOutput(
+            async_iterator=create_async_chunk_generator(
+                ctx.state, run_all_crews, ctx.output_holder
+            )
+        )
+
+        def set_results_wrapper(result: Any) -> None:
+            streaming_output._set_results(result)
+
+        streaming_output._set_result = set_results_wrapper  # type: ignore[method-assign]
+        ctx.output_holder.append(streaming_output)
+
+        return streaming_output
+
+    async_tasks: list[asyncio.Task[CrewOutput | CrewStreamingOutput]] = [
+        asyncio.create_task(kickoff_fn(crew_copy, input_data))
+        for crew_copy, input_data in zip(crew_copies, inputs, strict=True)
+    ]
+
+    results = await asyncio.gather(*async_tasks)
+
+    total_usage_metrics = UsageMetrics()
+    for crew_copy in crew_copies:
+        if crew_copy.usage_metrics:
+            total_usage_metrics.add_usage_metrics(crew_copy.usage_metrics)
+    crew.usage_metrics = total_usage_metrics
+
+    crew._task_output_handler.reset()
+    return list(results)

lib/crewai/src/crewai/events/event_listener.py

@@ -140,7 +140,9 @@ class EventListener(BaseEventListener):
         def on_crew_started(source: Any, event: CrewKickoffStartedEvent) -> None:
             with self._crew_tree_lock:
                 self.formatter.create_crew_tree(event.crew_name or "Crew", source.id)
-                self._telemetry.crew_execution_span(source, event.inputs)
+                source._execution_span = self._telemetry.crew_execution_span(
+                    source, event.inputs
+                )
                 self._crew_tree_lock.notify_all()
 
         @crewai_event_bus.on(CrewKickoffCompletedEvent)

lib/crewai/src/crewai/events/listeners/tracing/trace_batch_manager.py

@@ -76,7 +76,7 @@ class TraceBatchManager:
         use_ephemeral: bool = False,
     ) -> TraceBatch:
         """Initialize a new trace batch (thread-safe)"""
-        with self._init_lock:
+        with self._batch_ready_cv:
             if self.current_batch is not None:
                 logger.debug(
                     "Batch already initialized, skipping duplicate initialization"
@@ -99,7 +99,6 @@ class TraceBatchManager:
                 self.backend_initialized = True
 
             self._batch_ready_cv.notify_all()
-
             return self.current_batch
 
     def _initialize_backend_batch(
@@ -107,7 +106,7 @@ class TraceBatchManager:
         user_context: dict[str, str],
         execution_metadata: dict[str, Any],
         use_ephemeral: bool = False,
-    ):
+    ) -> None:
         """Send batch initialization to backend"""
 
         if not is_tracing_enabled_in_context():
@@ -204,7 +203,7 @@ class TraceBatchManager:
                     return False
         return True
 
-    def add_event(self, trace_event: TraceEvent):
+    def add_event(self, trace_event: TraceEvent) -> None:
         """Add event to buffer"""
         self.event_buffer.append(trace_event)
 
@@ -300,7 +299,7 @@ class TraceBatchManager:
 
         return finalized_batch
 
-    def _finalize_backend_batch(self, events_count: int = 0):
+    def _finalize_backend_batch(self, events_count: int = 0) -> None:
         """Send batch finalization to backend
 
         Args:
@@ -366,7 +365,7 @@ class TraceBatchManager:
             logger.error(f"❌ Error finalizing trace batch: {e}")
             self.plus_api.mark_trace_batch_as_failed(self.trace_batch_id, str(e))
 
-    def _cleanup_batch_data(self):
+    def _cleanup_batch_data(self) -> None:
         """Clean up batch data after successful finalization to free memory"""
         try:
             if hasattr(self, "event_buffer") and self.event_buffer:
@@ -411,7 +410,7 @@ class TraceBatchManager:
                 lambda: self.current_batch is not None, timeout=timeout
             )
 
-    def record_start_time(self, key: str):
+    def record_start_time(self, key: str) -> None:
         """Record start time for duration calculation"""
         self.execution_start_times[key] = datetime.now(timezone.utc)
 

lib/crewai/src/crewai/events/listeners/tracing/trace_listener.py

@@ -71,6 +71,7 @@ from crewai.events.types.reasoning_events import (
     AgentReasoningFailedEvent,
     AgentReasoningStartedEvent,
 )
+from crewai.events.types.system_events import SignalEvent, on_signal
 from crewai.events.types.task_events import (
     TaskCompletedEvent,
     TaskFailedEvent,
@@ -159,6 +160,7 @@ class TraceCollectionListener(BaseEventListener):
         self._register_flow_event_handlers(crewai_event_bus)
         self._register_context_event_handlers(crewai_event_bus)
         self._register_action_event_handlers(crewai_event_bus)
+        self._register_system_event_handlers(crewai_event_bus)
 
         self._listeners_setup = True
 
@@ -458,6 +460,15 @@ class TraceCollectionListener(BaseEventListener):
         ) -> None:
             self._handle_action_event("knowledge_query_failed", source, event)
 
+    def _register_system_event_handlers(self, event_bus: CrewAIEventsBus) -> None:
+        """Register handlers for system signal events (SIGTERM, SIGINT, etc.)."""
+
+        @on_signal
+        def handle_signal(source: Any, event: SignalEvent) -> None:
+            """Flush trace batch on system signals to prevent data loss."""
+            if self.batch_manager.is_batch_initialized():
+                self.batch_manager.finalize_batch()
+
     def _initialize_crew_batch(self, source: Any, event: Any) -> None:
         """Initialize trace batch.
 

lib/crewai/src/crewai/events/types/system_events.py

@@ -0,0 +1,102 @@
+"""System signal event types for CrewAI.
+
+This module contains event types for system-level signals like SIGTERM,
+allowing listeners to perform cleanup operations before process termination.
+"""
+
+from collections.abc import Callable
+from enum import IntEnum
+import signal
+from typing import Annotated, Literal, TypeVar
+
+from pydantic import Field, TypeAdapter
+
+from crewai.events.base_events import BaseEvent
+
+
+class SignalType(IntEnum):
+    """Enumeration of supported system signals."""
+
+    SIGTERM = signal.SIGTERM
+    SIGINT = signal.SIGINT
+    SIGHUP = getattr(signal, "SIGHUP", 1)
+    SIGTSTP = getattr(signal, "SIGTSTP", 20)
+    SIGCONT = getattr(signal, "SIGCONT", 18)
+
+
+class SigTermEvent(BaseEvent):
+    """Event emitted when SIGTERM is received."""
+
+    type: Literal["SIGTERM"] = "SIGTERM"
+    signal_number: SignalType = SignalType.SIGTERM
+    reason: str | None = None
+
+
+class SigIntEvent(BaseEvent):
+    """Event emitted when SIGINT is received."""
+
+    type: Literal["SIGINT"] = "SIGINT"
+    signal_number: SignalType = SignalType.SIGINT
+    reason: str | None = None
+
+
+class SigHupEvent(BaseEvent):
+    """Event emitted when SIGHUP is received."""
+
+    type: Literal["SIGHUP"] = "SIGHUP"
+    signal_number: SignalType = SignalType.SIGHUP
+    reason: str | None = None
+
+
+class SigTStpEvent(BaseEvent):
+    """Event emitted when SIGTSTP is received.
+
+    Note: SIGSTOP cannot be caught - it immediately suspends the process.
+    """
+
+    type: Literal["SIGTSTP"] = "SIGTSTP"
+    signal_number: SignalType = SignalType.SIGTSTP
+    reason: str | None = None
+
+
+class SigContEvent(BaseEvent):
+    """Event emitted when SIGCONT is received."""
+
+    type: Literal["SIGCONT"] = "SIGCONT"
+    signal_number: SignalType = SignalType.SIGCONT
+    reason: str | None = None
+
+
+SignalEvent = Annotated[
+    SigTermEvent | SigIntEvent | SigHupEvent | SigTStpEvent | SigContEvent,
+    Field(discriminator="type"),
+]
+
+signal_event_adapter: TypeAdapter[SignalEvent] = TypeAdapter(SignalEvent)
+
+SIGNAL_EVENT_TYPES: tuple[type[BaseEvent], ...] = (
+    SigTermEvent,
+    SigIntEvent,
+    SigHupEvent,
+    SigTStpEvent,
+    SigContEvent,
+)
+
+
+T = TypeVar("T", bound=Callable[[object, SignalEvent], None])
+
+
+def on_signal(func: T) -> T:
+    """Decorator to register a handler for all signal events.
+
+    Args:
+        func: Handler function that receives (source, event) arguments.
+
+    Returns:
+        The original function, registered for all signal event types.
+    """
+    from crewai.events.event_bus import crewai_event_bus
+
+    for event_type in SIGNAL_EVENT_TYPES:
+        crewai_event_bus.on(event_type)(func)
+    return func

lib/crewai/src/crewai/flow/flow.py

@@ -1032,6 +1032,20 @@ class Flow(Generic[T], metaclass=FlowMeta):
         finally:
             detach(flow_token)
 
+    async def akickoff(
+        self, inputs: dict[str, Any] | None = None
+    ) -> Any | FlowStreamingOutput:
+        """Native async method to start the flow execution. Alias for kickoff_async.
+
+
+        Args:
+            inputs: Optional dictionary containing input values and/or a state ID for restoration.
+
+        Returns:
+            The final output from the flow, which is the result of the last executed method.
+        """
+        return await self.kickoff_async(inputs)
+
     async def _execute_start_method(self, start_method_name: FlowMethodName) -> None:
         """Executes a flow's start method and its triggered listeners.
 

lib/crewai/src/crewai/hooks/llm_hooks.py

@@ -1,6 +1,6 @@
 from __future__ import annotations
 
-from typing import TYPE_CHECKING
+from typing import TYPE_CHECKING, Any, cast
 
 from crewai.events.event_listener import event_listener
 from crewai.hooks.types import AfterLLMCallHookType, BeforeLLMCallHookType
@@ -9,17 +9,22 @@ from crewai.utilities.printer import Printer
 
 if TYPE_CHECKING:
     from crewai.agents.crew_agent_executor import CrewAgentExecutor
+    from crewai.lite_agent import LiteAgent
+    from crewai.llms.base_llm import BaseLLM
+    from crewai.utilities.types import LLMMessage
 
 
 class LLMCallHookContext:
-    """Context object passed to LLM call hooks with full executor access.
+    """Context object passed to LLM call hooks.
 
-    Provides hooks with complete access to the executor state, allowing
+    Provides hooks with complete access to the execution state, allowing
     modification of messages, responses, and executor attributes.
 
+    Supports both executor-based calls (agents in crews/flows) and direct LLM calls.
+
     Attributes:
-        executor: Full reference to the CrewAgentExecutor instance
-        messages: Direct reference to executor.messages (mutable list).
+        executor: Reference to the executor (CrewAgentExecutor/LiteAgent) or None for direct calls
+        messages: Direct reference to messages (mutable list).
             Can be modified in both before_llm_call and after_llm_call hooks.
             Modifications in after_llm_call hooks persist to the next iteration,
             allowing hooks to modify conversation history for subsequent LLM calls.
@@ -27,33 +32,75 @@ class LLMCallHookContext:
             Do NOT replace the list (e.g., context.messages = []), as this will break
             the executor. Use context.messages.append() or context.messages.extend()
             instead of assignment.
-        agent: Reference to the agent executing the task
-        task: Reference to the task being executed
-        crew: Reference to the crew instance
+        agent: Reference to the agent executing the task (None for direct LLM calls)
+        task: Reference to the task being executed (None for direct LLM calls or LiteAgent)
+        crew: Reference to the crew instance (None for direct LLM calls or LiteAgent)
         llm: Reference to the LLM instance
-        iterations: Current iteration count
+        iterations: Current iteration count (0 for direct LLM calls)
         response: LLM response string (only set for after_llm_call hooks).
             Can be modified by returning a new string from after_llm_call hook.
     """
 
+    executor: CrewAgentExecutor | LiteAgent | None
+    messages: list[LLMMessage]
+    agent: Any
+    task: Any
+    crew: Any
+    llm: BaseLLM | None | str | Any
+    iterations: int
+    response: str | None
+
     def __init__(
         self,
-        executor: CrewAgentExecutor,
+        executor: CrewAgentExecutor | LiteAgent | None = None,
         response: str | None = None,
+        messages: list[LLMMessage] | None = None,
+        llm: BaseLLM | str | Any | None = None,  # TODO: look into
+        agent: Any | None = None,
+        task: Any | None = None,
+        crew: Any | None = None,
     ) -> None:
-        """Initialize hook context with executor reference.
+        """Initialize hook context with executor reference or direct parameters.
 
         Args:
-            executor: The CrewAgentExecutor instance
+            executor: The CrewAgentExecutor or LiteAgent instance (None for direct LLM calls)
             response: Optional response string (for after_llm_call hooks)
+            messages: Optional messages list (for direct LLM calls when executor is None)
+            llm: Optional LLM instance (for direct LLM calls when executor is None)
+            agent: Optional agent reference (for direct LLM calls when executor is None)
+            task: Optional task reference (for direct LLM calls when executor is None)
+            crew: Optional crew reference (for direct LLM calls when executor is None)
         """
-        self.executor = executor
-        self.messages = executor.messages
-        self.agent = executor.agent
-        self.task = executor.task
-        self.crew = executor.crew
-        self.llm = executor.llm
-        self.iterations = executor.iterations
+        if executor is not None:
+            # Existing path: extract from executor
+            self.executor = executor
+            self.messages = executor.messages
+            self.llm = executor.llm
+            self.iterations = executor.iterations
+            # Handle CrewAgentExecutor vs LiteAgent differences
+            if hasattr(executor, "agent"):
+                self.agent = executor.agent
+                self.task = cast("CrewAgentExecutor", executor).task
+                self.crew = cast("CrewAgentExecutor", executor).crew
+            else:
+                # LiteAgent case - is the agent itself, doesn't have task/crew
+                self.agent = (
+                    executor.original_agent
+                    if hasattr(executor, "original_agent")
+                    else executor
+                )
+                self.task = None
+                self.crew = None
+        else:
+            # New path: direct LLM call with explicit parameters
+            self.executor = None
+            self.messages = messages or []
+            self.llm = llm
+            self.agent = agent
+            self.task = task
+            self.crew = crew
+            self.iterations = 0
+
         self.response = response
 
     def request_human_input(

lib/crewai/src/crewai/knowledge/knowledge.py

@@ -32,8 +32,8 @@ class Knowledge(BaseModel):
         sources: list[BaseKnowledgeSource],
         embedder: EmbedderConfig | None = None,
         storage: KnowledgeStorage | None = None,
-        **data,
-    ):
+        **data: object,
+    ) -> None:
         super().__init__(**data)
         if storage:
             self.storage = storage
@@ -75,3 +75,44 @@ class Knowledge(BaseModel):
             self.storage.reset()
         else:
             raise ValueError("Storage is not initialized.")
+
+    async def aquery(
+        self, query: list[str], results_limit: int = 5, score_threshold: float = 0.6
+    ) -> list[SearchResult]:
+        """Query across all knowledge sources asynchronously.
+
+        Args:
+            query: List of query strings.
+            results_limit: Maximum number of results to return.
+            score_threshold: Minimum similarity score for results.
+
+        Returns:
+            The top results matching the query.
+
+        Raises:
+            ValueError: If storage is not initialized.
+        """
+        if self.storage is None:
+            raise ValueError("Storage is not initialized.")
+
+        return await self.storage.asearch(
+            query,
+            limit=results_limit,
+            score_threshold=score_threshold,
+        )
+
+    async def aadd_sources(self) -> None:
+        """Add all knowledge sources to storage asynchronously."""
+        try:
+            for source in self.sources:
+                source.storage = self.storage
+                await source.aadd()
+        except Exception as e:
+            raise e
+
+    async def areset(self) -> None:
+        """Reset the knowledge base asynchronously."""
+        if self.storage:
+            await self.storage.areset()
+        else:
+            raise ValueError("Storage is not initialized.")

lib/crewai/src/crewai/knowledge/source/base_file_knowledge_source.py

@@ -1,5 +1,6 @@
 from abc import ABC, abstractmethod
 from pathlib import Path
+from typing import Any
 
 from pydantic import Field, field_validator
 
@@ -25,7 +26,10 @@ class BaseFileKnowledgeSource(BaseKnowledgeSource, ABC):
     safe_file_paths: list[Path] = Field(default_factory=list)
 
     @field_validator("file_path", "file_paths", mode="before")
-    def validate_file_path(cls, v, info):  # noqa: N805
+    @classmethod
+    def validate_file_path(
+        cls, v: Path | list[Path] | str | list[str] | None, info: Any
+    ) -> Path | list[Path] | str | list[str] | None:
         """Validate that at least one of file_path or file_paths is provided."""
         # Single check if both are None, O(1) instead of nested conditions
         if (
@@ -38,7 +42,7 @@ class BaseFileKnowledgeSource(BaseKnowledgeSource, ABC):
             raise ValueError("Either file_path or file_paths must be provided")
         return v
 
-    def model_post_init(self, _):
+    def model_post_init(self, _: Any) -> None:
         """Post-initialization method to load content."""
         self.safe_file_paths = self._process_file_paths()
         self.validate_content()
@@ -48,7 +52,7 @@ class BaseFileKnowledgeSource(BaseKnowledgeSource, ABC):
     def load_content(self) -> dict[Path, str]:
         """Load and preprocess file content. Should be overridden by subclasses. Assume that the file path is relative to the project root in the knowledge directory."""
 
-    def validate_content(self):
+    def validate_content(self) -> None:
         """Validate the paths."""
         for path in self.safe_file_paths:
             if not path.exists():
@@ -65,13 +69,20 @@ class BaseFileKnowledgeSource(BaseKnowledgeSource, ABC):
                     color="red",
                 )
 
-    def _save_documents(self):
+    def _save_documents(self) -> None:
         """Save the documents to the storage."""
         if self.storage:
             self.storage.save(self.chunks)
         else:
             raise ValueError("No storage found to save documents.")
 
+    async def _asave_documents(self) -> None:
+        """Save the documents to the storage asynchronously."""
+        if self.storage:
+            await self.storage.asave(self.chunks)
+        else:
+            raise ValueError("No storage found to save documents.")
+
     def convert_to_path(self, path: Path | str) -> Path:
         """Convert a path to a Path object."""
         return Path(KNOWLEDGE_DIRECTORY + "/" + path) if isinstance(path, str) else path

lib/crewai/src/crewai/knowledge/source/base_knowledge_source.py

@@ -39,12 +39,32 @@ class BaseKnowledgeSource(BaseModel, ABC):
             for i in range(0, len(text), self.chunk_size - self.chunk_overlap)
         ]
 
-    def _save_documents(self):
-        """
-        Save the documents to the storage.
+    def _save_documents(self) -> None:
+        """Save the documents to the storage.
+
         This method should be called after the chunks and embeddings are generated.
+
+        Raises:
+            ValueError: If no storage is configured.
         """
         if self.storage:
             self.storage.save(self.chunks)
         else:
             raise ValueError("No storage found to save documents.")
+
+    @abstractmethod
+    async def aadd(self) -> None:
+        """Process content, chunk it, compute embeddings, and save them asynchronously."""
+
+    async def _asave_documents(self) -> None:
+        """Save the documents to the storage asynchronously.
+
+        This method should be called after the chunks and embeddings are generated.
+
+        Raises:
+            ValueError: If no storage is configured.
+        """
+        if self.storage:
+            await self.storage.asave(self.chunks)
+        else:
+            raise ValueError("No storage found to save documents.")

lib/crewai/src/crewai/knowledge/source/crew_docling_source.py

@@ -2,27 +2,24 @@ from __future__ import annotations
 
 from collections.abc import Iterator
 from pathlib import Path
+from typing import TYPE_CHECKING, Any
 from urllib.parse import urlparse
 
 
 try:
-    from docling.datamodel.base_models import (  # type: ignore[import-not-found]
-        InputFormat,
-    )
-    from docling.document_converter import (  # type: ignore[import-not-found]
-        DocumentConverter,
-    )
-    from docling.exceptions import ConversionError  # type: ignore[import-not-found]
-    from docling_core.transforms.chunker.hierarchical_chunker import (  # type: ignore[import-not-found]
-        HierarchicalChunker,
-    )
-    from docling_core.types.doc.document import (  # type: ignore[import-not-found]
-        DoclingDocument,
-    )
+    from docling.datamodel.base_models import InputFormat
+    from docling.document_converter import DocumentConverter
+    from docling.exceptions import ConversionError
+    from docling_core.transforms.chunker.hierarchical_chunker import HierarchicalChunker
+    from docling_core.types.doc.document import DoclingDocument
 
     DOCLING_AVAILABLE = True
 except ImportError:
     DOCLING_AVAILABLE = False
+    # Provide type stubs for when docling is not available
+    if TYPE_CHECKING:
+        from docling.document_converter import DocumentConverter
+        from docling_core.types.doc.document import DoclingDocument
 
 from pydantic import Field
 
@@ -32,11 +29,13 @@ from crewai.utilities.logger import Logger
 
 
 class CrewDoclingSource(BaseKnowledgeSource):
-    """Default Source class for converting documents to markdown or json
-    This will auto support PDF, DOCX, and TXT, XLSX, Images, and HTML files without any additional dependencies and follows the docling package as the source of truth.
+    """Default Source class for converting documents to markdown or json.
+
+    This will auto support PDF, DOCX, and TXT, XLSX, Images, and HTML files without
+    any additional dependencies and follows the docling package as the source of truth.
     """
 
-    def __init__(self, *args, **kwargs):
+    def __init__(self, *args: Any, **kwargs: Any) -> None:
         if not DOCLING_AVAILABLE:
             raise ImportError(
                 "The docling package is required to use CrewDoclingSource. "
@@ -66,7 +65,7 @@ class CrewDoclingSource(BaseKnowledgeSource):
         )
     )
 
-    def model_post_init(self, _) -> None:
+    def model_post_init(self, _: Any) -> None:
         if self.file_path:
             self._logger.log(
                 "warning",
@@ -99,6 +98,15 @@ class CrewDoclingSource(BaseKnowledgeSource):
             self.chunks.extend(list(new_chunks_iterable))
         self._save_documents()
 
+    async def aadd(self) -> None:
+        """Add docling content asynchronously."""
+        if self.content is None:
+            return
+        for doc in self.content:
+            new_chunks_iterable = self._chunk_doc(doc)
+            self.chunks.extend(list(new_chunks_iterable))
+        await self._asave_documents()
+
     def _convert_source_to_docling_documents(self) -> list[DoclingDocument]:
         conv_results_iter = self.document_converter.convert_all(self.safe_file_paths)
         return [result.document for result in conv_results_iter]

lib/crewai/src/crewai/knowledge/source/csv_knowledge_source.py

@@ -31,6 +31,15 @@ class CSVKnowledgeSource(BaseFileKnowledgeSource):
         self.chunks.extend(new_chunks)
         self._save_documents()
 
+    async def aadd(self) -> None:
+        """Add CSV file content asynchronously."""
+        content_str = (
+            str(self.content) if isinstance(self.content, dict) else self.content
+        )
+        new_chunks = self._chunk_text(content_str)
+        self.chunks.extend(new_chunks)
+        await self._asave_documents()
+
     def _chunk_text(self, text: str) -> list[str]:
         """Utility method to split text into chunks."""
         return [

lib/crewai/src/crewai/knowledge/source/excel_knowledge_source.py

@@ -1,4 +1,6 @@
 from pathlib import Path
+from types import ModuleType
+from typing import Any
 
 from pydantic import Field, field_validator
 
@@ -26,7 +28,10 @@ class ExcelKnowledgeSource(BaseKnowledgeSource):
     safe_file_paths: list[Path] = Field(default_factory=list)
 
     @field_validator("file_path", "file_paths", mode="before")
-    def validate_file_path(cls, v, info):  # noqa: N805
+    @classmethod
+    def validate_file_path(
+        cls, v: Path | list[Path] | str | list[str] | None, info: Any
+    ) -> Path | list[Path] | str | list[str] | None:
         """Validate that at least one of file_path or file_paths is provided."""
         # Single check if both are None, O(1) instead of nested conditions
         if (
@@ -69,7 +74,7 @@ class ExcelKnowledgeSource(BaseKnowledgeSource):
 
         return [self.convert_to_path(path) for path in path_list]
 
-    def validate_content(self):
+    def validate_content(self) -> None:
         """Validate the paths."""
         for path in self.safe_file_paths:
             if not path.exists():
@@ -86,7 +91,7 @@ class ExcelKnowledgeSource(BaseKnowledgeSource):
                     color="red",
                 )
 
-    def model_post_init(self, _) -> None:
+    def model_post_init(self, _: Any) -> None:
         if self.file_path:
             self._logger.log(
                 "warning",
@@ -128,12 +133,12 @@ class ExcelKnowledgeSource(BaseKnowledgeSource):
         """Convert a path to a Path object."""
         return Path(KNOWLEDGE_DIRECTORY + "/" + path) if isinstance(path, str) else path
 
-    def _import_dependencies(self):
+    def _import_dependencies(self) -> ModuleType:
         """Dynamically import dependencies."""
         try:
-            import pandas as pd  # type: ignore[import-untyped,import-not-found]
+            import pandas as pd  # type: ignore[import-untyped]
 
-            return pd
+            return pd  # type: ignore[no-any-return]
         except ImportError as e:
             missing_package = str(e).split()[-1]
             raise ImportError(
@@ -159,6 +164,20 @@ class ExcelKnowledgeSource(BaseKnowledgeSource):
         self.chunks.extend(new_chunks)
         self._save_documents()
 
+    async def aadd(self) -> None:
+        """Add Excel file content asynchronously."""
+        content_str = ""
+        for value in self.content.values():
+            if isinstance(value, dict):
+                for sheet_value in value.values():
+                    content_str += str(sheet_value) + "\n"
+            else:
+                content_str += str(value) + "\n"
+
+        new_chunks = self._chunk_text(content_str)
+        self.chunks.extend(new_chunks)
+        await self._asave_documents()
+
     def _chunk_text(self, text: str) -> list[str]:
         """Utility method to split text into chunks."""
         return [

lib/crewai/src/crewai/knowledge/source/json_knowledge_source.py

@@ -44,6 +44,15 @@ class JSONKnowledgeSource(BaseFileKnowledgeSource):
         self.chunks.extend(new_chunks)
         self._save_documents()
 
+    async def aadd(self) -> None:
+        """Add JSON file content asynchronously."""
+        content_str = (
+            str(self.content) if isinstance(self.content, dict) else self.content
+        )
+        new_chunks = self._chunk_text(content_str)
+        self.chunks.extend(new_chunks)
+        await self._asave_documents()
+
     def _chunk_text(self, text: str) -> list[str]:
         """Utility method to split text into chunks."""
         return [

lib/crewai/src/crewai/knowledge/source/pdf_knowledge_source.py

@@ -1,4 +1,5 @@
 from pathlib import Path
+from types import ModuleType
 
 from crewai.knowledge.source.base_file_knowledge_source import BaseFileKnowledgeSource
 
@@ -23,7 +24,7 @@ class PDFKnowledgeSource(BaseFileKnowledgeSource):
             content[path] = text
         return content
 
-    def _import_pdfplumber(self):
+    def _import_pdfplumber(self) -> ModuleType:
         """Dynamically import pdfplumber."""
         try:
             import pdfplumber
@@ -44,6 +45,13 @@ class PDFKnowledgeSource(BaseFileKnowledgeSource):
             self.chunks.extend(new_chunks)
         self._save_documents()
 
+    async def aadd(self) -> None:
+        """Add PDF file content asynchronously."""
+        for text in self.content.values():
+            new_chunks = self._chunk_text(text)
+            self.chunks.extend(new_chunks)
+        await self._asave_documents()
+
     def _chunk_text(self, text: str) -> list[str]:
         """Utility method to split text into chunks."""
         return [

lib/crewai/src/crewai/knowledge/source/string_knowledge_source.py

@@ -1,3 +1,5 @@
+from typing import Any
+
 from pydantic import Field
 
 from crewai.knowledge.source.base_knowledge_source import BaseKnowledgeSource
@@ -9,11 +11,11 @@ class StringKnowledgeSource(BaseKnowledgeSource):
     content: str = Field(...)
     collection_name: str | None = Field(default=None)
 
-    def model_post_init(self, _):
+    def model_post_init(self, _: Any) -> None:
         """Post-initialization method to validate content."""
         self.validate_content()
 
-    def validate_content(self):
+    def validate_content(self) -> None:
         """Validate string content."""
         if not isinstance(self.content, str):
             raise ValueError("StringKnowledgeSource only accepts string content")
@@ -24,6 +26,12 @@ class StringKnowledgeSource(BaseKnowledgeSource):
         self.chunks.extend(new_chunks)
         self._save_documents()
 
+    async def aadd(self) -> None:
+        """Add string content asynchronously."""
+        new_chunks = self._chunk_text(self.content)
+        self.chunks.extend(new_chunks)
+        await self._asave_documents()
+
     def _chunk_text(self, text: str) -> list[str]:
         """Utility method to split text into chunks."""
         return [

lib/crewai/src/crewai/knowledge/source/text_file_knowledge_source.py

@@ -25,6 +25,13 @@ class TextFileKnowledgeSource(BaseFileKnowledgeSource):
             self.chunks.extend(new_chunks)
         self._save_documents()
 
+    async def aadd(self) -> None:
+        """Add text file content asynchronously."""
+        for text in self.content.values():
+            new_chunks = self._chunk_text(text)
+            self.chunks.extend(new_chunks)
+        await self._asave_documents()
+
     def _chunk_text(self, text: str) -> list[str]:
         """Utility method to split text into chunks."""
         return [

lib/crewai/src/crewai/knowledge/storage/base_knowledge_storage.py

@@ -21,10 +21,28 @@ class BaseKnowledgeStorage(ABC):
     ) -> list[SearchResult]:
         """Search for documents in the knowledge base."""
 
+    @abstractmethod
+    async def asearch(
+        self,
+        query: list[str],
+        limit: int = 5,
+        metadata_filter: dict[str, Any] | None = None,
+        score_threshold: float = 0.6,
+    ) -> list[SearchResult]:
+        """Search for documents in the knowledge base asynchronously."""
+
     @abstractmethod
     def save(self, documents: list[str]) -> None:
         """Save documents to the knowledge base."""
 
+    @abstractmethod
+    async def asave(self, documents: list[str]) -> None:
+        """Save documents to the knowledge base asynchronously."""
+
     @abstractmethod
     def reset(self) -> None:
         """Reset the knowledge base."""
+
+    @abstractmethod
+    async def areset(self) -> None:
+        """Reset the knowledge base asynchronously."""

lib/crewai/src/crewai/knowledge/storage/knowledge_storage.py

@@ -25,8 +25,8 @@ class KnowledgeStorage(BaseKnowledgeStorage):
     def __init__(
         self,
         embedder: ProviderSpec
-        | BaseEmbeddingsProvider
-        | type[BaseEmbeddingsProvider]
+        | BaseEmbeddingsProvider[Any]
+        | type[BaseEmbeddingsProvider[Any]]
         | None = None,
         collection_name: str | None = None,
     ) -> None:
@@ -127,3 +127,96 @@ class KnowledgeStorage(BaseKnowledgeStorage):
                 ) from e
             Logger(verbose=True).log("error", f"Failed to upsert documents: {e}", "red")
             raise
+
+    async def asearch(
+        self,
+        query: list[str],
+        limit: int = 5,
+        metadata_filter: dict[str, Any] | None = None,
+        score_threshold: float = 0.6,
+    ) -> list[SearchResult]:
+        """Search for documents in the knowledge base asynchronously.
+
+        Args:
+            query: List of query strings.
+            limit: Maximum number of results to return.
+            metadata_filter: Optional metadata filter for the search.
+            score_threshold: Minimum similarity score for results.
+
+        Returns:
+            List of search results.
+        """
+        try:
+            if not query:
+                raise ValueError("Query cannot be empty")
+
+            client = self._get_client()
+            collection_name = (
+                f"knowledge_{self.collection_name}"
+                if self.collection_name
+                else "knowledge"
+            )
+            query_text = " ".join(query) if len(query) > 1 else query[0]
+
+            return await client.asearch(
+                collection_name=collection_name,
+                query=query_text,
+                limit=limit,
+                metadata_filter=metadata_filter,
+                score_threshold=score_threshold,
+            )
+        except Exception as e:
+            logging.error(
+                f"Error during knowledge search: {e!s}\n{traceback.format_exc()}"
+            )
+            return []
+
+    async def asave(self, documents: list[str]) -> None:
+        """Save documents to the knowledge base asynchronously.
+
+        Args:
+            documents: List of document strings to save.
+        """
+        try:
+            client = self._get_client()
+            collection_name = (
+                f"knowledge_{self.collection_name}"
+                if self.collection_name
+                else "knowledge"
+            )
+            await client.aget_or_create_collection(collection_name=collection_name)
+
+            rag_documents: list[BaseRecord] = [{"content": doc} for doc in documents]
+
+            await client.aadd_documents(
+                collection_name=collection_name, documents=rag_documents
+            )
+        except Exception as e:
+            if "dimension mismatch" in str(e).lower():
+                Logger(verbose=True).log(
+                    "error",
+                    "Embedding dimension mismatch. This usually happens when mixing different embedding models. Try resetting the collection using `crewai reset-memories -a`",
+                    "red",
+                )
+                raise ValueError(
+                    "Embedding dimension mismatch. Make sure you're using the same embedding model "
+                    "across all operations with this collection."
+                    "Try resetting the collection using `crewai reset-memories -a`"
+                ) from e
+            Logger(verbose=True).log("error", f"Failed to upsert documents: {e}", "red")
+            raise
+
+    async def areset(self) -> None:
+        """Reset the knowledge base asynchronously."""
+        try:
+            client = self._get_client()
+            collection_name = (
+                f"knowledge_{self.collection_name}"
+                if self.collection_name
+                else "knowledge"
+            )
+            await client.adelete_collection(collection_name=collection_name)
+        except Exception as e:
+            logging.error(
+                f"Error during knowledge reset: {e!s}\n{traceback.format_exc()}"
+            )

lib/crewai/src/crewai/lite_agent.py

@@ -38,6 +38,8 @@ from crewai.events.types.agent_events import (
 )
 from crewai.events.types.logging_events import AgentLogsExecutionEvent
 from crewai.flow.flow_trackable import FlowTrackable
+from crewai.hooks.llm_hooks import get_after_llm_call_hooks, get_before_llm_call_hooks
+from crewai.hooks.types import AfterLLMCallHookType, BeforeLLMCallHookType
 from crewai.lite_agent_output import LiteAgentOutput
 from crewai.llm import LLM
 from crewai.llms.base_llm import BaseLLM
@@ -155,6 +157,12 @@ class LiteAgent(FlowTrackable, BaseModel):
     _guardrail: GuardrailCallable | None = PrivateAttr(default=None)
     _guardrail_retry_count: int = PrivateAttr(default=0)
     _callbacks: list[TokenCalcHandler] = PrivateAttr(default_factory=list)
+    _before_llm_call_hooks: list[BeforeLLMCallHookType] = PrivateAttr(
+        default_factory=get_before_llm_call_hooks
+    )
+    _after_llm_call_hooks: list[AfterLLMCallHookType] = PrivateAttr(
+        default_factory=get_after_llm_call_hooks
+    )
 
     @model_validator(mode="after")
     def setup_llm(self) -> Self:
@@ -246,6 +254,26 @@ class LiteAgent(FlowTrackable, BaseModel):
         """Return the original role for compatibility with tool interfaces."""
         return self.role
 
+    @property
+    def before_llm_call_hooks(self) -> list[BeforeLLMCallHookType]:
+        """Get the before_llm_call hooks for this agent."""
+        return self._before_llm_call_hooks
+
+    @property
+    def after_llm_call_hooks(self) -> list[AfterLLMCallHookType]:
+        """Get the after_llm_call hooks for this agent."""
+        return self._after_llm_call_hooks
+
+    @property
+    def messages(self) -> list[LLMMessage]:
+        """Get the messages list for hook context compatibility."""
+        return self._messages
+
+    @property
+    def iterations(self) -> int:
+        """Get the current iteration count for hook context compatibility."""
+        return self._iterations
+
     def kickoff(
         self,
         messages: str | list[LLMMessage],
@@ -504,7 +532,7 @@ class LiteAgent(FlowTrackable, BaseModel):
             AgentFinish: The final result of the agent execution.
         """
         # Execute the agent loop
-        formatted_answer = None
+        formatted_answer: AgentAction | AgentFinish | None = None
         while not isinstance(formatted_answer, AgentFinish):
             try:
                 if has_reached_max_iterations(self._iterations, self.max_iterations):
@@ -526,6 +554,7 @@ class LiteAgent(FlowTrackable, BaseModel):
                         callbacks=self._callbacks,
                         printer=self._printer,
                         from_agent=self,
+                        executor_context=self,
                     )
 
                 except Exception as e:

lib/crewai/src/crewai/llm.py

@@ -57,11 +57,17 @@ if TYPE_CHECKING:
     from litellm.litellm_core_utils.get_supported_openai_params import (
         get_supported_openai_params,
     )
-    from litellm.types.utils import ChatCompletionDeltaToolCall, Choices, ModelResponse
+    from litellm.types.utils import (
+        ChatCompletionDeltaToolCall,
+        Choices,
+        Function,
+        ModelResponse,
+    )
     from litellm.utils import supports_response_schema
 
     from crewai.agent.core import Agent
     from crewai.llms.hooks.base import BaseInterceptor
+    from crewai.llms.providers.anthropic.completion import AnthropicThinkingConfig
     from crewai.task import Task
     from crewai.tools.base_tool import BaseTool
     from crewai.utilities.types import LLMMessage
@@ -73,7 +79,12 @@ try:
     from litellm.litellm_core_utils.get_supported_openai_params import (
         get_supported_openai_params,
     )
-    from litellm.types.utils import ChatCompletionDeltaToolCall, Choices, ModelResponse
+    from litellm.types.utils import (
+        ChatCompletionDeltaToolCall,
+        Choices,
+        Function,
+        ModelResponse,
+    )
     from litellm.utils import supports_response_schema
 
     LITELLM_AVAILABLE = True
@@ -84,6 +95,7 @@ except ImportError:
     ContextWindowExceededError = Exception  # type: ignore
     get_supported_openai_params = None  # type: ignore
     ChatCompletionDeltaToolCall = None  # type: ignore
+    Function = None  # type: ignore
     ModelResponse = None  # type: ignore
     supports_response_schema = None  # type: ignore
     CustomLogger = None  # type: ignore
@@ -574,6 +586,7 @@ class LLM(BaseLLM):
         reasoning_effort: Literal["none", "low", "medium", "high"] | None = None,
         stream: bool = False,
         interceptor: BaseInterceptor[httpx.Request, httpx.Response] | None = None,
+        thinking: AnthropicThinkingConfig | dict[str, Any] | None = None,
         **kwargs: Any,
     ) -> None:
         """Initialize LLM instance.
@@ -610,7 +623,9 @@ class LLM(BaseLLM):
         self.callbacks = callbacks
         self.context_window_size = 0
         self.reasoning_effort = reasoning_effort
-        self.additional_params = kwargs
+        self.additional_params = {
+            k: v for k, v in kwargs.items() if k not in ("is_litellm", "provider")
+        }
         self.is_anthropic = self._is_anthropic_model(model)
         self.stream = stream
         self.interceptor = interceptor
@@ -1204,6 +1219,281 @@ class LLM(BaseLLM):
         )
         return text_response
 
+    async def _ahandle_non_streaming_response(
+        self,
+        params: dict[str, Any],
+        callbacks: list[Any] | None = None,
+        available_functions: dict[str, Any] | None = None,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> str | Any:
+        """Handle an async non-streaming response from the LLM.
+
+        Args:
+            params: Parameters for the completion call
+            callbacks: Optional list of callback functions
+            available_functions: Dict of available functions
+            from_task: Optional Task that invoked the LLM
+            from_agent: Optional Agent that invoked the LLM
+            response_model: Optional Response model
+
+        Returns:
+            str: The response text
+        """
+        if response_model and self.is_litellm:
+            from crewai.utilities.internal_instructor import InternalInstructor
+
+            messages = params.get("messages", [])
+            if not messages:
+                raise ValueError("Messages are required when using response_model")
+
+            combined_content = "\n\n".join(
+                f"{msg['role'].upper()}: {msg['content']}" for msg in messages
+            )
+
+            instructor_instance = InternalInstructor(
+                content=combined_content,
+                model=response_model,
+                llm=self,
+            )
+            result = instructor_instance.to_pydantic()
+            structured_response = result.model_dump_json()
+            self._handle_emit_call_events(
+                response=structured_response,
+                call_type=LLMCallType.LLM_CALL,
+                from_task=from_task,
+                from_agent=from_agent,
+                messages=params["messages"],
+            )
+            return structured_response
+
+        try:
+            if response_model:
+                params["response_model"] = response_model
+            response = await litellm.acompletion(**params)
+
+        except ContextWindowExceededError as e:
+            raise LLMContextLengthExceededError(str(e)) from e
+
+        if response_model is not None:
+            if isinstance(response, BaseModel):
+                structured_response = response.model_dump_json()
+                self._handle_emit_call_events(
+                    response=structured_response,
+                    call_type=LLMCallType.LLM_CALL,
+                    from_task=from_task,
+                    from_agent=from_agent,
+                    messages=params["messages"],
+                )
+                return structured_response
+
+        response_message = cast(Choices, cast(ModelResponse, response).choices)[
+            0
+        ].message
+        text_response = response_message.content or ""
+
+        if callbacks and len(callbacks) > 0:
+            for callback in callbacks:
+                if hasattr(callback, "log_success_event"):
+                    usage_info = getattr(response, "usage", None)
+                    if usage_info:
+                        callback.log_success_event(
+                            kwargs=params,
+                            response_obj={"usage": usage_info},
+                            start_time=0,
+                            end_time=0,
+                        )
+
+        tool_calls = getattr(response_message, "tool_calls", [])
+
+        if (not tool_calls or not available_functions) and text_response:
+            self._handle_emit_call_events(
+                response=text_response,
+                call_type=LLMCallType.LLM_CALL,
+                from_task=from_task,
+                from_agent=from_agent,
+                messages=params["messages"],
+            )
+            return text_response
+
+        if tool_calls and not available_functions and not text_response:
+            return tool_calls
+
+        tool_result = self._handle_tool_call(
+            tool_calls, available_functions, from_task, from_agent
+        )
+        if tool_result is not None:
+            return tool_result
+
+        self._handle_emit_call_events(
+            response=text_response,
+            call_type=LLMCallType.LLM_CALL,
+            from_task=from_task,
+            from_agent=from_agent,
+            messages=params["messages"],
+        )
+        return text_response
+
+    async def _ahandle_streaming_response(
+        self,
+        params: dict[str, Any],
+        callbacks: list[Any] | None = None,
+        available_functions: dict[str, Any] | None = None,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> Any:
+        """Handle an async streaming response from the LLM.
+
+        Args:
+            params: Parameters for the completion call
+            callbacks: Optional list of callback functions
+            available_functions: Dict of available functions
+            from_task: Optional task object
+            from_agent: Optional agent object
+            response_model: Optional response model
+
+        Returns:
+            str: The complete response text
+        """
+        full_response = ""
+        chunk_count = 0
+        usage_info = None
+
+        accumulated_tool_args: defaultdict[int, AccumulatedToolArgs] = defaultdict(
+            AccumulatedToolArgs
+        )
+
+        params["stream"] = True
+        params["stream_options"] = {"include_usage": True}
+
+        try:
+            async for chunk in await litellm.acompletion(**params):
+                chunk_count += 1
+                chunk_content = None
+
+                try:
+                    choices = None
+                    if isinstance(chunk, dict) and "choices" in chunk:
+                        choices = chunk["choices"]
+                    elif hasattr(chunk, "choices"):
+                        if not isinstance(chunk.choices, type):
+                            choices = chunk.choices
+
+                    if hasattr(chunk, "usage") and chunk.usage is not None:
+                        usage_info = chunk.usage
+
+                    if choices and len(choices) > 0:
+                        first_choice = choices[0]
+                        delta = None
+
+                        if isinstance(first_choice, dict):
+                            delta = first_choice.get("delta", {})
+                        elif hasattr(first_choice, "delta"):
+                            delta = first_choice.delta
+
+                        if delta:
+                            if isinstance(delta, dict):
+                                chunk_content = delta.get("content")
+                            elif hasattr(delta, "content"):
+                                chunk_content = delta.content
+
+                            tool_calls: list[ChatCompletionDeltaToolCall] | None = None
+                            if isinstance(delta, dict):
+                                tool_calls = delta.get("tool_calls")
+                            elif hasattr(delta, "tool_calls"):
+                                tool_calls = delta.tool_calls
+
+                            if tool_calls:
+                                for tool_call in tool_calls:
+                                    idx = tool_call.index
+                                    if tool_call.function:
+                                        if tool_call.function.name:
+                                            accumulated_tool_args[
+                                                idx
+                                            ].function.name = tool_call.function.name
+                                        if tool_call.function.arguments:
+                                            accumulated_tool_args[
+                                                idx
+                                            ].function.arguments += (
+                                                tool_call.function.arguments
+                                            )
+
+                except (AttributeError, KeyError, IndexError, TypeError):
+                    pass
+
+                if chunk_content:
+                    full_response += chunk_content
+                    crewai_event_bus.emit(
+                        self,
+                        event=LLMStreamChunkEvent(
+                            chunk=chunk_content,
+                            from_task=from_task,
+                            from_agent=from_agent,
+                        ),
+                    )
+
+            if callbacks and len(callbacks) > 0 and usage_info:
+                for callback in callbacks:
+                    if hasattr(callback, "log_success_event"):
+                        callback.log_success_event(
+                            kwargs=params,
+                            response_obj={"usage": usage_info},
+                            start_time=0,
+                            end_time=0,
+                        )
+
+            if accumulated_tool_args and available_functions:
+                # Convert accumulated tool args to ChatCompletionDeltaToolCall objects
+                tool_calls_list: list[ChatCompletionDeltaToolCall] = [
+                    ChatCompletionDeltaToolCall(
+                        index=idx,
+                        function=Function(
+                            name=tool_arg.function.name,
+                            arguments=tool_arg.function.arguments,
+                        ),
+                    )
+                    for idx, tool_arg in accumulated_tool_args.items()
+                    if tool_arg.function.name
+                ]
+
+                if tool_calls_list:
+                    result = self._handle_streaming_tool_calls(
+                        tool_calls=tool_calls_list,
+                        accumulated_tool_args=accumulated_tool_args,
+                        available_functions=available_functions,
+                        from_task=from_task,
+                        from_agent=from_agent,
+                    )
+                    if result is not None:
+                        return result
+
+            self._handle_emit_call_events(
+                response=full_response,
+                call_type=LLMCallType.LLM_CALL,
+                from_task=from_task,
+                from_agent=from_agent,
+                messages=params.get("messages"),
+            )
+            return full_response
+
+        except ContextWindowExceededError as e:
+            raise LLMContextLengthExceededError(str(e)) from e
+        except Exception:
+            if chunk_count == 0:
+                raise
+            if full_response:
+                self._handle_emit_call_events(
+                    response=full_response,
+                    call_type=LLMCallType.LLM_CALL,
+                    from_task=from_task,
+                    from_agent=from_agent,
+                    messages=params.get("messages"),
+                )
+                return full_response
+            raise
+
     def _handle_tool_call(
         self,
         tool_calls: list[Any],
@@ -1354,6 +1644,10 @@ class LLM(BaseLLM):
                 if message.get("role") == "system":
                     msg_role: Literal["assistant"] = "assistant"
                     message["role"] = msg_role
+
+        if not self._invoke_before_llm_call_hooks(messages, from_agent):
+            raise ValueError("LLM call blocked by before_llm_call hook")
+
         # --- 5) Set up callbacks if provided
         with suppress_warnings():
             if callbacks and len(callbacks) > 0:
@@ -1363,7 +1657,16 @@ class LLM(BaseLLM):
                 params = self._prepare_completion_params(messages, tools)
                 # --- 7) Make the completion call and handle response
                 if self.stream:
-                    return self._handle_streaming_response(
+                    result = self._handle_streaming_response(
+                        params=params,
+                        callbacks=callbacks,
+                        available_functions=available_functions,
+                        from_task=from_task,
+                        from_agent=from_agent,
+                        response_model=response_model,
+                    )
+                else:
+                    result = self._handle_non_streaming_response(
                         params=params,
                         callbacks=callbacks,
                         available_functions=available_functions,
@@ -1372,14 +1675,12 @@ class LLM(BaseLLM):
                         response_model=response_model,
                     )
 
-                return self._handle_non_streaming_response(
-                    params=params,
-                    callbacks=callbacks,
-                    available_functions=available_functions,
-                    from_task=from_task,
-                    from_agent=from_agent,
-                    response_model=response_model,
-                )
+                if isinstance(result, str):
+                    result = self._invoke_after_llm_call_hooks(
+                        messages, result, from_agent
+                    )
+
+                return result
             except LLMContextLengthExceededError:
                 # Re-raise LLMContextLengthExceededError as it should be handled
                 # by the CrewAgentExecutor._invoke_loop method, which can then decide
@@ -1421,6 +1722,128 @@ class LLM(BaseLLM):
                 )
                 raise
 
+    async def acall(
+        self,
+        messages: str | list[LLMMessage],
+        tools: list[dict[str, BaseTool]] | None = None,
+        callbacks: list[Any] | None = None,
+        available_functions: dict[str, Any] | None = None,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> str | Any:
+        """Async high-level LLM call method.
+
+        Args:
+            messages: Input messages for the LLM.
+                     Can be a string or list of message dictionaries.
+                     If string, it will be converted to a single user message.
+                     If list, each dict must have 'role' and 'content' keys.
+            tools: Optional list of tool schemas for function calling.
+                  Each tool should define its name, description, and parameters.
+            callbacks: Optional list of callback functions to be executed
+                      during and after the LLM call.
+            available_functions: Optional dict mapping function names to callables
+                               that can be invoked by the LLM.
+            from_task: Optional Task that invoked the LLM
+            from_agent: Optional Agent that invoked the LLM
+            response_model: Optional Model that contains a pydantic response model.
+
+        Returns:
+            Union[str, Any]: Either a text response from the LLM (str) or
+                           the result of a tool function call (Any).
+
+        Raises:
+            TypeError: If messages format is invalid
+            ValueError: If response format is not supported
+            LLMContextLengthExceededError: If input exceeds model's context limit
+        """
+        crewai_event_bus.emit(
+            self,
+            event=LLMCallStartedEvent(
+                messages=messages,
+                tools=tools,
+                callbacks=callbacks,
+                available_functions=available_functions,
+                from_task=from_task,
+                from_agent=from_agent,
+                model=self.model,
+            ),
+        )
+
+        self._validate_call_params()
+
+        if isinstance(messages, str):
+            messages = [{"role": "user", "content": messages}]
+
+        if "o1" in self.model.lower():
+            for message in messages:
+                if message.get("role") == "system":
+                    msg_role: Literal["assistant"] = "assistant"
+                    message["role"] = msg_role
+
+        with suppress_warnings():
+            if callbacks and len(callbacks) > 0:
+                self.set_callbacks(callbacks)
+            try:
+                params = self._prepare_completion_params(messages, tools)
+
+                if self.stream:
+                    return await self._ahandle_streaming_response(
+                        params=params,
+                        callbacks=callbacks,
+                        available_functions=available_functions,
+                        from_task=from_task,
+                        from_agent=from_agent,
+                        response_model=response_model,
+                    )
+
+                return await self._ahandle_non_streaming_response(
+                    params=params,
+                    callbacks=callbacks,
+                    available_functions=available_functions,
+                    from_task=from_task,
+                    from_agent=from_agent,
+                    response_model=response_model,
+                )
+            except LLMContextLengthExceededError:
+                raise
+            except Exception as e:
+                unsupported_stop = "Unsupported parameter" in str(
+                    e
+                ) and "'stop'" in str(e)
+
+                if unsupported_stop:
+                    if (
+                        "additional_drop_params" in self.additional_params
+                        and isinstance(
+                            self.additional_params["additional_drop_params"], list
+                        )
+                    ):
+                        self.additional_params["additional_drop_params"].append("stop")
+                    else:
+                        self.additional_params = {"additional_drop_params": ["stop"]}
+
+                    logging.info("Retrying LLM call without the unsupported 'stop'")
+
+                    return await self.acall(
+                        messages,
+                        tools=tools,
+                        callbacks=callbacks,
+                        available_functions=available_functions,
+                        from_task=from_task,
+                        from_agent=from_agent,
+                        response_model=response_model,
+                    )
+
+                crewai_event_bus.emit(
+                    self,
+                    event=LLMCallFailedEvent(
+                        error=str(e), from_task=from_task, from_agent=from_agent
+                    ),
+                )
+                raise
+
     def _handle_emit_call_events(
         self,
         response: Any,

lib/crewai/src/crewai/llms/base_llm.py

@@ -158,6 +158,44 @@ class BaseLLM(ABC):
             RuntimeError: If the LLM request fails for other reasons.
         """
 
+    async def acall(
+        self,
+        messages: str | list[LLMMessage],
+        tools: list[dict[str, BaseTool]] | None = None,
+        callbacks: list[Any] | None = None,
+        available_functions: dict[str, Any] | None = None,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> str | Any:
+        """Call the LLM with the given messages.
+
+        Args:
+            messages: Input messages for the LLM.
+                     Can be a string or list of message dictionaries.
+                     If string, it will be converted to a single user message.
+                     If list, each dict must have 'role' and 'content' keys.
+            tools: Optional list of tool schemas for function calling.
+                  Each tool should define its name, description, and parameters.
+            callbacks: Optional list of callback functions to be executed
+                      during and after the LLM call.
+            available_functions: Optional dict mapping function names to callables
+                               that can be invoked by the LLM.
+            from_task: Optional task caller to be used for the LLM call.
+            from_agent: Optional agent caller to be used for the LLM call.
+            response_model: Optional response model to be used for the LLM call.
+
+        Returns:
+            Either a text response from the LLM (str) or
+            the result of a tool function call (Any).
+
+        Raises:
+            ValueError: If the messages format is invalid.
+            TimeoutError: If the LLM request times out.
+            RuntimeError: If the LLM request fails for other reasons.
+        """
+        raise NotImplementedError
+
     def _convert_tools_for_interference(
         self, tools: list[dict[str, BaseTool]]
     ) -> list[dict[str, BaseTool]]:
@@ -276,7 +314,7 @@ class BaseLLM(ABC):
         call_type: LLMCallType,
         from_task: Task | None = None,
         from_agent: Agent | None = None,
-        messages: str | list[dict[str, Any]] | None = None,
+        messages: str | list[LLMMessage] | None = None,
     ) -> None:
         """Emit LLM call completed event."""
         crewai_event_bus.emit(
@@ -548,3 +586,134 @@ class BaseLLM(ABC):
             Dictionary with token usage totals
         """
         return UsageMetrics(**self._token_usage)
+
+    def _invoke_before_llm_call_hooks(
+        self,
+        messages: list[LLMMessage],
+        from_agent: Agent | None = None,
+    ) -> bool:
+        """Invoke before_llm_call hooks for direct LLM calls (no agent context).
+
+        This method should be called by native provider implementations before
+        making the actual LLM call when from_agent is None (direct calls).
+
+        Args:
+            messages: The messages being sent to the LLM
+            from_agent: The agent making the call (None for direct calls)
+
+        Returns:
+            True if LLM call should proceed, False if blocked by hook
+
+        Example:
+            >>> # In a native provider's call() method:
+            >>> if from_agent is None and not self._invoke_before_llm_call_hooks(
+            ...     messages, from_agent
+            ... ):
+            ...     raise ValueError("LLM call blocked by hook")
+        """
+        # Only invoke hooks for direct calls (no agent context)
+        if from_agent is not None:
+            return True
+
+        from crewai.hooks.llm_hooks import (
+            LLMCallHookContext,
+            get_before_llm_call_hooks,
+        )
+        from crewai.utilities.printer import Printer
+
+        before_hooks = get_before_llm_call_hooks()
+        if not before_hooks:
+            return True
+
+        hook_context = LLMCallHookContext(
+            executor=None,
+            messages=messages,
+            llm=self,
+            agent=None,
+            task=None,
+            crew=None,
+        )
+        printer = Printer()
+
+        try:
+            for hook in before_hooks:
+                result = hook(hook_context)
+                if result is False:
+                    printer.print(
+                        content="LLM call blocked by before_llm_call hook",
+                        color="yellow",
+                    )
+                    return False
+        except Exception as e:
+            printer.print(
+                content=f"Error in before_llm_call hook: {e}",
+                color="yellow",
+            )
+
+        return True
+
+    def _invoke_after_llm_call_hooks(
+        self,
+        messages: list[LLMMessage],
+        response: str,
+        from_agent: Agent | None = None,
+    ) -> str:
+        """Invoke after_llm_call hooks for direct LLM calls (no agent context).
+
+        This method should be called by native provider implementations after
+        receiving the LLM response when from_agent is None (direct calls).
+
+        Args:
+            messages: The messages that were sent to the LLM
+            response: The response from the LLM
+            from_agent: The agent that made the call (None for direct calls)
+
+        Returns:
+            The potentially modified response string
+
+        Example:
+            >>> # In a native provider's call() method:
+            >>> if from_agent is None and isinstance(result, str):
+            ...     result = self._invoke_after_llm_call_hooks(
+            ...         messages, result, from_agent
+            ...     )
+        """
+        # Only invoke hooks for direct calls (no agent context)
+        if from_agent is not None or not isinstance(response, str):
+            return response
+
+        from crewai.hooks.llm_hooks import (
+            LLMCallHookContext,
+            get_after_llm_call_hooks,
+        )
+        from crewai.utilities.printer import Printer
+
+        after_hooks = get_after_llm_call_hooks()
+        if not after_hooks:
+            return response
+
+        hook_context = LLMCallHookContext(
+            executor=None,
+            messages=messages,
+            llm=self,
+            agent=None,
+            task=None,
+            crew=None,
+            response=response,
+        )
+        printer = Printer()
+        modified_response = response
+
+        try:
+            for hook in after_hooks:
+                result = hook(hook_context)
+                if result is not None and isinstance(result, str):
+                    modified_response = result
+                    hook_context.response = modified_response
+        except Exception as e:
+            printer.print(
+                content=f"Error in after_llm_call hook: {e}",
+                color="yellow",
+            )
+
+        return modified_response

lib/crewai/src/crewai/llms/constants.py

@@ -256,6 +256,7 @@ GeminiModels: TypeAlias = Literal[
     "gemini-2.5-flash-preview-tts",
     "gemini-2.5-pro-preview-tts",
     "gemini-2.5-computer-use-preview-10-2025",
+    "gemini-2.5-pro-exp-03-25",
     "gemini-2.0-flash",
     "gemini-2.0-flash-001",
     "gemini-2.0-flash-exp",
@@ -309,6 +310,7 @@ GEMINI_MODELS: list[GeminiModels] = [
     "gemini-2.5-flash-preview-tts",
     "gemini-2.5-pro-preview-tts",
     "gemini-2.5-computer-use-preview-10-2025",
+    "gemini-2.5-pro-exp-03-25",
     "gemini-2.0-flash",
     "gemini-2.0-flash-001",
     "gemini-2.0-flash-exp",

lib/crewai/src/crewai/llms/providers/anthropic/completion.py

@@ -3,13 +3,14 @@ from __future__ import annotations
 import json
 import logging
 import os
-from typing import TYPE_CHECKING, Any, cast
+from typing import TYPE_CHECKING, Any, Literal, cast
 
+from anthropic.types import ThinkingBlock
 from pydantic import BaseModel
 
 from crewai.events.types.llm_events import LLMCallType
 from crewai.llms.base_llm import BaseLLM
-from crewai.llms.hooks.transport import HTTPTransport
+from crewai.llms.hooks.transport import AsyncHTTPTransport, HTTPTransport
 from crewai.utilities.agent_utils import is_context_length_exceeded
 from crewai.utilities.exceptions.context_window_exceeding_exception import (
     LLMContextLengthExceededError,
@@ -21,9 +22,8 @@ if TYPE_CHECKING:
     from crewai.llms.hooks.base import BaseInterceptor
 
 try:
-    from anthropic import Anthropic
-    from anthropic.types import Message
-    from anthropic.types.tool_use_block import ToolUseBlock
+    from anthropic import Anthropic, AsyncAnthropic
+    from anthropic.types import Message, TextBlock, ThinkingBlock, ToolUseBlock
     import httpx
 except ImportError:
     raise ImportError(
@@ -31,6 +31,11 @@ except ImportError:
     ) from None
 
 
+class AnthropicThinkingConfig(BaseModel):
+    type: Literal["enabled", "disabled"]
+    budget_tokens: int | None = None
+
+
 class AnthropicCompletion(BaseLLM):
     """Anthropic native completion implementation.
 
@@ -52,6 +57,7 @@ class AnthropicCompletion(BaseLLM):
         stream: bool = False,
         client_params: dict[str, Any] | None = None,
         interceptor: BaseInterceptor[httpx.Request, httpx.Response] | None = None,
+        thinking: AnthropicThinkingConfig | None = None,
         **kwargs: Any,
     ):
         """Initialize Anthropic chat completion client.
@@ -84,15 +90,24 @@ class AnthropicCompletion(BaseLLM):
 
         self.client = Anthropic(**self._get_client_params())
 
+        async_client_params = self._get_client_params()
+        if self.interceptor:
+            async_transport = AsyncHTTPTransport(interceptor=self.interceptor)
+            async_http_client = httpx.AsyncClient(transport=async_transport)
+            async_client_params["http_client"] = async_http_client
+
+        self.async_client = AsyncAnthropic(**async_client_params)
+
         # Store completion parameters
         self.max_tokens = max_tokens
         self.top_p = top_p
         self.stream = stream
         self.stop_sequences = stop_sequences or []
-
+        self.thinking = thinking
+        self.previous_thinking_blocks: list[ThinkingBlock] = []
         # Model-specific settings
         self.is_claude_3 = "claude-3" in model.lower()
-        self.supports_tools = self.is_claude_3  # Claude 3+ supports tool use
+        self.supports_tools = True
 
     @property
     def stop(self) -> list[str]:
@@ -182,6 +197,9 @@ class AnthropicCompletion(BaseLLM):
                 messages
             )
 
+            if not self._invoke_before_llm_call_hooks(formatted_messages, from_agent):
+                raise ValueError("LLM call blocked by before_llm_call hook")
+
             # Prepare completion parameters
             completion_params = self._prepare_completion_params(
                 formatted_messages, system_message, tools
@@ -213,6 +231,72 @@ class AnthropicCompletion(BaseLLM):
             )
             raise
 
+    async def acall(
+        self,
+        messages: str | list[LLMMessage],
+        tools: list[dict[str, Any]] | None = None,
+        callbacks: list[Any] | None = None,
+        available_functions: dict[str, Any] | None = None,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> str | Any:
+        """Async call to Anthropic messages API.
+
+        Args:
+            messages: Input messages for the chat completion
+            tools: List of tool/function definitions
+            callbacks: Callback functions (not used in native implementation)
+            available_functions: Available functions for tool calling
+            from_task: Task that initiated the call
+            from_agent: Agent that initiated the call
+
+        Returns:
+            Chat completion response or tool call result
+        """
+        try:
+            self._emit_call_started_event(
+                messages=messages,
+                tools=tools,
+                callbacks=callbacks,
+                available_functions=available_functions,
+                from_task=from_task,
+                from_agent=from_agent,
+            )
+
+            formatted_messages, system_message = self._format_messages_for_anthropic(
+                messages
+            )
+
+            completion_params = self._prepare_completion_params(
+                formatted_messages, system_message, tools
+            )
+
+            if self.stream:
+                return await self._ahandle_streaming_completion(
+                    completion_params,
+                    available_functions,
+                    from_task,
+                    from_agent,
+                    response_model,
+                )
+
+            return await self._ahandle_completion(
+                completion_params,
+                available_functions,
+                from_task,
+                from_agent,
+                response_model,
+            )
+
+        except Exception as e:
+            error_msg = f"Anthropic API call failed: {e!s}"
+            logging.error(error_msg)
+            self._emit_call_failed_event(
+                error=error_msg, from_task=from_task, from_agent=from_agent
+            )
+            raise
+
     def _prepare_completion_params(
         self,
         messages: list[LLMMessage],
@@ -252,6 +336,12 @@ class AnthropicCompletion(BaseLLM):
         if tools and self.supports_tools:
             params["tools"] = self._convert_tools_for_interference(tools)
 
+        if self.thinking:
+            if isinstance(self.thinking, AnthropicThinkingConfig):
+                params["thinking"] = self.thinking.model_dump()
+            else:
+                params["thinking"] = self.thinking
+
         return params
 
     def _convert_tools_for_interference(
@@ -291,6 +381,34 @@ class AnthropicCompletion(BaseLLM):
 
         return anthropic_tools
 
+    def _extract_thinking_block(
+        self, content_block: Any
+    ) -> ThinkingBlock | dict[str, Any] | None:
+        """Extract and format thinking block from content block.
+
+        Args:
+            content_block: Content block from Anthropic response
+
+        Returns:
+            Dictionary with thinking block data including signature, or None if not a thinking block
+        """
+        if content_block.type == "thinking":
+            thinking_block = {
+                "type": "thinking",
+                "thinking": content_block.thinking,
+            }
+            if hasattr(content_block, "signature"):
+                thinking_block["signature"] = content_block.signature
+            return thinking_block
+        if content_block.type == "redacted_thinking":
+            redacted_block = {"type": "redacted_thinking"}
+            if hasattr(content_block, "thinking"):
+                redacted_block["thinking"] = content_block.thinking
+            if hasattr(content_block, "signature"):
+                redacted_block["signature"] = content_block.signature
+            return redacted_block
+        return None
+
     def _format_messages_for_anthropic(
         self, messages: str | list[LLMMessage]
     ) -> tuple[list[LLMMessage], str | None]:
@@ -300,6 +418,7 @@ class AnthropicCompletion(BaseLLM):
         - System messages are separate from conversation messages
         - Messages must alternate between user and assistant
         - First message must be from user
+        - When thinking is enabled, assistant messages must start with thinking blocks
 
         Args:
             messages: Input messages
@@ -324,8 +443,29 @@ class AnthropicCompletion(BaseLLM):
                     system_message = cast(str, content)
             else:
                 role_str = role if role is not None else "user"
-                content_str = content if content is not None else ""
-                formatted_messages.append({"role": role_str, "content": content_str})
+
+                if isinstance(content, list):
+                    formatted_messages.append({"role": role_str, "content": content})
+                elif (
+                    role_str == "assistant"
+                    and self.thinking
+                    and self.previous_thinking_blocks
+                ):
+                    structured_content = cast(
+                        list[dict[str, Any]],
+                        [
+                            *self.previous_thinking_blocks,
+                            {"type": "text", "text": content if content else ""},
+                        ],
+                    )
+                    formatted_messages.append(
+                        LLMMessage(role=role_str, content=structured_content)
+                    )
+                else:
+                    content_str = content if content is not None else ""
+                    formatted_messages.append(
+                        LLMMessage(role=role_str, content=content_str)
+                    )
 
         # Ensure first message is from user (Anthropic requirement)
         if not formatted_messages:
@@ -375,7 +515,6 @@ class AnthropicCompletion(BaseLLM):
             if tool_uses and tool_uses[0].name == "structured_output":
                 structured_data = tool_uses[0].input
                 structured_json = json.dumps(structured_data)
-
                 self._emit_call_completed_event(
                     response=structured_json,
                     call_type=LLMCallType.LLM_CALL,
@@ -403,15 +542,22 @@ class AnthropicCompletion(BaseLLM):
                     from_agent,
                 )
 
-        # Extract text content
         content = ""
+        thinking_blocks: list[ThinkingBlock] = []
+
         if response.content:
             for content_block in response.content:
                 if hasattr(content_block, "text"):
                     content += content_block.text
+                else:
+                    thinking_block = self._extract_thinking_block(content_block)
+                    if thinking_block:
+                        thinking_blocks.append(cast(ThinkingBlock, thinking_block))
+
+        if thinking_blocks:
+            self.previous_thinking_blocks = thinking_blocks
 
         content = self._apply_stop_words(content)
-
         self._emit_call_completed_event(
             response=content,
             call_type=LLMCallType.LLM_CALL,
@@ -423,7 +569,9 @@ class AnthropicCompletion(BaseLLM):
         if usage.get("total_tokens", 0) > 0:
             logging.info(f"Anthropic API usage: {usage}")
 
-        return content
+        return self._invoke_after_llm_call_hooks(
+            params["messages"], content, from_agent
+        )
 
     def _handle_streaming_completion(
         self,
@@ -464,6 +612,16 @@ class AnthropicCompletion(BaseLLM):
 
             final_message: Message = stream.get_final_message()
 
+        thinking_blocks: list[ThinkingBlock] = []
+        if final_message.content:
+            for content_block in final_message.content:
+                thinking_block = self._extract_thinking_block(content_block)
+                if thinking_block:
+                    thinking_blocks.append(cast(ThinkingBlock, thinking_block))
+
+        if thinking_blocks:
+            self.previous_thinking_blocks = thinking_blocks
+
         usage = self._extract_anthropic_token_usage(final_message)
         self._track_token_usage_internal(usage)
 
@@ -517,7 +675,52 @@ class AnthropicCompletion(BaseLLM):
             messages=params["messages"],
         )
 
-        return full_response
+        return self._invoke_after_llm_call_hooks(
+            params["messages"], full_response, from_agent
+        )
+
+    def _execute_tools_and_collect_results(
+        self,
+        tool_uses: list[ToolUseBlock],
+        available_functions: dict[str, Any],
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+    ) -> list[dict[str, Any]]:
+        """Execute tools and collect results in Anthropic format.
+
+        Args:
+            tool_uses: List of tool use blocks from Claude's response
+            available_functions: Available functions for tool calling
+            from_task: Task that initiated the call
+            from_agent: Agent that initiated the call
+
+        Returns:
+            List of tool result dictionaries in Anthropic format
+        """
+        tool_results = []
+
+        for tool_use in tool_uses:
+            function_name = tool_use.name
+            function_args = tool_use.input
+
+            result = self._handle_tool_execution(
+                function_name=function_name,
+                function_args=cast(dict[str, Any], function_args),
+                available_functions=available_functions,
+                from_task=from_task,
+                from_agent=from_agent,
+            )
+
+            tool_result = {
+                "type": "tool_result",
+                "tool_use_id": tool_use.id,
+                "content": str(result)
+                if result is not None
+                else "Tool execution completed",
+            }
+            tool_results.append(tool_result)
+
+        return tool_results
 
     def _handle_tool_use_conversation(
         self,
@@ -536,37 +739,33 @@ class AnthropicCompletion(BaseLLM):
         3. We send tool results back to Claude
         4. Claude processes results and generates final response
         """
-        # Execute all requested tools and collect results
-        tool_results = []
+        tool_results = self._execute_tools_and_collect_results(
+            tool_uses, available_functions, from_task, from_agent
+        )
 
-        for tool_use in tool_uses:
-            function_name = tool_use.name
-            function_args = tool_use.input
-
-            # Execute the tool
-            result = self._handle_tool_execution(
-                function_name=function_name,
-                function_args=function_args,  # type: ignore
-                available_functions=available_functions,
-                from_task=from_task,
-                from_agent=from_agent,
-            )
-
-            # Create tool result in Anthropic format
-            tool_result = {
-                "type": "tool_result",
-                "tool_use_id": tool_use.id,
-                "content": str(result)
-                if result is not None
-                else "Tool execution completed",
-            }
-            tool_results.append(tool_result)
-
-        # Prepare follow-up conversation with tool results
         follow_up_params = params.copy()
 
         # Add Claude's tool use response to conversation
-        assistant_message = {"role": "assistant", "content": initial_response.content}
+        assistant_content: list[
+            ThinkingBlock | ToolUseBlock | TextBlock | dict[str, Any]
+        ] = []
+        for block in initial_response.content:
+            thinking_block = self._extract_thinking_block(block)
+            if thinking_block:
+                assistant_content.append(thinking_block)
+            elif block.type == "tool_use":
+                assistant_content.append(
+                    {
+                        "type": "tool_use",
+                        "id": block.id,
+                        "name": block.name,
+                        "input": block.input,
+                    }
+                )
+            elif hasattr(block, "text"):
+                assistant_content.append({"type": "text", "text": block.text})
+
+        assistant_message = {"role": "assistant", "content": assistant_content}
 
         # Add user message with tool results
         user_message = {"role": "user", "content": tool_results}
@@ -585,12 +784,20 @@ class AnthropicCompletion(BaseLLM):
             follow_up_usage = self._extract_anthropic_token_usage(final_response)
             self._track_token_usage_internal(follow_up_usage)
 
-            # Extract final text content
             final_content = ""
+            thinking_blocks: list[ThinkingBlock] = []
+
             if final_response.content:
                 for content_block in final_response.content:
                     if hasattr(content_block, "text"):
                         final_content += content_block.text
+                    else:
+                        thinking_block = self._extract_thinking_block(content_block)
+                        if thinking_block:
+                            thinking_blocks.append(cast(ThinkingBlock, thinking_block))
+
+            if thinking_blocks:
+                self.previous_thinking_blocks = thinking_blocks
 
             final_content = self._apply_stop_words(final_content)
 
@@ -623,7 +830,257 @@ class AnthropicCompletion(BaseLLM):
             logging.error(f"Tool follow-up conversation failed: {e}")
             # Fallback: return the first tool result if follow-up fails
             if tool_results:
-                return tool_results[0]["content"]
+                return cast(str, tool_results[0]["content"])
+            raise e
+
+    async def _ahandle_completion(
+        self,
+        params: dict[str, Any],
+        available_functions: dict[str, Any] | None = None,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> str | Any:
+        """Handle non-streaming async message completion."""
+        if response_model:
+            structured_tool = {
+                "name": "structured_output",
+                "description": "Returns structured data according to the schema",
+                "input_schema": response_model.model_json_schema(),
+            }
+
+            params["tools"] = [structured_tool]
+            params["tool_choice"] = {"type": "tool", "name": "structured_output"}
+
+        try:
+            response: Message = await self.async_client.messages.create(**params)
+
+        except Exception as e:
+            if is_context_length_exceeded(e):
+                logging.error(f"Context window exceeded: {e}")
+                raise LLMContextLengthExceededError(str(e)) from e
+            raise e from e
+
+        usage = self._extract_anthropic_token_usage(response)
+        self._track_token_usage_internal(usage)
+
+        if response_model and response.content:
+            tool_uses = [
+                block for block in response.content if isinstance(block, ToolUseBlock)
+            ]
+            if tool_uses and tool_uses[0].name == "structured_output":
+                structured_data = tool_uses[0].input
+                structured_json = json.dumps(structured_data)
+
+                self._emit_call_completed_event(
+                    response=structured_json,
+                    call_type=LLMCallType.LLM_CALL,
+                    from_task=from_task,
+                    from_agent=from_agent,
+                    messages=params["messages"],
+                )
+
+                return structured_json
+
+        if response.content and available_functions:
+            tool_uses = [
+                block for block in response.content if isinstance(block, ToolUseBlock)
+            ]
+
+            if tool_uses:
+                return await self._ahandle_tool_use_conversation(
+                    response,
+                    tool_uses,
+                    params,
+                    available_functions,
+                    from_task,
+                    from_agent,
+                )
+
+        content = ""
+        if response.content:
+            for content_block in response.content:
+                if hasattr(content_block, "text"):
+                    content += content_block.text
+
+        content = self._apply_stop_words(content)
+
+        self._emit_call_completed_event(
+            response=content,
+            call_type=LLMCallType.LLM_CALL,
+            from_task=from_task,
+            from_agent=from_agent,
+            messages=params["messages"],
+        )
+
+        if usage.get("total_tokens", 0) > 0:
+            logging.info(f"Anthropic API usage: {usage}")
+
+        return content
+
+    async def _ahandle_streaming_completion(
+        self,
+        params: dict[str, Any],
+        available_functions: dict[str, Any] | None = None,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> str:
+        """Handle async streaming message completion."""
+        if response_model:
+            structured_tool = {
+                "name": "structured_output",
+                "description": "Returns structured data according to the schema",
+                "input_schema": response_model.model_json_schema(),
+            }
+
+            params["tools"] = [structured_tool]
+            params["tool_choice"] = {"type": "tool", "name": "structured_output"}
+
+        full_response = ""
+
+        stream_params = {k: v for k, v in params.items() if k != "stream"}
+
+        async with self.async_client.messages.stream(**stream_params) as stream:
+            async for event in stream:
+                if hasattr(event, "delta") and hasattr(event.delta, "text"):
+                    text_delta = event.delta.text
+                    full_response += text_delta
+                    self._emit_stream_chunk_event(
+                        chunk=text_delta,
+                        from_task=from_task,
+                        from_agent=from_agent,
+                    )
+
+            final_message: Message = await stream.get_final_message()
+
+        usage = self._extract_anthropic_token_usage(final_message)
+        self._track_token_usage_internal(usage)
+
+        if response_model and final_message.content:
+            tool_uses = [
+                block
+                for block in final_message.content
+                if isinstance(block, ToolUseBlock)
+            ]
+            if tool_uses and tool_uses[0].name == "structured_output":
+                structured_data = tool_uses[0].input
+                structured_json = json.dumps(structured_data)
+
+                self._emit_call_completed_event(
+                    response=structured_json,
+                    call_type=LLMCallType.LLM_CALL,
+                    from_task=from_task,
+                    from_agent=from_agent,
+                    messages=params["messages"],
+                )
+
+                return structured_json
+
+        if final_message.content and available_functions:
+            tool_uses = [
+                block
+                for block in final_message.content
+                if isinstance(block, ToolUseBlock)
+            ]
+
+            if tool_uses:
+                return await self._ahandle_tool_use_conversation(
+                    final_message,
+                    tool_uses,
+                    params,
+                    available_functions,
+                    from_task,
+                    from_agent,
+                )
+
+        full_response = self._apply_stop_words(full_response)
+
+        self._emit_call_completed_event(
+            response=full_response,
+            call_type=LLMCallType.LLM_CALL,
+            from_task=from_task,
+            from_agent=from_agent,
+            messages=params["messages"],
+        )
+
+        return full_response
+
+    async def _ahandle_tool_use_conversation(
+        self,
+        initial_response: Message,
+        tool_uses: list[ToolUseBlock],
+        params: dict[str, Any],
+        available_functions: dict[str, Any],
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+    ) -> str:
+        """Handle the complete async tool use conversation flow.
+
+        This implements the proper Anthropic tool use pattern:
+        1. Claude requests tool use
+        2. We execute the tools
+        3. We send tool results back to Claude
+        4. Claude processes results and generates final response
+        """
+        tool_results = self._execute_tools_and_collect_results(
+            tool_uses, available_functions, from_task, from_agent
+        )
+
+        follow_up_params = params.copy()
+
+        assistant_message = {"role": "assistant", "content": initial_response.content}
+
+        user_message = {"role": "user", "content": tool_results}
+
+        follow_up_params["messages"] = params["messages"] + [
+            assistant_message,
+            user_message,
+        ]
+
+        try:
+            final_response: Message = await self.async_client.messages.create(
+                **follow_up_params
+            )
+
+            follow_up_usage = self._extract_anthropic_token_usage(final_response)
+            self._track_token_usage_internal(follow_up_usage)
+
+            final_content = ""
+            if final_response.content:
+                for content_block in final_response.content:
+                    if hasattr(content_block, "text"):
+                        final_content += content_block.text
+
+            final_content = self._apply_stop_words(final_content)
+
+            self._emit_call_completed_event(
+                response=final_content,
+                call_type=LLMCallType.LLM_CALL,
+                from_task=from_task,
+                from_agent=from_agent,
+                messages=follow_up_params["messages"],
+            )
+
+            total_usage = {
+                "input_tokens": follow_up_usage.get("input_tokens", 0),
+                "output_tokens": follow_up_usage.get("output_tokens", 0),
+                "total_tokens": follow_up_usage.get("total_tokens", 0),
+            }
+
+            if total_usage.get("total_tokens", 0) > 0:
+                logging.info(f"Anthropic API tool conversation usage: {total_usage}")
+
+            return final_content
+
+        except Exception as e:
+            if is_context_length_exceeded(e):
+                logging.error(f"Context window exceeded in tool follow-up: {e}")
+                raise LLMContextLengthExceededError(str(e)) from e
+
+            logging.error(f"Tool follow-up conversation failed: {e}")
+            if tool_results:
+                return cast(str, tool_results[0]["content"])
             raise e
 
     def supports_function_calling(self) -> bool:
@@ -659,7 +1116,8 @@ class AnthropicCompletion(BaseLLM):
         # Default context window size for Claude models
         return int(200000 * CONTEXT_WINDOW_USAGE_RATIO)
 
-    def _extract_anthropic_token_usage(self, response: Message) -> dict[str, Any]:
+    @staticmethod
+    def _extract_anthropic_token_usage(response: Message) -> dict[str, Any]:
         """Extract token usage from Anthropic response."""
         if hasattr(response, "usage") and response.usage:
             usage = response.usage

lib/crewai/src/crewai/llms/providers/azure/completion.py

@@ -3,30 +3,35 @@ from __future__ import annotations
 import json
 import logging
 import os
-from typing import TYPE_CHECKING, Any
+from typing import TYPE_CHECKING, Any, TypedDict
 
 from pydantic import BaseModel
+from typing_extensions import Self
 
 from crewai.utilities.agent_utils import is_context_length_exceeded
-from crewai.utilities.converter import generate_model_description
 from crewai.utilities.exceptions.context_window_exceeding_exception import (
     LLMContextLengthExceededError,
 )
+from crewai.utilities.pydantic_schema_utils import generate_model_description
 from crewai.utilities.types import LLMMessage
 
 
 if TYPE_CHECKING:
     from crewai.llms.hooks.base import BaseInterceptor
-    from crewai.tools.base_tool import BaseTool
 
 
 try:
     from azure.ai.inference import (
         ChatCompletionsClient,
     )
+    from azure.ai.inference.aio import (
+        ChatCompletionsClient as AsyncChatCompletionsClient,
+    )
     from azure.ai.inference.models import (
         ChatCompletions,
         ChatCompletionsToolCall,
+        ChatCompletionsToolDefinition,
+        FunctionDefinition,
         JsonSchemaFormat,
         StreamingChatCompletionsUpdate,
     )
@@ -46,6 +51,24 @@ except ImportError:
     ) from None
 
 
+class AzureCompletionParams(TypedDict, total=False):
+    """Type definition for Azure chat completion parameters."""
+
+    messages: list[LLMMessage]
+    stream: bool
+    model_extras: dict[str, Any]
+    response_format: JsonSchemaFormat
+    model: str
+    temperature: float
+    top_p: float
+    frequency_penalty: float
+    presence_penalty: float
+    max_tokens: int
+    stop: list[str]
+    tools: list[ChatCompletionsToolDefinition]
+    tool_choice: str
+
+
 class AzureCompletion(BaseLLM):
     """Azure AI Inference native completion implementation.
 
@@ -135,6 +158,8 @@ class AzureCompletion(BaseLLM):
 
         self.client = ChatCompletionsClient(**client_kwargs)  # type: ignore[arg-type]
 
+        self.async_client = AsyncChatCompletionsClient(**client_kwargs)  # type: ignore[arg-type]
+
         self.top_p = top_p
         self.frequency_penalty = frequency_penalty
         self.presence_penalty = presence_penalty
@@ -150,7 +175,8 @@ class AzureCompletion(BaseLLM):
             and "/openai/deployments/" in self.endpoint
         )
 
-    def _validate_and_fix_endpoint(self, endpoint: str, model: str) -> str:
+    @staticmethod
+    def _validate_and_fix_endpoint(endpoint: str, model: str) -> str:
         """Validate and fix Azure endpoint URL format.
 
         Azure OpenAI endpoints should be in the format:
@@ -173,10 +199,75 @@ class AzureCompletion(BaseLLM):
 
         return endpoint
 
+    def _handle_api_error(
+        self,
+        error: Exception,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+    ) -> None:
+        """Handle API errors with appropriate logging and events.
+
+        Args:
+            error: The exception that occurred
+            from_task: Task that initiated the call
+            from_agent: Agent that initiated the call
+
+        Raises:
+            The original exception after logging and emitting events
+        """
+        if isinstance(error, HttpResponseError):
+            if error.status_code == 401:
+                error_msg = "Azure authentication failed. Check your API key."
+            elif error.status_code == 404:
+                error_msg = (
+                    f"Azure endpoint not found. Check endpoint URL: {self.endpoint}"
+                )
+            elif error.status_code == 429:
+                error_msg = "Azure API rate limit exceeded. Please retry later."
+            else:
+                error_msg = (
+                    f"Azure API HTTP error: {error.status_code} - {error.message}"
+                )
+        else:
+            error_msg = f"Azure API call failed: {error!s}"
+
+        logging.error(error_msg)
+        self._emit_call_failed_event(
+            error=error_msg, from_task=from_task, from_agent=from_agent
+        )
+        raise error
+
+    def _handle_completion_error(
+        self,
+        error: Exception,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+    ) -> None:
+        """Handle completion-specific errors including context length checks.
+
+        Args:
+            error: The exception that occurred
+            from_task: Task that initiated the call
+            from_agent: Agent that initiated the call
+
+        Raises:
+            LLMContextLengthExceededError if context window exceeded, otherwise the original exception
+        """
+        if is_context_length_exceeded(error):
+            logging.error(f"Context window exceeded: {error}")
+            raise LLMContextLengthExceededError(str(error)) from error
+
+        error_msg = f"Azure API call failed: {error!s}"
+        logging.error(error_msg)
+        self._emit_call_failed_event(
+            error=error_msg, from_task=from_task, from_agent=from_agent
+        )
+        raise error
+
     def call(
         self,
         messages: str | list[LLMMessage],
-        tools: list[dict[str, BaseTool]] | None = None,
+        tools: list[dict[str, Any]] | None = None,
         callbacks: list[Any] | None = None,
         available_functions: dict[str, Any] | None = None,
         from_task: Any | None = None,
@@ -192,6 +283,7 @@ class AzureCompletion(BaseLLM):
             available_functions: Available functions for tool calling
             from_task: Task that initiated the call
             from_agent: Agent that initiated the call
+            response_model: Response model
 
         Returns:
             Chat completion response or tool call result
@@ -210,6 +302,9 @@ class AzureCompletion(BaseLLM):
             # Format messages for Azure
             formatted_messages = self._format_messages_for_azure(messages)
 
+            if not self._invoke_before_llm_call_hooks(formatted_messages, from_agent):
+                raise ValueError("LLM call blocked by before_llm_call hook")
+
             # Prepare completion parameters
             completion_params = self._prepare_completion_params(
                 formatted_messages, tools, response_model
@@ -233,37 +328,75 @@ class AzureCompletion(BaseLLM):
                 response_model,
             )
 
-        except HttpResponseError as e:
-            if e.status_code == 401:
-                error_msg = "Azure authentication failed. Check your API key."
-            elif e.status_code == 404:
-                error_msg = (
-                    f"Azure endpoint not found. Check endpoint URL: {self.endpoint}"
-                )
-            elif e.status_code == 429:
-                error_msg = "Azure API rate limit exceeded. Please retry later."
-            else:
-                error_msg = f"Azure API HTTP error: {e.status_code} - {e.message}"
-
-            logging.error(error_msg)
-            self._emit_call_failed_event(
-                error=error_msg, from_task=from_task, from_agent=from_agent
-            )
-            raise
         except Exception as e:
-            error_msg = f"Azure API call failed: {e!s}"
-            logging.error(error_msg)
-            self._emit_call_failed_event(
-                error=error_msg, from_task=from_task, from_agent=from_agent
+            return self._handle_api_error(e, from_task, from_agent)  # type: ignore[func-returns-value]
+
+    async def acall(  # type: ignore[return]
+        self,
+        messages: str | list[LLMMessage],
+        tools: list[dict[str, Any]] | None = None,
+        callbacks: list[Any] | None = None,
+        available_functions: dict[str, Any] | None = None,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> str | Any:
+        """Call Azure AI Inference chat completions API asynchronously.
+
+        Args:
+            messages: Input messages for the chat completion
+            tools: List of tool/function definitions
+            callbacks: Callback functions (not used in native implementation)
+            available_functions: Available functions for tool calling
+            from_task: Task that initiated the call
+            from_agent: Agent that initiated the call
+            response_model: Pydantic model for structured output
+
+        Returns:
+            Chat completion response or tool call result
+        """
+        try:
+            self._emit_call_started_event(
+                messages=messages,
+                tools=tools,
+                callbacks=callbacks,
+                available_functions=available_functions,
+                from_task=from_task,
+                from_agent=from_agent,
             )
-            raise
+
+            formatted_messages = self._format_messages_for_azure(messages)
+
+            completion_params = self._prepare_completion_params(
+                formatted_messages, tools, response_model
+            )
+
+            if self.stream:
+                return await self._ahandle_streaming_completion(
+                    completion_params,
+                    available_functions,
+                    from_task,
+                    from_agent,
+                    response_model,
+                )
+
+            return await self._ahandle_completion(
+                completion_params,
+                available_functions,
+                from_task,
+                from_agent,
+                response_model,
+            )
+
+        except Exception as e:
+            self._handle_api_error(e, from_task, from_agent)
 
     def _prepare_completion_params(
         self,
         messages: list[LLMMessage],
         tools: list[dict[str, Any]] | None = None,
         response_model: type[BaseModel] | None = None,
-    ) -> dict[str, Any]:
+    ) -> AzureCompletionParams:
         """Prepare parameters for Azure AI Inference chat completion.
 
         Args:
@@ -274,11 +407,14 @@ class AzureCompletion(BaseLLM):
         Returns:
             Parameters dictionary for Azure API
         """
-        params = {
+        params: AzureCompletionParams = {
             "messages": messages,
             "stream": self.stream,
         }
 
+        if self.stream:
+            params["model_extras"] = {"stream_options": {"include_usage": True}}
+
         if response_model and self.is_openai_model:
             model_description = generate_model_description(response_model)
             json_schema_info = model_description["json_schema"]
@@ -321,37 +457,42 @@ class AzureCompletion(BaseLLM):
 
         if drop_params and isinstance(additional_drop_params, list):
             for drop_param in additional_drop_params:
-                params.pop(drop_param, None)
+                if isinstance(drop_param, str):
+                    params.pop(drop_param, None)  # type: ignore[misc]
 
         return params
 
-    def _convert_tools_for_interference(
+    def _convert_tools_for_interference(  # type: ignore[override]
         self, tools: list[dict[str, Any]]
-    ) -> list[dict[str, Any]]:
-        """Convert CrewAI tool format to Azure OpenAI function calling format."""
+    ) -> list[ChatCompletionsToolDefinition]:
+        """Convert CrewAI tool format to Azure OpenAI function calling format.
 
+        Args:
+            tools: List of CrewAI tool definitions
+
+        Returns:
+            List of Azure ChatCompletionsToolDefinition objects
+        """
         from crewai.llms.providers.utils.common import safe_tool_conversion
 
-        azure_tools = []
+        azure_tools: list[ChatCompletionsToolDefinition] = []
 
         for tool in tools:
             name, description, parameters = safe_tool_conversion(tool, "Azure")
 
-            azure_tool = {
-                "type": "function",
-                "function": {
-                    "name": name,
-                    "description": description,
-                },
-            }
+            function_def = FunctionDefinition(
+                name=name,
+                description=description,
+                parameters=parameters
+                if isinstance(parameters, dict)
+                else dict(parameters)
+                if parameters
+                else None,
+            )
 
-            if parameters:
-                if isinstance(parameters, dict):
-                    azure_tool["function"]["parameters"] = parameters  # type: ignore
-                else:
-                    azure_tool["function"]["parameters"] = dict(parameters)
+            tool_def = ChatCompletionsToolDefinition(function=function_def)
 
-            azure_tools.append(azure_tool)
+            azure_tools.append(tool_def)
 
         return azure_tools
 
@@ -380,144 +521,239 @@ class AzureCompletion(BaseLLM):
 
         return azure_messages
 
-    def _handle_completion(
+    def _validate_and_emit_structured_output(
         self,
-        params: dict[str, Any],
-        available_functions: dict[str, Any] | None = None,
+        content: str,
+        response_model: type[BaseModel],
+        params: AzureCompletionParams,
         from_task: Any | None = None,
         from_agent: Any | None = None,
-        response_model: type[BaseModel] | None = None,
-    ) -> str | Any:
-        """Handle non-streaming chat completion."""
-        # Make API call
+    ) -> str:
+        """Validate content against response model and emit completion event.
+
+        Args:
+            content: Response content to validate
+            response_model: Pydantic model for validation
+            params: Completion parameters containing messages
+            from_task: Task that initiated the call
+            from_agent: Agent that initiated the call
+
+        Returns:
+            Validated and serialized JSON string
+
+        Raises:
+            ValueError: If validation fails
+        """
         try:
-            response: ChatCompletions = self.client.complete(**params)
+            structured_data = response_model.model_validate_json(content)
+            structured_json = structured_data.model_dump_json()
 
-            if not response.choices:
-                raise ValueError("No choices returned from Azure API")
-
-            choice = response.choices[0]
-            message = choice.message
-
-            # Extract and track token usage
-            usage = self._extract_azure_token_usage(response)
-            self._track_token_usage_internal(usage)
-
-            if response_model and self.is_openai_model:
-                content = message.content or ""
-                try:
-                    structured_data = response_model.model_validate_json(content)
-                    structured_json = structured_data.model_dump_json()
-
-                    self._emit_call_completed_event(
-                        response=structured_json,
-                        call_type=LLMCallType.LLM_CALL,
-                        from_task=from_task,
-                        from_agent=from_agent,
-                        messages=params["messages"],
-                    )
-
-                    return structured_json
-                except Exception as e:
-                    error_msg = f"Failed to validate structured output with model {response_model.__name__}: {e}"
-                    logging.error(error_msg)
-                    raise ValueError(error_msg) from e
-
-            # Handle tool calls
-            if message.tool_calls and available_functions:
-                tool_call = message.tool_calls[0]  # Handle first tool call
-                if isinstance(tool_call, ChatCompletionsToolCall):
-                    function_name = tool_call.function.name
-
-                    try:
-                        function_args = json.loads(tool_call.function.arguments)
-                    except json.JSONDecodeError as e:
-                        logging.error(f"Failed to parse tool arguments: {e}")
-                        function_args = {}
-
-                    # Execute tool
-                    result = self._handle_tool_execution(
-                        function_name=function_name,
-                        function_args=function_args,
-                        available_functions=available_functions,
-                        from_task=from_task,
-                        from_agent=from_agent,
-                    )
-
-                    if result is not None:
-                        return result
-
-            # Extract content
-            content = message.content or ""
-
-            # Apply stop words
-            content = self._apply_stop_words(content)
-
-            # Emit completion event and return content
             self._emit_call_completed_event(
-                response=content,
+                response=structured_json,
                 call_type=LLMCallType.LLM_CALL,
                 from_task=from_task,
                 from_agent=from_agent,
                 messages=params["messages"],
             )
 
+            return structured_json
         except Exception as e:
-            if is_context_length_exceeded(e):
-                logging.error(f"Context window exceeded: {e}")
-                raise LLMContextLengthExceededError(str(e)) from e
-
-            error_msg = f"Azure API call failed: {e!s}"
+            error_msg = f"Failed to validate structured output with model {response_model.__name__}: {e}"
             logging.error(error_msg)
-            self._emit_call_failed_event(
-                error=error_msg, from_task=from_task, from_agent=from_agent
-            )
-            raise e
+            raise ValueError(error_msg) from e
 
-        return content
-
-    def _handle_streaming_completion(
+    def _process_completion_response(
         self,
-        params: dict[str, Any],
+        response: ChatCompletions,
+        params: AzureCompletionParams,
         available_functions: dict[str, Any] | None = None,
         from_task: Any | None = None,
         from_agent: Any | None = None,
         response_model: type[BaseModel] | None = None,
+    ) -> str | Any:
+        """Process completion response with usage tracking, tool execution, and events.
+
+        Args:
+            response: Chat completion response from Azure API
+            params: Completion parameters containing messages
+            available_functions: Available functions for tool calling
+            from_task: Task that initiated the call
+            from_agent: Agent that initiated the call
+            response_model: Pydantic model for structured output
+
+        Returns:
+            Response content or structured output
+        """
+        if not response.choices:
+            raise ValueError("No choices returned from Azure API")
+
+        choice = response.choices[0]
+        message = choice.message
+
+        # Extract and track token usage
+        usage = self._extract_azure_token_usage(response)
+        self._track_token_usage_internal(usage)
+
+        if response_model and self.is_openai_model:
+            content = message.content or ""
+            return self._validate_and_emit_structured_output(
+                content=content,
+                response_model=response_model,
+                params=params,
+                from_task=from_task,
+                from_agent=from_agent,
+            )
+
+        # Handle tool calls
+        if message.tool_calls and available_functions:
+            tool_call = message.tool_calls[0]  # Handle first tool call
+            if isinstance(tool_call, ChatCompletionsToolCall):
+                function_name = tool_call.function.name
+
+                try:
+                    function_args = json.loads(tool_call.function.arguments)
+                except json.JSONDecodeError as e:
+                    logging.error(f"Failed to parse tool arguments: {e}")
+                    function_args = {}
+
+                # Execute tool
+                result = self._handle_tool_execution(
+                    function_name=function_name,
+                    function_args=function_args,
+                    available_functions=available_functions,
+                    from_task=from_task,
+                    from_agent=from_agent,
+                )
+
+                if result is not None:
+                    return result
+
+        # Extract content
+        content = message.content or ""
+
+        # Apply stop words
+        content = self._apply_stop_words(content)
+
+        # Emit completion event and return content
+        self._emit_call_completed_event(
+            response=content,
+            call_type=LLMCallType.LLM_CALL,
+            from_task=from_task,
+            from_agent=from_agent,
+            messages=params["messages"],
+        )
+
+        return self._invoke_after_llm_call_hooks(
+            params["messages"], content, from_agent
+        )
+
+    def _handle_completion(
+        self,
+        params: AzureCompletionParams,
+        available_functions: dict[str, Any] | None = None,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> str | Any:
+        """Handle non-streaming chat completion."""
+        try:
+            # Cast params to Any to avoid type checking issues with TypedDict unpacking
+            response: ChatCompletions = self.client.complete(**params)  # type: ignore[assignment,arg-type]
+            return self._process_completion_response(
+                response=response,
+                params=params,
+                available_functions=available_functions,
+                from_task=from_task,
+                from_agent=from_agent,
+                response_model=response_model,
+            )
+        except Exception as e:
+            return self._handle_completion_error(e, from_task, from_agent)  # type: ignore[func-returns-value]
+
+    def _process_streaming_update(
+        self,
+        update: StreamingChatCompletionsUpdate,
+        full_response: str,
+        tool_calls: dict[str, dict[str, str]],
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
     ) -> str:
-        """Handle streaming chat completion."""
-        full_response = ""
-        tool_calls = {}
+        """Process a single streaming update chunk.
 
-        # Make streaming API call
-        for update in self.client.complete(**params):
-            if isinstance(update, StreamingChatCompletionsUpdate):
-                if update.choices:
-                    choice = update.choices[0]
-                    if choice.delta and choice.delta.content:
-                        content_delta = choice.delta.content
-                        full_response += content_delta
-                        self._emit_stream_chunk_event(
-                            chunk=content_delta,
-                            from_task=from_task,
-                            from_agent=from_agent,
-                        )
+        Args:
+            update: Streaming update from Azure API
+            full_response: Accumulated response content
+            tool_calls: Dictionary of accumulated tool calls
+            from_task: Task that initiated the call
+            from_agent: Agent that initiated the call
 
-                    # Handle tool call streaming
-                    if choice.delta and choice.delta.tool_calls:
-                        for tool_call in choice.delta.tool_calls:
-                            call_id = tool_call.id or "default"
-                            if call_id not in tool_calls:
-                                tool_calls[call_id] = {
-                                    "name": "",
-                                    "arguments": "",
-                                }
+        Returns:
+            Updated full_response string
+        """
+        if update.choices:
+            choice = update.choices[0]
+            if choice.delta and choice.delta.content:
+                content_delta = choice.delta.content
+                full_response += content_delta
+                self._emit_stream_chunk_event(
+                    chunk=content_delta,
+                    from_task=from_task,
+                    from_agent=from_agent,
+                )
 
-                            if tool_call.function and tool_call.function.name:
-                                tool_calls[call_id]["name"] = tool_call.function.name
-                            if tool_call.function and tool_call.function.arguments:
-                                tool_calls[call_id]["arguments"] += (
-                                    tool_call.function.arguments
-                                )
+            if choice.delta and choice.delta.tool_calls:
+                for tool_call in choice.delta.tool_calls:
+                    call_id = tool_call.id or "default"
+                    if call_id not in tool_calls:
+                        tool_calls[call_id] = {
+                            "name": "",
+                            "arguments": "",
+                        }
+
+                    if tool_call.function and tool_call.function.name:
+                        tool_calls[call_id]["name"] = tool_call.function.name
+                    if tool_call.function and tool_call.function.arguments:
+                        tool_calls[call_id]["arguments"] += tool_call.function.arguments
+
+        return full_response
+
+    def _finalize_streaming_response(
+        self,
+        full_response: str,
+        tool_calls: dict[str, dict[str, str]],
+        usage_data: dict[str, int],
+        params: AzureCompletionParams,
+        available_functions: dict[str, Any] | None = None,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> str | Any:
+        """Finalize streaming response with usage tracking, tool execution, and events.
+
+        Args:
+            full_response: The complete streamed response content
+            tool_calls: Dictionary of tool calls accumulated during streaming
+            usage_data: Token usage data from the stream
+            params: Completion parameters containing messages
+            available_functions: Available functions for tool calling
+            from_task: Task that initiated the call
+            from_agent: Agent that initiated the call
+            response_model: Pydantic model for structured output validation
+
+        Returns:
+            Final response content after processing, or structured output
+        """
+        self._track_token_usage_internal(usage_data)
+
+        # Handle structured output validation
+        if response_model and self.is_openai_model:
+            return self._validate_and_emit_structured_output(
+                content=full_response,
+                response_model=response_model,
+                params=params,
+                from_task=from_task,
+                from_agent=from_agent,
+            )
 
         # Handle completed tool calls
         if tool_calls and available_functions:
@@ -554,7 +790,120 @@ class AzureCompletion(BaseLLM):
             messages=params["messages"],
         )
 
-        return full_response
+        return self._invoke_after_llm_call_hooks(
+            params["messages"], full_response, from_agent
+        )
+
+    def _handle_streaming_completion(
+        self,
+        params: AzureCompletionParams,
+        available_functions: dict[str, Any] | None = None,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> str | Any:
+        """Handle streaming chat completion."""
+        full_response = ""
+        tool_calls: dict[str, dict[str, Any]] = {}
+
+        usage_data = {"total_tokens": 0}
+        for update in self.client.complete(**params):  # type: ignore[arg-type]
+            if isinstance(update, StreamingChatCompletionsUpdate):
+                if update.usage:
+                    usage = update.usage
+                    usage_data = {
+                        "prompt_tokens": usage.prompt_tokens,
+                        "completion_tokens": usage.completion_tokens,
+                        "total_tokens": usage.total_tokens,
+                    }
+                    continue
+
+                full_response = self._process_streaming_update(
+                    update=update,
+                    full_response=full_response,
+                    tool_calls=tool_calls,
+                    from_task=from_task,
+                    from_agent=from_agent,
+                )
+
+        return self._finalize_streaming_response(
+            full_response=full_response,
+            tool_calls=tool_calls,
+            usage_data=usage_data,
+            params=params,
+            available_functions=available_functions,
+            from_task=from_task,
+            from_agent=from_agent,
+            response_model=response_model,
+        )
+
+    async def _ahandle_completion(
+        self,
+        params: AzureCompletionParams,
+        available_functions: dict[str, Any] | None = None,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> str | Any:
+        """Handle non-streaming chat completion asynchronously."""
+        try:
+            # Cast params to Any to avoid type checking issues with TypedDict unpacking
+            response: ChatCompletions = await self.async_client.complete(**params)  # type: ignore[assignment,arg-type]
+            return self._process_completion_response(
+                response=response,
+                params=params,
+                available_functions=available_functions,
+                from_task=from_task,
+                from_agent=from_agent,
+                response_model=response_model,
+            )
+        except Exception as e:
+            return self._handle_completion_error(e, from_task, from_agent)  # type: ignore[func-returns-value]
+
+    async def _ahandle_streaming_completion(
+        self,
+        params: AzureCompletionParams,
+        available_functions: dict[str, Any] | None = None,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> str | Any:
+        """Handle streaming chat completion asynchronously."""
+        full_response = ""
+        tool_calls: dict[str, dict[str, Any]] = {}
+
+        usage_data = {"total_tokens": 0}
+
+        stream = await self.async_client.complete(**params)  # type: ignore[arg-type]
+        async for update in stream:  # type: ignore[union-attr]
+            if isinstance(update, StreamingChatCompletionsUpdate):
+                if hasattr(update, "usage") and update.usage:
+                    usage = update.usage
+                    usage_data = {
+                        "prompt_tokens": getattr(usage, "prompt_tokens", 0),
+                        "completion_tokens": getattr(usage, "completion_tokens", 0),
+                        "total_tokens": getattr(usage, "total_tokens", 0),
+                    }
+                    continue
+
+                full_response = self._process_streaming_update(
+                    update=update,
+                    full_response=full_response,
+                    tool_calls=tool_calls,
+                    from_task=from_task,
+                    from_agent=from_agent,
+                )
+
+        return self._finalize_streaming_response(
+            full_response=full_response,
+            tool_calls=tool_calls,
+            usage_data=usage_data,
+            params=params,
+            available_functions=available_functions,
+            from_task=from_task,
+            from_agent=from_agent,
+            response_model=response_model,
+        )
 
     def supports_function_calling(self) -> bool:
         """Check if the model supports function calling."""
@@ -599,7 +948,8 @@ class AzureCompletion(BaseLLM):
         # Default context window size
         return int(8192 * CONTEXT_WINDOW_USAGE_RATIO)
 
-    def _extract_azure_token_usage(self, response: ChatCompletions) -> dict[str, Any]:
+    @staticmethod
+    def _extract_azure_token_usage(response: ChatCompletions) -> dict[str, Any]:
         """Extract token usage from Azure response."""
         if hasattr(response, "usage") and response.usage:
             usage = response.usage
@@ -609,3 +959,20 @@ class AzureCompletion(BaseLLM):
                 "total_tokens": getattr(usage, "total_tokens", 0),
             }
         return {"total_tokens": 0}
+
+    async def aclose(self) -> None:
+        """Close the async client and clean up resources.
+
+        This ensures proper cleanup of the underlying aiohttp session
+        to avoid unclosed connector warnings.
+        """
+        if hasattr(self.async_client, "close"):
+            await self.async_client.close()
+
+    async def __aenter__(self) -> Self:
+        """Async context manager entry."""
+        return self
+
+    async def __aexit__(self, exc_type: Any, exc_val: Any, exc_tb: Any) -> None:
+        """Async context manager exit."""
+        await self.aclose()

lib/crewai/src/crewai/llms/providers/bedrock/completion.py

@@ -1,6 +1,8 @@
 from __future__ import annotations
 
 from collections.abc import Mapping, Sequence
+from contextlib import AsyncExitStack
+import json
 import logging
 import os
 from typing import TYPE_CHECKING, Any, TypedDict, cast
@@ -42,6 +44,16 @@ except ImportError:
         'AWS Bedrock native provider not available, to install: uv add "crewai[bedrock]"'
     ) from None
 
+try:
+    from aiobotocore.session import (  # type: ignore[import-untyped]
+        get_session as get_aiobotocore_session,
+    )
+
+    AIOBOTOCORE_AVAILABLE = True
+except ImportError:
+    AIOBOTOCORE_AVAILABLE = False
+    get_aiobotocore_session = None
+
 
 if TYPE_CHECKING:
 
@@ -221,6 +233,15 @@ class BedrockCompletion(BaseLLM):
         self.client = session.client("bedrock-runtime", config=config)
         self.region_name = region_name
 
+        self.aws_access_key_id = aws_access_key_id or os.getenv("AWS_ACCESS_KEY_ID")
+        self.aws_secret_access_key = aws_secret_access_key or os.getenv(
+            "AWS_SECRET_ACCESS_KEY"
+        )
+        self.aws_session_token = aws_session_token or os.getenv("AWS_SESSION_TOKEN")
+
+        self._async_exit_stack = AsyncExitStack() if AIOBOTOCORE_AVAILABLE else None
+        self._async_client_initialized = False
+
         # Store completion parameters
         self.max_tokens = max_tokens
         self.top_p = top_p
@@ -291,9 +312,14 @@ class BedrockCompletion(BaseLLM):
 
             # Format messages for Converse API
             formatted_messages, system_message = self._format_messages_for_converse(
-                messages  # type: ignore[arg-type]
+                messages
             )
 
+            if not self._invoke_before_llm_call_hooks(
+                cast(list[LLMMessage], formatted_messages), from_agent
+            ):
+                raise ValueError("LLM call blocked by before_llm_call hook")
+
             # Prepare request body
             body: BedrockConverseRequestBody = {
                 "inferenceConfig": self._get_inference_config(),
@@ -335,10 +361,122 @@ class BedrockCompletion(BaseLLM):
 
             if self.stream:
                 return self._handle_streaming_converse(
-                    formatted_messages, body, available_functions, from_task, from_agent
+                    cast(list[LLMMessage], formatted_messages),
+                    body,
+                    available_functions,
+                    from_task,
+                    from_agent,
                 )
 
             return self._handle_converse(
+                cast(list[LLMMessage], formatted_messages),
+                body,
+                available_functions,
+                from_task,
+                from_agent,
+            )
+
+        except Exception as e:
+            if is_context_length_exceeded(e):
+                logging.error(f"Context window exceeded: {e}")
+                raise LLMContextLengthExceededError(str(e)) from e
+
+            error_msg = f"AWS Bedrock API call failed: {e!s}"
+            logging.error(error_msg)
+            self._emit_call_failed_event(
+                error=error_msg, from_task=from_task, from_agent=from_agent
+            )
+            raise
+
+    async def acall(
+        self,
+        messages: str | list[LLMMessage],
+        tools: list[dict[Any, Any]] | None = None,
+        callbacks: list[Any] | None = None,
+        available_functions: dict[str, Any] | None = None,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> str | Any:
+        """Async call to AWS Bedrock Converse API.
+
+        Args:
+            messages: Input messages as string or list of message dicts.
+            tools: Optional list of tool definitions.
+            callbacks: Optional list of callback handlers.
+            available_functions: Optional dict mapping function names to callables.
+            from_task: Optional task context for events.
+            from_agent: Optional agent context for events.
+            response_model: Optional Pydantic model for structured output.
+
+        Returns:
+            Generated text response or structured output.
+
+        Raises:
+            NotImplementedError: If aiobotocore is not installed.
+            LLMContextLengthExceededError: If context window is exceeded.
+        """
+        if not AIOBOTOCORE_AVAILABLE:
+            raise NotImplementedError(
+                "Async support for AWS Bedrock requires aiobotocore. "
+                'Install with: uv add "crewai[bedrock-async]"'
+            )
+
+        try:
+            self._emit_call_started_event(
+                messages=messages,
+                tools=tools,
+                callbacks=callbacks,
+                available_functions=available_functions,
+                from_task=from_task,
+                from_agent=from_agent,
+            )
+
+            formatted_messages, system_message = self._format_messages_for_converse(
+                messages  # type: ignore[arg-type]
+            )
+
+            body: BedrockConverseRequestBody = {
+                "inferenceConfig": self._get_inference_config(),
+            }
+
+            if system_message:
+                body["system"] = cast(
+                    "list[SystemContentBlockTypeDef]",
+                    cast(object, [{"text": system_message}]),
+                )
+
+            if tools:
+                tool_config: ToolConfigurationTypeDef = {
+                    "tools": cast(
+                        "Sequence[ToolTypeDef]",
+                        cast(object, self._format_tools_for_converse(tools)),
+                    )
+                }
+                body["toolConfig"] = tool_config
+
+            if self.guardrail_config:
+                guardrail_config: GuardrailConfigurationTypeDef = cast(
+                    "GuardrailConfigurationTypeDef", cast(object, self.guardrail_config)
+                )
+                body["guardrailConfig"] = guardrail_config
+
+            if self.additional_model_request_fields:
+                body["additionalModelRequestFields"] = (
+                    self.additional_model_request_fields
+                )
+
+            if self.additional_model_response_field_paths:
+                body["additionalModelResponseFieldPaths"] = (
+                    self.additional_model_response_field_paths
+                )
+
+            if self.stream:
+                return await self._ahandle_streaming_converse(
+                    formatted_messages, body, available_functions, from_task, from_agent
+                )
+
+            return await self._ahandle_converse(
                 formatted_messages, body, available_functions, from_task, from_agent
             )
 
@@ -356,7 +494,7 @@ class BedrockCompletion(BaseLLM):
 
     def _handle_converse(
         self,
-        messages: list[dict[str, Any]],
+        messages: list[LLMMessage],
         body: BedrockConverseRequestBody,
         available_functions: Mapping[str, Any] | None = None,
         from_task: Any | None = None,
@@ -480,7 +618,11 @@ class BedrockCompletion(BaseLLM):
                 messages=messages,
             )
 
-            return text_content
+            return self._invoke_after_llm_call_hooks(
+                messages,
+                text_content,
+                from_agent,
+            )
 
         except ClientError as e:
             # Handle all AWS ClientError exceptions as per documentation
@@ -537,7 +679,7 @@ class BedrockCompletion(BaseLLM):
 
     def _handle_streaming_converse(
         self,
-        messages: list[dict[str, Any]],
+        messages: list[LLMMessage],
         body: BedrockConverseRequestBody,
         available_functions: dict[str, Any] | None = None,
         from_task: Any | None = None,
@@ -565,6 +707,341 @@ class BedrockCompletion(BaseLLM):
                         role = event["messageStart"].get("role")
                         logging.debug(f"Streaming message started with role: {role}")
 
+                    elif "contentBlockStart" in event:
+                        start = event["contentBlockStart"].get("start", {})
+                        if "toolUse" in start:
+                            current_tool_use = start["toolUse"]
+                            tool_use_id = current_tool_use.get("toolUseId")
+                        logging.debug(
+                            f"Tool use started in stream: {json.dumps(current_tool_use)} (ID: {tool_use_id})"
+                        )
+
+                    elif "contentBlockDelta" in event:
+                        delta = event["contentBlockDelta"]["delta"]
+                        if "text" in delta:
+                            text_chunk = delta["text"]
+                            logging.debug(f"Streaming text chunk: {text_chunk[:50]}...")
+                            full_response += text_chunk
+                            self._emit_stream_chunk_event(
+                                chunk=text_chunk,
+                                from_task=from_task,
+                                from_agent=from_agent,
+                            )
+                        elif "toolUse" in delta and current_tool_use:
+                            tool_input = delta["toolUse"].get("input", "")
+                            if tool_input:
+                                logging.debug(f"Tool input delta: {tool_input}")
+                    elif "contentBlockStop" in event:
+                        logging.debug("Content block stopped in stream")
+                        if current_tool_use and available_functions:
+                            function_name = current_tool_use["name"]
+                            function_args = cast(
+                                dict[str, Any], current_tool_use.get("input", {})
+                            )
+                            tool_result = self._handle_tool_execution(
+                                function_name=function_name,
+                                function_args=function_args,
+                                available_functions=available_functions,
+                                from_task=from_task,
+                                from_agent=from_agent,
+                            )
+                            if tool_result is not None and tool_use_id:
+                                messages.append(
+                                    {
+                                        "role": "assistant",
+                                        "content": [{"toolUse": current_tool_use}],
+                                    }
+                                )
+                                messages.append(
+                                    {
+                                        "role": "user",
+                                        "content": [
+                                            {
+                                                "toolResult": {
+                                                    "toolUseId": tool_use_id,
+                                                    "content": [
+                                                        {"text": str(tool_result)}
+                                                    ],
+                                                }
+                                            }
+                                        ],
+                                    }
+                                )
+                                return self._handle_converse(
+                                    messages,
+                                    body,
+                                    available_functions,
+                                    from_task,
+                                    from_agent,
+                                )
+                            current_tool_use = None
+                            tool_use_id = None
+                    elif "messageStop" in event:
+                        stop_reason = event["messageStop"].get("stopReason")
+                        logging.debug(f"Streaming message stopped: {stop_reason}")
+                        if stop_reason == "max_tokens":
+                            logging.warning(
+                                "Streaming response truncated due to max_tokens"
+                            )
+                        elif stop_reason == "content_filtered":
+                            logging.warning(
+                                "Streaming response filtered due to content policy"
+                            )
+                            break
+                    elif "metadata" in event:
+                        metadata = event["metadata"]
+                        if "usage" in metadata:
+                            usage_metrics = metadata["usage"]
+                            self._track_token_usage_internal(usage_metrics)
+                            logging.debug(f"Token usage: {usage_metrics}")
+                            if "trace" in metadata:
+                                logging.debug(
+                                    f"Trace information available: {metadata['trace']}"
+                                )
+
+        except ClientError as e:
+            error_msg = self._handle_client_error(e)
+            raise RuntimeError(error_msg) from e
+        except BotoCoreError as e:
+            error_msg = f"Bedrock streaming connection error: {e}"
+            logging.error(error_msg)
+            raise ConnectionError(error_msg) from e
+
+        full_response = self._apply_stop_words(full_response)
+
+        if not full_response or full_response.strip() == "":
+            logging.warning("Bedrock streaming returned empty content, using fallback")
+            full_response = (
+                "I apologize, but I couldn't generate a response. Please try again."
+            )
+
+        self._emit_call_completed_event(
+            response=full_response,
+            call_type=LLMCallType.LLM_CALL,
+            from_task=from_task,
+            from_agent=from_agent,
+            messages=messages,
+        )
+
+        return full_response
+
+    async def _ensure_async_client(self) -> Any:
+        """Ensure async client is initialized and return it."""
+        if not self._async_client_initialized and get_aiobotocore_session:
+            if self._async_exit_stack is None:
+                raise RuntimeError(
+                    "Async exit stack not initialized - aiobotocore not available"
+                )
+            session = get_aiobotocore_session()
+            client = await self._async_exit_stack.enter_async_context(
+                session.create_client(
+                    "bedrock-runtime",
+                    region_name=self.region_name,
+                    aws_access_key_id=self.aws_access_key_id,
+                    aws_secret_access_key=self.aws_secret_access_key,
+                    aws_session_token=self.aws_session_token,
+                )
+            )
+            self._async_client = client
+            self._async_client_initialized = True
+        return self._async_client
+
+    async def _ahandle_converse(
+        self,
+        messages: list[dict[str, Any]],
+        body: BedrockConverseRequestBody,
+        available_functions: Mapping[str, Any] | None = None,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+    ) -> str:
+        """Handle async non-streaming converse API call."""
+        try:
+            if not messages:
+                raise ValueError("Messages cannot be empty")
+
+            for i, msg in enumerate(messages):
+                if (
+                    not isinstance(msg, dict)
+                    or "role" not in msg
+                    or "content" not in msg
+                ):
+                    raise ValueError(f"Invalid message format at index {i}")
+
+            async_client = await self._ensure_async_client()
+            response = await async_client.converse(
+                modelId=self.model_id,
+                messages=cast(
+                    "Sequence[MessageTypeDef | MessageOutputTypeDef]",
+                    cast(object, messages),
+                ),
+                **body,
+            )
+
+            if "usage" in response:
+                self._track_token_usage_internal(response["usage"])
+
+            stop_reason = response.get("stopReason")
+            if stop_reason:
+                logging.debug(f"Response stop reason: {stop_reason}")
+                if stop_reason == "max_tokens":
+                    logging.warning("Response truncated due to max_tokens limit")
+                elif stop_reason == "content_filtered":
+                    logging.warning("Response was filtered due to content policy")
+
+            output = response.get("output", {})
+            message = output.get("message", {})
+            content = message.get("content", [])
+
+            if not content:
+                logging.warning("No content in Bedrock response")
+                return (
+                    "I apologize, but I received an empty response. Please try again."
+                )
+
+            text_content = ""
+
+            for content_block in content:
+                if "text" in content_block:
+                    text_content += content_block["text"]
+
+                elif "toolUse" in content_block and available_functions:
+                    tool_use_block = content_block["toolUse"]
+                    tool_use_id = tool_use_block.get("toolUseId")
+                    function_name = tool_use_block["name"]
+                    function_args = tool_use_block.get("input", {})
+
+                    logging.debug(
+                        f"Tool use requested: {function_name} with ID {tool_use_id}"
+                    )
+
+                    tool_result = self._handle_tool_execution(
+                        function_name=function_name,
+                        function_args=function_args,
+                        available_functions=dict(available_functions),
+                        from_task=from_task,
+                        from_agent=from_agent,
+                    )
+
+                    if tool_result is not None:
+                        messages.append(
+                            {
+                                "role": "assistant",
+                                "content": [{"toolUse": tool_use_block}],
+                            }
+                        )
+
+                        messages.append(
+                            {
+                                "role": "user",
+                                "content": [
+                                    {
+                                        "toolResult": {
+                                            "toolUseId": tool_use_id,
+                                            "content": [{"text": str(tool_result)}],
+                                        }
+                                    }
+                                ],
+                            }
+                        )
+
+                        return await self._ahandle_converse(
+                            messages, body, available_functions, from_task, from_agent
+                        )
+
+            text_content = self._apply_stop_words(text_content)
+
+            if not text_content or text_content.strip() == "":
+                logging.warning("Extracted empty text content from Bedrock response")
+                text_content = "I apologize, but I couldn't generate a proper response. Please try again."
+
+            self._emit_call_completed_event(
+                response=text_content,
+                call_type=LLMCallType.LLM_CALL,
+                from_task=from_task,
+                from_agent=from_agent,
+                messages=messages,
+            )
+
+            return text_content
+
+        except ClientError as e:
+            error_code = e.response.get("Error", {}).get("Code", "Unknown")
+            error_msg = e.response.get("Error", {}).get("Message", str(e))
+            logging.error(f"AWS Bedrock ClientError ({error_code}): {error_msg}")
+
+            if error_code == "ValidationException":
+                if "last turn" in error_msg and "user message" in error_msg:
+                    raise ValueError(
+                        f"Conversation format error: {error_msg}. Check message alternation."
+                    ) from e
+                raise ValueError(f"Request validation failed: {error_msg}") from e
+            if error_code == "AccessDeniedException":
+                raise PermissionError(
+                    f"Access denied to model {self.model_id}: {error_msg}"
+                ) from e
+            if error_code == "ResourceNotFoundException":
+                raise ValueError(f"Model {self.model_id} not found: {error_msg}") from e
+            if error_code == "ThrottlingException":
+                raise RuntimeError(
+                    f"API throttled, please retry later: {error_msg}"
+                ) from e
+            if error_code == "ModelTimeoutException":
+                raise TimeoutError(f"Model request timed out: {error_msg}") from e
+            if error_code == "ServiceQuotaExceededException":
+                raise RuntimeError(f"Service quota exceeded: {error_msg}") from e
+            if error_code == "ModelNotReadyException":
+                raise RuntimeError(
+                    f"Model {self.model_id} not ready: {error_msg}"
+                ) from e
+            if error_code == "ModelErrorException":
+                raise RuntimeError(f"Model error: {error_msg}") from e
+            if error_code == "InternalServerException":
+                raise RuntimeError(f"Internal server error: {error_msg}") from e
+            if error_code == "ServiceUnavailableException":
+                raise RuntimeError(f"Service unavailable: {error_msg}") from e
+
+            raise RuntimeError(f"Bedrock API error ({error_code}): {error_msg}") from e
+
+        except BotoCoreError as e:
+            error_msg = f"Bedrock connection error: {e}"
+            logging.error(error_msg)
+            raise ConnectionError(error_msg) from e
+        except Exception as e:
+            error_msg = f"Unexpected error in Bedrock converse call: {e}"
+            logging.error(error_msg)
+            raise RuntimeError(error_msg) from e
+
+    async def _ahandle_streaming_converse(
+        self,
+        messages: list[dict[str, Any]],
+        body: BedrockConverseRequestBody,
+        available_functions: dict[str, Any] | None = None,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+    ) -> str:
+        """Handle async streaming converse API call."""
+        full_response = ""
+        current_tool_use = None
+        tool_use_id = None
+
+        try:
+            async_client = await self._ensure_async_client()
+            response = await async_client.converse_stream(
+                modelId=self.model_id,
+                messages=cast(
+                    "Sequence[MessageTypeDef | MessageOutputTypeDef]",
+                    cast(object, messages),
+                ),
+                **body,
+            )
+
+            stream = response.get("stream")
+            if stream:
+                async for event in stream:
+                    if "messageStart" in event:
+                        role = event["messageStart"].get("role")
+                        logging.debug(f"Streaming message started with role: {role}")
+
                     elif "contentBlockStart" in event:
                         start = event["contentBlockStart"].get("start", {})
                         if "toolUse" in start:
@@ -590,17 +1067,14 @@ class BedrockCompletion(BaseLLM):
                             if tool_input:
                                 logging.debug(f"Tool input delta: {tool_input}")
 
-                    # Content block stop - end of a content block
                     elif "contentBlockStop" in event:
                         logging.debug("Content block stopped in stream")
-                        # If we were accumulating a tool use, it's now complete
                         if current_tool_use and available_functions:
                             function_name = current_tool_use["name"]
                             function_args = cast(
                                 dict[str, Any], current_tool_use.get("input", {})
                             )
 
-                            # Execute tool
                             tool_result = self._handle_tool_execution(
                                 function_name=function_name,
                                 function_args=function_args,
@@ -610,7 +1084,6 @@ class BedrockCompletion(BaseLLM):
                             )
 
                             if tool_result is not None and tool_use_id:
-                                # Continue conversation with tool result
                                 messages.append(
                                     {
                                         "role": "assistant",
@@ -634,8 +1107,7 @@ class BedrockCompletion(BaseLLM):
                                     }
                                 )
 
-                                # Recursive call - note this switches to non-streaming
-                                return self._handle_converse(
+                                return await self._ahandle_converse(
                                     messages,
                                     body,
                                     available_functions,
@@ -643,10 +1115,9 @@ class BedrockCompletion(BaseLLM):
                                     from_agent,
                                 )
 
-                            current_tool_use = None
-                            tool_use_id = None
+                                current_tool_use = None
+                                tool_use_id = None
 
-                    # Message stop - end of entire message
                     elif "messageStop" in event:
                         stop_reason = event["messageStop"].get("stopReason")
                         logging.debug(f"Streaming message stopped: {stop_reason}")
@@ -660,7 +1131,6 @@ class BedrockCompletion(BaseLLM):
                             )
                         break
 
-                    # Metadata - contains usage information and trace details
                     elif "metadata" in event:
                         metadata = event["metadata"]
                         if "usage" in metadata:
@@ -680,17 +1150,14 @@ class BedrockCompletion(BaseLLM):
             logging.error(error_msg)
             raise ConnectionError(error_msg) from e
 
-        # Apply stop words to full response
         full_response = self._apply_stop_words(full_response)
 
-        # Ensure we don't return empty content
         if not full_response or full_response.strip() == "":
             logging.warning("Bedrock streaming returned empty content, using fallback")
             full_response = (
                 "I apologize, but I couldn't generate a response. Please try again."
             )
 
-        # Emit completion event
         self._emit_call_completed_event(
             response=full_response,
             call_type=LLMCallType.LLM_CALL,
@@ -699,16 +1166,25 @@ class BedrockCompletion(BaseLLM):
             messages=messages,
         )
 
-        return full_response
+        return self._invoke_after_llm_call_hooks(
+            messages,
+            full_response,
+            from_agent,
+        )
 
     def _format_messages_for_converse(
-        self, messages: str | list[dict[str, str]]
+        self, messages: str | list[LLMMessage]
     ) -> tuple[list[dict[str, Any]], str | None]:
-        """Format messages for Converse API following AWS documentation."""
-        # Use base class formatting first
-        formatted_messages = self._format_messages(messages)  # type: ignore[arg-type]
+        """Format messages for Converse API following AWS documentation.
 
-        converse_messages = []
+        Note: Returns dict[str, Any] instead of LLMMessage because Bedrock uses
+        a different content structure: {"role": str, "content": [{"text": str}]}
+        rather than the standard {"role": str, "content": str}.
+        """
+        # Use base class formatting first
+        formatted_messages = self._format_messages(messages)
+
+        converse_messages: list[dict[str, Any]] = []
         system_message: str | None = None
 
         for message in formatted_messages:

lib/crewai/src/crewai/llms/providers/gemini/completion.py

@@ -1,13 +1,14 @@
+from __future__ import annotations
+
 import logging
 import os
 import re
-from typing import Any, cast
+from typing import TYPE_CHECKING, Any, Literal, cast
 
 from pydantic import BaseModel
 
 from crewai.events.types.llm_events import LLMCallType
 from crewai.llms.base_llm import BaseLLM
-from crewai.llms.hooks.base import BaseInterceptor
 from crewai.utilities.agent_utils import is_context_length_exceeded
 from crewai.utilities.exceptions.context_window_exceeding_exception import (
     LLMContextLengthExceededError,
@@ -15,10 +16,15 @@ from crewai.utilities.exceptions.context_window_exceeding_exception import (
 from crewai.utilities.types import LLMMessage
 
 
+if TYPE_CHECKING:
+    from crewai.llms.hooks.base import BaseInterceptor
+
+
 try:
-    from google import genai  # type: ignore[import-untyped]
-    from google.genai import types  # type: ignore[import-untyped]
-    from google.genai.errors import APIError  # type: ignore[import-untyped]
+    from google import genai
+    from google.genai import types
+    from google.genai.errors import APIError
+    from google.genai.types import GenerateContentResponse, Schema
 except ImportError:
     raise ImportError(
         'Google Gen AI native provider not available, to install: uv add "crewai[google-genai]"'
@@ -99,10 +105,13 @@ class GeminiCompletion(BaseLLM):
         self.stream = stream
         self.safety_settings = safety_settings or {}
         self.stop_sequences = stop_sequences or []
+        self.tools: list[dict[str, Any]] | None = None
 
         # Model-specific settings
         version_match = re.search(r"gemini-(\d+(?:\.\d+)?)", model.lower())
-        self.supports_tools = bool(version_match and float(version_match.group(1)) >= 1.5)
+        self.supports_tools = bool(
+            version_match and float(version_match.group(1)) >= 1.5
+        )
 
     @property
     def stop(self) -> list[str]:
@@ -128,7 +137,7 @@ class GeminiCompletion(BaseLLM):
         else:
             self.stop_sequences = []
 
-    def _initialize_client(self, use_vertexai: bool = False) -> genai.Client:  # type: ignore[no-any-unimported]
+    def _initialize_client(self, use_vertexai: bool = False) -> genai.Client:
         """Initialize the Google Gen AI client with proper parameter handling.
 
         Args:
@@ -215,10 +224,93 @@ class GeminiCompletion(BaseLLM):
         Args:
             messages: Input messages for the chat completion
             tools: List of tool/function definitions
-            callbacks: Callback functions (not used as token counts are handled by the reponse)
+            callbacks: Callback functions (not used as token counts are handled by the response)
             available_functions: Available functions for tool calling
             from_task: Task that initiated the call
             from_agent: Agent that initiated the call
+            response_model: Response model to use.
+
+        Returns:
+            Chat completion response or tool call result
+        """
+        try:
+            self._emit_call_started_event(
+                messages=messages,
+                tools=tools,
+                callbacks=callbacks,
+                available_functions=available_functions,
+                from_task=from_task,
+                from_agent=from_agent,
+            )
+            self.tools = tools
+
+            formatted_content, system_instruction = self._format_messages_for_gemini(
+                messages
+            )
+
+            messages_for_hooks = self._convert_contents_to_dict(formatted_content)
+
+            if not self._invoke_before_llm_call_hooks(messages_for_hooks, from_agent):
+                raise ValueError("LLM call blocked by before_llm_call hook")
+
+            config = self._prepare_generation_config(
+                system_instruction, tools, response_model
+            )
+
+            if self.stream:
+                return self._handle_streaming_completion(
+                    formatted_content,
+                    config,
+                    available_functions,
+                    from_task,
+                    from_agent,
+                    response_model,
+                )
+
+            return self._handle_completion(
+                formatted_content,
+                config,
+                available_functions,
+                from_task,
+                from_agent,
+                response_model,
+            )
+
+        except APIError as e:
+            error_msg = f"Google Gemini API error: {e.code} - {e.message}"
+            logging.error(error_msg)
+            self._emit_call_failed_event(
+                error=error_msg, from_task=from_task, from_agent=from_agent
+            )
+            raise
+        except Exception as e:
+            error_msg = f"Google Gemini API call failed: {e!s}"
+            logging.error(error_msg)
+            self._emit_call_failed_event(
+                error=error_msg, from_task=from_task, from_agent=from_agent
+            )
+            raise
+
+    async def acall(
+        self,
+        messages: str | list[LLMMessage],
+        tools: list[dict[str, Any]] | None = None,
+        callbacks: list[Any] | None = None,
+        available_functions: dict[str, Any] | None = None,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> str | Any:
+        """Async call to Google Gemini generate content API.
+
+        Args:
+            messages: Input messages for the chat completion
+            tools: List of tool/function definitions
+            callbacks: Callback functions (not used as token counts are handled by the response)
+            available_functions: Available functions for tool calling
+            from_task: Task that initiated the call
+            from_agent: Agent that initiated the call
+            response_model: Response model to use.
 
         Returns:
             Chat completion response or tool call result
@@ -243,7 +335,7 @@ class GeminiCompletion(BaseLLM):
             )
 
             if self.stream:
-                return self._handle_streaming_completion(
+                return await self._ahandle_streaming_completion(
                     formatted_content,
                     config,
                     available_functions,
@@ -252,9 +344,8 @@ class GeminiCompletion(BaseLLM):
                     response_model,
                 )
 
-            return self._handle_completion(
+            return await self._ahandle_completion(
                 formatted_content,
-                system_instruction,
                 config,
                 available_functions,
                 from_task,
@@ -277,7 +368,7 @@ class GeminiCompletion(BaseLLM):
             )
             raise
 
-    def _prepare_generation_config(  # type: ignore[no-any-unimported]
+    def _prepare_generation_config(
         self,
         system_instruction: str | None = None,
         tools: list[dict[str, Any]] | None = None,
@@ -294,7 +385,7 @@ class GeminiCompletion(BaseLLM):
             GenerateContentConfig object for Gemini API
         """
         self.tools = tools
-        config_params = {}
+        config_params: dict[str, Any] = {}
 
         # Add system instruction if present
         if system_instruction:
@@ -329,7 +420,7 @@ class GeminiCompletion(BaseLLM):
 
         return types.GenerateContentConfig(**config_params)
 
-    def _convert_tools_for_interference(  # type: ignore[no-any-unimported]
+    def _convert_tools_for_interference(  # type: ignore[override]
         self, tools: list[dict[str, Any]]
     ) -> list[types.Tool]:
         """Convert CrewAI tool format to Gemini function declaration format."""
@@ -346,7 +437,7 @@ class GeminiCompletion(BaseLLM):
             )
 
             # Add parameters if present - ensure parameters is a dict
-            if parameters and isinstance(parameters, dict):
+            if parameters and isinstance(parameters, Schema):
                 function_declaration.parameters = parameters
 
             gemini_tool = types.Tool(function_declarations=[function_declaration])
@@ -354,7 +445,7 @@ class GeminiCompletion(BaseLLM):
 
         return gemini_tools
 
-    def _format_messages_for_gemini(  # type: ignore[no-any-unimported]
+    def _format_messages_for_gemini(
         self, messages: str | list[LLMMessage]
     ) -> tuple[list[types.Content], str | None]:
         """Format messages for Gemini API.
@@ -373,88 +464,113 @@ class GeminiCompletion(BaseLLM):
         # Use base class formatting first
         base_formatted = super()._format_messages(messages)
 
-        contents = []
+        contents: list[types.Content] = []
         system_instruction: str | None = None
 
         for message in base_formatted:
-            role = message.get("role")
-            content = message.get("content", "")
+            role = message["role"]
+            content = message["content"]
+
+            # Convert content to string if it's a list
+            if isinstance(content, list):
+                text_content = " ".join(
+                    str(item.get("text", "")) if isinstance(item, dict) else str(item)
+                    for item in content
+                )
+            else:
+                text_content = str(content) if content else ""
 
             if role == "system":
                 # Extract system instruction - Gemini handles it separately
                 if system_instruction:
-                    system_instruction += f"\n\n{content}"
+                    system_instruction += f"\n\n{text_content}"
                 else:
-                    system_instruction = cast(str, content)
+                    system_instruction = text_content
             else:
                 # Convert role for Gemini (assistant -> model)
                 gemini_role = "model" if role == "assistant" else "user"
 
                 # Create Content object
                 gemini_content = types.Content(
-                    role=gemini_role, parts=[types.Part.from_text(text=content)]
+                    role=gemini_role, parts=[types.Part.from_text(text=text_content)]
                 )
                 contents.append(gemini_content)
 
         return contents, system_instruction
 
-    def _handle_completion(  # type: ignore[no-any-unimported]
+    def _validate_and_emit_structured_output(
         self,
-        contents: list[types.Content],
-        system_instruction: str | None,
-        config: types.GenerateContentConfig,
-        available_functions: dict[str, Any] | None = None,
+        content: str,
+        response_model: type[BaseModel],
+        messages_for_event: list[LLMMessage],
         from_task: Any | None = None,
         from_agent: Any | None = None,
-        response_model: type[BaseModel] | None = None,
-    ) -> str | Any:
-        """Handle non-streaming content generation."""
-        api_params = {
-            "model": self.model,
-            "contents": contents,
-            "config": config,
-        }
+    ) -> str:
+        """Validate content against response model and emit completion event.
 
+        Args:
+            content: Response content to validate
+            response_model: Pydantic model for validation
+            messages_for_event: Messages to include in event
+            from_task: Task that initiated the call
+            from_agent: Agent that initiated the call
+
+        Returns:
+            Validated and serialized JSON string
+
+        Raises:
+            ValueError: If validation fails
+        """
         try:
-            response = self.client.models.generate_content(**api_params)
+            structured_data = response_model.model_validate_json(content)
+            structured_json = structured_data.model_dump_json()
 
-            usage = self._extract_token_usage(response)
+            self._emit_call_completed_event(
+                response=structured_json,
+                call_type=LLMCallType.LLM_CALL,
+                from_task=from_task,
+                from_agent=from_agent,
+                messages=messages_for_event,
+            )
+
+            return structured_json
         except Exception as e:
-            if is_context_length_exceeded(e):
-                logging.error(f"Context window exceeded: {e}")
-                raise LLMContextLengthExceededError(str(e)) from e
-            raise e from e
+            error_msg = f"Failed to validate structured output with model {response_model.__name__}: {e}"
+            logging.error(error_msg)
+            raise ValueError(error_msg) from e
 
-        self._track_token_usage_internal(usage)
+    def _finalize_completion_response(
+        self,
+        content: str,
+        contents: list[types.Content],
+        response_model: type[BaseModel] | None = None,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+    ) -> str:
+        """Finalize completion response with validation and event emission.
 
-        if response.candidates and (self.tools or available_functions):
-            candidate = response.candidates[0]
-            if candidate.content and candidate.content.parts:
-                for part in candidate.content.parts:
-                    if hasattr(part, "function_call") and part.function_call:
-                        function_name = part.function_call.name
-                        function_args = (
-                            dict(part.function_call.args)
-                            if part.function_call.args
-                            else {}
-                        )
-
-                        result = self._handle_tool_execution(
-                            function_name=function_name,
-                            function_args=function_args,
-                            available_functions=available_functions,  # type: ignore
-                            from_task=from_task,
-                            from_agent=from_agent,
-                        )
-
-                        if result is not None:
-                            return result
-
-        content = response.text if hasattr(response, "text") else ""
-        content = self._apply_stop_words(content)
+        Args:
+            content: The response content
+            contents: Original contents for event conversion
+            response_model: Pydantic model for structured output validation
+            from_task: Task that initiated the call
+            from_agent: Agent that initiated the call
 
+        Returns:
+            Final response content after processing
+        """
         messages_for_event = self._convert_contents_to_dict(contents)
 
+        # Handle structured output validation
+        if response_model:
+            return self._validate_and_emit_structured_output(
+                content=content,
+                response_model=response_model,
+                messages_for_event=messages_for_event,
+                from_task=from_task,
+                from_agent=from_agent,
+            )
+
         self._emit_call_completed_event(
             response=content,
             call_type=LLMCallType.LLM_CALL,
@@ -463,49 +579,144 @@ class GeminiCompletion(BaseLLM):
             messages=messages_for_event,
         )
 
-        return content
+        return self._invoke_after_llm_call_hooks(
+            messages_for_event, content, from_agent
+        )
 
-    def _handle_streaming_completion(  # type: ignore[no-any-unimported]
+    def _process_response_with_tools(
         self,
+        response: GenerateContentResponse,
+        contents: list[types.Content],
+        available_functions: dict[str, Any] | None = None,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> str | Any:
+        """Process response, execute function calls, and finalize completion.
+
+        Args:
+            response: The completion response
+            contents: Original contents for event conversion
+            available_functions: Available functions for function calling
+            from_task: Task that initiated the call
+            from_agent: Agent that initiated the call
+            response_model: Pydantic model for structured output validation
+
+        Returns:
+            Final response content or function call result
+        """
+        if response.candidates and (self.tools or available_functions):
+            candidate = response.candidates[0]
+            if candidate.content and candidate.content.parts:
+                for part in candidate.content.parts:
+                    if hasattr(part, "function_call") and part.function_call:
+                        function_name = part.function_call.name
+                        if function_name is None:
+                            continue
+                        function_args = (
+                            dict(part.function_call.args)
+                            if part.function_call.args
+                            else {}
+                        )
+
+                        result = self._handle_tool_execution(
+                            function_name=function_name,
+                            function_args=function_args,
+                            available_functions=available_functions or {},
+                            from_task=from_task,
+                            from_agent=from_agent,
+                        )
+
+                        if result is not None:
+                            return result
+
+        content = response.text or ""
+        content = self._apply_stop_words(content)
+
+        return self._finalize_completion_response(
+            content=content,
+            contents=contents,
+            response_model=response_model,
+            from_task=from_task,
+            from_agent=from_agent,
+        )
+
+    def _process_stream_chunk(
+        self,
+        chunk: GenerateContentResponse,
+        full_response: str,
+        function_calls: dict[str, dict[str, Any]],
+        usage_data: dict[str, int],
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+    ) -> tuple[str, dict[str, dict[str, Any]], dict[str, int]]:
+        """Process a single streaming chunk.
+
+        Args:
+            chunk: The streaming chunk response
+            full_response: Accumulated response text
+            function_calls: Accumulated function calls
+            usage_data: Accumulated usage data
+            from_task: Task that initiated the call
+            from_agent: Agent that initiated the call
+
+        Returns:
+            Tuple of (updated full_response, updated function_calls, updated usage_data)
+        """
+        if chunk.usage_metadata:
+            usage_data = self._extract_token_usage(chunk)
+
+        if chunk.text:
+            full_response += chunk.text
+            self._emit_stream_chunk_event(
+                chunk=chunk.text,
+                from_task=from_task,
+                from_agent=from_agent,
+            )
+
+        if chunk.candidates:
+            candidate = chunk.candidates[0]
+            if candidate.content and candidate.content.parts:
+                for part in candidate.content.parts:
+                    if hasattr(part, "function_call") and part.function_call:
+                        call_id = part.function_call.name or "default"
+                        if call_id not in function_calls:
+                            function_calls[call_id] = {
+                                "name": part.function_call.name,
+                                "args": dict(part.function_call.args)
+                                if part.function_call.args
+                                else {},
+                            }
+
+        return full_response, function_calls, usage_data
+
+    def _finalize_streaming_response(
+        self,
+        full_response: str,
+        function_calls: dict[str, dict[str, Any]],
+        usage_data: dict[str, int],
         contents: list[types.Content],
-        config: types.GenerateContentConfig,
         available_functions: dict[str, Any] | None = None,
         from_task: Any | None = None,
         from_agent: Any | None = None,
         response_model: type[BaseModel] | None = None,
     ) -> str:
-        """Handle streaming content generation."""
-        full_response = ""
-        function_calls = {}
+        """Finalize streaming response with usage tracking, function execution, and events.
 
-        api_params = {
-            "model": self.model,
-            "contents": contents,
-            "config": config,
-        }
+        Args:
+            full_response: The complete streamed response content
+            function_calls: Dictionary of function calls accumulated during streaming
+            usage_data: Token usage data from the stream
+            contents: Original contents for event conversion
+            available_functions: Available functions for function calling
+            from_task: Task that initiated the call
+            from_agent: Agent that initiated the call
+            response_model: Pydantic model for structured output validation
 
-        for chunk in self.client.models.generate_content_stream(**api_params):
-            if hasattr(chunk, "text") and chunk.text:
-                full_response += chunk.text
-                self._emit_stream_chunk_event(
-                    chunk=chunk.text,
-                    from_task=from_task,
-                    from_agent=from_agent,
-                )
-
-            if hasattr(chunk, "candidates") and chunk.candidates:
-                candidate = chunk.candidates[0]
-                if candidate.content and candidate.content.parts:
-                    for part in candidate.content.parts:
-                        if hasattr(part, "function_call") and part.function_call:
-                            call_id = part.function_call.name or "default"
-                            if call_id not in function_calls:
-                                function_calls[call_id] = {
-                                    "name": part.function_call.name,
-                                    "args": dict(part.function_call.args)
-                                    if part.function_call.args
-                                    else {},
-                                }
+        Returns:
+            Final response content after processing
+        """
+        self._track_token_usage_internal(usage_data)
 
         # Handle completed function calls
         if function_calls and available_functions:
@@ -513,6 +724,14 @@ class GeminiCompletion(BaseLLM):
                 function_name = call_data["name"]
                 function_args = call_data["args"]
 
+                # Skip if function_name is None
+                if not isinstance(function_name, str):
+                    continue
+
+                # Ensure function_args is a dict
+                if not isinstance(function_args, dict):
+                    function_args = {}
+
                 # Execute tool
                 result = self._handle_tool_execution(
                     function_name=function_name,
@@ -525,17 +744,170 @@ class GeminiCompletion(BaseLLM):
                 if result is not None:
                     return result
 
-        messages_for_event = self._convert_contents_to_dict(contents)
-
-        self._emit_call_completed_event(
-            response=full_response,
-            call_type=LLMCallType.LLM_CALL,
+        return self._finalize_completion_response(
+            content=full_response,
+            contents=contents,
+            response_model=response_model,
             from_task=from_task,
             from_agent=from_agent,
-            messages=messages_for_event,
         )
 
-        return full_response
+    def _handle_completion(
+        self,
+        contents: list[types.Content],
+        config: types.GenerateContentConfig,
+        available_functions: dict[str, Any] | None = None,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> str | Any:
+        """Handle non-streaming content generation."""
+        try:
+            # The API accepts list[Content] but mypy is overly strict about variance
+            contents_for_api: Any = contents
+            response = self.client.models.generate_content(
+                model=self.model,
+                contents=contents_for_api,
+                config=config,
+            )
+
+            usage = self._extract_token_usage(response)
+        except Exception as e:
+            if is_context_length_exceeded(e):
+                logging.error(f"Context window exceeded: {e}")
+                raise LLMContextLengthExceededError(str(e)) from e
+            raise e from e
+
+        self._track_token_usage_internal(usage)
+
+        return self._process_response_with_tools(
+            response=response,
+            contents=contents,
+            available_functions=available_functions,
+            from_task=from_task,
+            from_agent=from_agent,
+            response_model=response_model,
+        )
+
+    def _handle_streaming_completion(
+        self,
+        contents: list[types.Content],
+        config: types.GenerateContentConfig,
+        available_functions: dict[str, Any] | None = None,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> str:
+        """Handle streaming content generation."""
+        full_response = ""
+        function_calls: dict[str, dict[str, Any]] = {}
+        usage_data = {"total_tokens": 0}
+
+        # The API accepts list[Content] but mypy is overly strict about variance
+        contents_for_api: Any = contents
+        for chunk in self.client.models.generate_content_stream(
+            model=self.model,
+            contents=contents_for_api,
+            config=config,
+        ):
+            full_response, function_calls, usage_data = self._process_stream_chunk(
+                chunk=chunk,
+                full_response=full_response,
+                function_calls=function_calls,
+                usage_data=usage_data,
+                from_task=from_task,
+                from_agent=from_agent,
+            )
+
+        return self._finalize_streaming_response(
+            full_response=full_response,
+            function_calls=function_calls,
+            usage_data=usage_data,
+            contents=contents,
+            available_functions=available_functions,
+            from_task=from_task,
+            from_agent=from_agent,
+            response_model=response_model,
+        )
+
+    async def _ahandle_completion(
+        self,
+        contents: list[types.Content],
+        config: types.GenerateContentConfig,
+        available_functions: dict[str, Any] | None = None,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> str | Any:
+        """Handle async non-streaming content generation."""
+        try:
+            # The API accepts list[Content] but mypy is overly strict about variance
+            contents_for_api: Any = contents
+            response = await self.client.aio.models.generate_content(
+                model=self.model,
+                contents=contents_for_api,
+                config=config,
+            )
+
+            usage = self._extract_token_usage(response)
+        except Exception as e:
+            if is_context_length_exceeded(e):
+                logging.error(f"Context window exceeded: {e}")
+                raise LLMContextLengthExceededError(str(e)) from e
+            raise e from e
+
+        self._track_token_usage_internal(usage)
+
+        return self._process_response_with_tools(
+            response=response,
+            contents=contents,
+            available_functions=available_functions,
+            from_task=from_task,
+            from_agent=from_agent,
+            response_model=response_model,
+        )
+
+    async def _ahandle_streaming_completion(
+        self,
+        contents: list[types.Content],
+        config: types.GenerateContentConfig,
+        available_functions: dict[str, Any] | None = None,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> str:
+        """Handle async streaming content generation."""
+        full_response = ""
+        function_calls: dict[str, dict[str, Any]] = {}
+        usage_data = {"total_tokens": 0}
+
+        # The API accepts list[Content] but mypy is overly strict about variance
+        contents_for_api: Any = contents
+        stream = await self.client.aio.models.generate_content_stream(
+            model=self.model,
+            contents=contents_for_api,
+            config=config,
+        )
+        async for chunk in stream:
+            full_response, function_calls, usage_data = self._process_stream_chunk(
+                chunk=chunk,
+                full_response=full_response,
+                function_calls=function_calls,
+                usage_data=usage_data,
+                from_task=from_task,
+                from_agent=from_agent,
+            )
+
+        return self._finalize_streaming_response(
+            full_response=full_response,
+            function_calls=function_calls,
+            usage_data=usage_data,
+            contents=contents,
+            available_functions=available_functions,
+            from_task=from_task,
+            from_agent=from_agent,
+            response_model=response_model,
+        )
 
     def supports_function_calling(self) -> bool:
         """Check if the model supports function calling."""
@@ -583,9 +955,10 @@ class GeminiCompletion(BaseLLM):
         # Default context window size for Gemini models
         return int(1048576 * CONTEXT_WINDOW_USAGE_RATIO)  # 1M tokens
 
-    def _extract_token_usage(self, response: dict[str, Any]) -> dict[str, Any]:
+    @staticmethod
+    def _extract_token_usage(response: GenerateContentResponse) -> dict[str, Any]:
         """Extract token usage from Gemini response."""
-        if hasattr(response, "usage_metadata"):
+        if response.usage_metadata:
             usage = response.usage_metadata
             return {
                 "prompt_token_count": getattr(usage, "prompt_token_count", 0),
@@ -595,21 +968,28 @@ class GeminiCompletion(BaseLLM):
             }
         return {"total_tokens": 0}
 
-    def _convert_contents_to_dict(  # type: ignore[no-any-unimported]
-        self,
+    @staticmethod
+    def _convert_contents_to_dict(
         contents: list[types.Content],
-    ) -> list[dict[str, str]]:
+    ) -> list[LLMMessage]:
         """Convert contents to dict format."""
-        return [
-            {
-                "role": "assistant"
-                if content_obj.role == "model"
-                else content_obj.role,
-                "content": " ".join(
-                    part.text
-                    for part in content_obj.parts
-                    if hasattr(part, "text") and part.text
-                ),
-            }
-            for content_obj in contents
-        ]
+        result: list[LLMMessage] = []
+        for content_obj in contents:
+            role = content_obj.role
+            if role == "model":
+                role = "assistant"
+            elif role is None:
+                role = "user"
+
+            parts = content_obj.parts or []
+            content = " ".join(
+                part.text for part in parts if hasattr(part, "text") and part.text
+            )
+
+            result.append(
+                LLMMessage(
+                    role=cast(Literal["user", "assistant", "system"], role),
+                    content=content,
+                )
+            )
+        return result

lib/crewai/src/crewai/llms/providers/openai/completion.py

@@ -1,13 +1,14 @@
 from __future__ import annotations
 
-from collections.abc import Iterator
+from collections.abc import AsyncIterator
 import json
 import logging
 import os
 from typing import TYPE_CHECKING, Any
 
 import httpx
-from openai import APIConnectionError, NotFoundError, OpenAI
+from openai import APIConnectionError, AsyncOpenAI, NotFoundError, OpenAI, Stream
+from openai.lib.streaming.chat import ChatCompletionStream
 from openai.types.chat import ChatCompletion, ChatCompletionChunk
 from openai.types.chat.chat_completion import Choice
 from openai.types.chat.chat_completion_chunk import ChoiceDelta
@@ -15,12 +16,12 @@ from pydantic import BaseModel
 
 from crewai.events.types.llm_events import LLMCallType
 from crewai.llms.base_llm import BaseLLM
-from crewai.llms.hooks.transport import HTTPTransport
+from crewai.llms.hooks.transport import AsyncHTTPTransport, HTTPTransport
 from crewai.utilities.agent_utils import is_context_length_exceeded
-from crewai.utilities.converter import generate_model_description
 from crewai.utilities.exceptions.context_window_exceeding_exception import (
     LLMContextLengthExceededError,
 )
+from crewai.utilities.pydantic_schema_utils import generate_model_description
 from crewai.utilities.types import LLMMessage
 
 
@@ -101,6 +102,14 @@ class OpenAICompletion(BaseLLM):
 
         self.client = OpenAI(**client_config)
 
+        async_client_config = self._get_client_params()
+        if self.interceptor:
+            async_transport = AsyncHTTPTransport(interceptor=self.interceptor)
+            async_http_client = httpx.AsyncClient(transport=async_transport)
+            async_client_config["http_client"] = async_http_client
+
+        self.async_client = AsyncOpenAI(**async_client_config)
+
         # Completion parameters
         self.top_p = top_p
         self.frequency_penalty = frequency_penalty
@@ -181,6 +190,9 @@ class OpenAICompletion(BaseLLM):
 
             formatted_messages = self._format_messages(messages)
 
+            if not self._invoke_before_llm_call_hooks(formatted_messages, from_agent):
+                raise ValueError("LLM call blocked by before_llm_call hook")
+
             completion_params = self._prepare_completion_params(
                 messages=formatted_messages, tools=tools
             )
@@ -210,6 +222,71 @@ class OpenAICompletion(BaseLLM):
             )
             raise
 
+    async def acall(
+        self,
+        messages: str | list[LLMMessage],
+        tools: list[dict[str, BaseTool]] | None = None,
+        callbacks: list[Any] | None = None,
+        available_functions: dict[str, Any] | None = None,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> str | Any:
+        """Async call to OpenAI chat completion API.
+
+        Args:
+            messages: Input messages for the chat completion
+            tools: list of tool/function definitions
+            callbacks: Callback functions (not used in native implementation)
+            available_functions: Available functions for tool calling
+            from_task: Task that initiated the call
+            from_agent: Agent that initiated the call
+            response_model: Response model for structured output.
+
+        Returns:
+            Chat completion response or tool call result
+        """
+        try:
+            self._emit_call_started_event(
+                messages=messages,
+                tools=tools,
+                callbacks=callbacks,
+                available_functions=available_functions,
+                from_task=from_task,
+                from_agent=from_agent,
+            )
+
+            formatted_messages = self._format_messages(messages)
+
+            completion_params = self._prepare_completion_params(
+                messages=formatted_messages, tools=tools
+            )
+
+            if self.stream:
+                return await self._ahandle_streaming_completion(
+                    params=completion_params,
+                    available_functions=available_functions,
+                    from_task=from_task,
+                    from_agent=from_agent,
+                    response_model=response_model,
+                )
+
+            return await self._ahandle_completion(
+                params=completion_params,
+                available_functions=available_functions,
+                from_task=from_task,
+                from_agent=from_agent,
+                response_model=response_model,
+            )
+
+        except Exception as e:
+            error_msg = f"OpenAI API call failed: {e!s}"
+            logging.error(error_msg)
+            self._emit_call_failed_event(
+                error=error_msg, from_task=from_task, from_agent=from_agent
+            )
+            raise
+
     def _prepare_completion_params(
         self, messages: list[LLMMessage], tools: list[dict[str, BaseTool]] | None = None
     ) -> dict[str, Any]:
@@ -220,6 +297,7 @@ class OpenAICompletion(BaseLLM):
         }
         if self.stream:
             params["stream"] = self.stream
+            params["stream_options"] = {"include_usage": True}
 
         params.update(self.additional_params)
 
@@ -352,10 +430,283 @@ class OpenAICompletion(BaseLLM):
 
             if message.tool_calls and available_functions:
                 tool_call = message.tool_calls[0]
-                function_name = tool_call.function.name  # type: ignore[union-attr]
+                function_name = tool_call.function.name
 
                 try:
-                    function_args = json.loads(tool_call.function.arguments)  # type: ignore[union-attr]
+                    function_args = json.loads(tool_call.function.arguments)
+                except json.JSONDecodeError as e:
+                    logging.error(f"Failed to parse tool arguments: {e}")
+                    function_args = {}
+
+                result = self._handle_tool_execution(
+                    function_name=function_name,
+                    function_args=function_args,
+                    available_functions=available_functions,
+                    from_task=from_task,
+                    from_agent=from_agent,
+                )
+
+                if result is not None:
+                    return result
+
+            content = message.content or ""
+            content = self._apply_stop_words(content)
+
+            if self.response_format and isinstance(self.response_format, type):
+                try:
+                    structured_result = self._validate_structured_output(
+                        content, self.response_format
+                    )
+                    self._emit_call_completed_event(
+                        response=structured_result,
+                        call_type=LLMCallType.LLM_CALL,
+                        from_task=from_task,
+                        from_agent=from_agent,
+                        messages=params["messages"],
+                    )
+                    return structured_result
+                except ValueError as e:
+                    logging.warning(f"Structured output validation failed: {e}")
+
+            self._emit_call_completed_event(
+                response=content,
+                call_type=LLMCallType.LLM_CALL,
+                from_task=from_task,
+                from_agent=from_agent,
+                messages=params["messages"],
+            )
+
+            if usage.get("total_tokens", 0) > 0:
+                logging.info(f"OpenAI API usage: {usage}")
+
+            content = self._invoke_after_llm_call_hooks(
+                params["messages"], content, from_agent
+            )
+        except NotFoundError as e:
+            error_msg = f"Model {self.model} not found: {e}"
+            logging.error(error_msg)
+            self._emit_call_failed_event(
+                error=error_msg, from_task=from_task, from_agent=from_agent
+            )
+            raise ValueError(error_msg) from e
+        except APIConnectionError as e:
+            error_msg = f"Failed to connect to OpenAI API: {e}"
+            logging.error(error_msg)
+            self._emit_call_failed_event(
+                error=error_msg, from_task=from_task, from_agent=from_agent
+            )
+            raise ConnectionError(error_msg) from e
+        except Exception as e:
+            # Handle context length exceeded and other errors
+            if is_context_length_exceeded(e):
+                logging.error(f"Context window exceeded: {e}")
+                raise LLMContextLengthExceededError(str(e)) from e
+
+            error_msg = f"OpenAI API call failed: {e!s}"
+            logging.error(error_msg)
+            self._emit_call_failed_event(
+                error=error_msg, from_task=from_task, from_agent=from_agent
+            )
+            raise e from e
+
+        return content
+
+    def _handle_streaming_completion(
+        self,
+        params: dict[str, Any],
+        available_functions: dict[str, Any] | None = None,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> str:
+        """Handle streaming chat completion."""
+        full_response = ""
+        tool_calls = {}
+
+        if response_model:
+            parse_params = {
+                k: v
+                for k, v in params.items()
+                if k not in ("response_format", "stream")
+            }
+
+            stream: ChatCompletionStream[BaseModel]
+            with self.client.beta.chat.completions.stream(
+                **parse_params, response_format=response_model
+            ) as stream:
+                for chunk in stream:
+                    if chunk.type == "content.delta":
+                        delta_content = chunk.delta
+                        if delta_content:
+                            self._emit_stream_chunk_event(
+                                chunk=delta_content,
+                                from_task=from_task,
+                                from_agent=from_agent,
+                            )
+
+                final_completion = stream.get_final_completion()
+                if final_completion:
+                    usage = self._extract_openai_token_usage(final_completion)
+                    self._track_token_usage_internal(usage)
+                    if final_completion.choices:
+                        parsed_result = final_completion.choices[0].message.parsed
+                        if parsed_result:
+                            structured_json = parsed_result.model_dump_json()
+                            self._emit_call_completed_event(
+                                response=structured_json,
+                                call_type=LLMCallType.LLM_CALL,
+                                from_task=from_task,
+                                from_agent=from_agent,
+                                messages=params["messages"],
+                            )
+                            return structured_json
+
+            logging.error("Failed to get parsed result from stream")
+            return ""
+
+        completion_stream: Stream[ChatCompletionChunk] = (
+            self.client.chat.completions.create(**params)
+        )
+
+        usage_data = {"total_tokens": 0}
+
+        for completion_chunk in completion_stream:
+            if hasattr(completion_chunk, "usage") and completion_chunk.usage:
+                usage_data = self._extract_openai_token_usage(completion_chunk)
+                continue
+
+            if not completion_chunk.choices:
+                continue
+
+            choice = completion_chunk.choices[0]
+            chunk_delta: ChoiceDelta = choice.delta
+
+            if chunk_delta.content:
+                full_response += chunk_delta.content
+                self._emit_stream_chunk_event(
+                    chunk=chunk_delta.content,
+                    from_task=from_task,
+                    from_agent=from_agent,
+                )
+
+            if chunk_delta.tool_calls:
+                for tool_call in chunk_delta.tool_calls:
+                    call_id = tool_call.id or "default"
+                    if call_id not in tool_calls:
+                        tool_calls[call_id] = {
+                            "name": "",
+                            "arguments": "",
+                        }
+
+                    if tool_call.function and tool_call.function.name:
+                        tool_calls[call_id]["name"] = tool_call.function.name
+                    if tool_call.function and tool_call.function.arguments:
+                        tool_calls[call_id]["arguments"] += tool_call.function.arguments
+
+        self._track_token_usage_internal(usage_data)
+
+        if tool_calls and available_functions:
+            for call_data in tool_calls.values():
+                function_name = call_data["name"]
+                arguments = call_data["arguments"]
+
+                # Skip if function name is empty or arguments are empty
+                if not function_name or not arguments:
+                    continue
+
+                # Check if function exists in available functions
+                if function_name not in available_functions:
+                    logging.warning(
+                        f"Function '{function_name}' not found in available functions"
+                    )
+                    continue
+
+                try:
+                    function_args = json.loads(arguments)
+                except json.JSONDecodeError as e:
+                    logging.error(f"Failed to parse streamed tool arguments: {e}")
+                    continue
+
+                result = self._handle_tool_execution(
+                    function_name=function_name,
+                    function_args=function_args,
+                    available_functions=available_functions,
+                    from_task=from_task,
+                    from_agent=from_agent,
+                )
+
+                if result is not None:
+                    return result
+
+        full_response = self._apply_stop_words(full_response)
+
+        self._emit_call_completed_event(
+            response=full_response,
+            call_type=LLMCallType.LLM_CALL,
+            from_task=from_task,
+            from_agent=from_agent,
+            messages=params["messages"],
+        )
+
+        return self._invoke_after_llm_call_hooks(
+            params["messages"], full_response, from_agent
+        )
+
+    async def _ahandle_completion(
+        self,
+        params: dict[str, Any],
+        available_functions: dict[str, Any] | None = None,
+        from_task: Any | None = None,
+        from_agent: Any | None = None,
+        response_model: type[BaseModel] | None = None,
+    ) -> str | Any:
+        """Handle non-streaming async chat completion."""
+        try:
+            if response_model:
+                parse_params = {
+                    k: v for k, v in params.items() if k != "response_format"
+                }
+                parsed_response = await self.async_client.beta.chat.completions.parse(
+                    **parse_params,
+                    response_format=response_model,
+                )
+                math_reasoning = parsed_response.choices[0].message
+
+                if math_reasoning.refusal:
+                    pass
+
+                usage = self._extract_openai_token_usage(parsed_response)
+                self._track_token_usage_internal(usage)
+
+                parsed_object = parsed_response.choices[0].message.parsed
+                if parsed_object:
+                    structured_json = parsed_object.model_dump_json()
+                    self._emit_call_completed_event(
+                        response=structured_json,
+                        call_type=LLMCallType.LLM_CALL,
+                        from_task=from_task,
+                        from_agent=from_agent,
+                        messages=params["messages"],
+                    )
+                    return structured_json
+
+            response: ChatCompletion = await self.async_client.chat.completions.create(
+                **params
+            )
+
+            usage = self._extract_openai_token_usage(response)
+
+            self._track_token_usage_internal(usage)
+
+            choice: Choice = response.choices[0]
+            message = choice.message
+
+            if message.tool_calls and available_functions:
+                tool_call = message.tool_calls[0]
+                function_name = tool_call.function.name
+
+                try:
+                    function_args = json.loads(tool_call.function.arguments)
                 except json.JSONDecodeError as e:
                     logging.error(f"Failed to parse tool arguments: {e}")
                     function_args = {}
@@ -415,7 +766,6 @@ class OpenAICompletion(BaseLLM):
             )
             raise ConnectionError(error_msg) from e
         except Exception as e:
-            # Handle context length exceeded and other errors
             if is_context_length_exceeded(e):
                 logging.error(f"Context window exceeded: {e}")
                 raise LLMContextLengthExceededError(str(e)) from e
@@ -429,7 +779,7 @@ class OpenAICompletion(BaseLLM):
 
         return content
 
-    def _handle_streaming_completion(
+    async def _ahandle_streaming_completion(
         self,
         params: dict[str, Any],
         available_functions: dict[str, Any] | None = None,
@@ -437,17 +787,22 @@ class OpenAICompletion(BaseLLM):
         from_agent: Any | None = None,
         response_model: type[BaseModel] | None = None,
     ) -> str:
-        """Handle streaming chat completion."""
+        """Handle async streaming chat completion."""
         full_response = ""
         tool_calls = {}
 
         if response_model:
-            completion_stream: Iterator[ChatCompletionChunk] = (
-                self.client.chat.completions.create(**params)
-            )
+            completion_stream: AsyncIterator[
+                ChatCompletionChunk
+            ] = await self.async_client.chat.completions.create(**params)
 
             accumulated_content = ""
-            for chunk in completion_stream:
+            usage_data = {"total_tokens": 0}
+            async for chunk in completion_stream:
+                if hasattr(chunk, "usage") and chunk.usage:
+                    usage_data = self._extract_openai_token_usage(chunk)
+                    continue
+
                 if not chunk.choices:
                     continue
 
@@ -462,6 +817,8 @@ class OpenAICompletion(BaseLLM):
                         from_agent=from_agent,
                     )
 
+            self._track_token_usage_internal(usage_data)
+
             try:
                 parsed_object = response_model.model_validate_json(accumulated_content)
                 structured_json = parsed_object.model_dump_json()
@@ -486,11 +843,17 @@ class OpenAICompletion(BaseLLM):
                 )
                 return accumulated_content
 
-        stream: Iterator[ChatCompletionChunk] = self.client.chat.completions.create(
-            **params
-        )
+        stream: AsyncIterator[
+            ChatCompletionChunk
+        ] = await self.async_client.chat.completions.create(**params)
+
+        usage_data = {"total_tokens": 0}
+
+        async for chunk in stream:
+            if hasattr(chunk, "usage") and chunk.usage:
+                usage_data = self._extract_openai_token_usage(chunk)
+                continue
 
-        for chunk in stream:
             if not chunk.choices:
                 continue
 
@@ -519,16 +882,16 @@ class OpenAICompletion(BaseLLM):
                     if tool_call.function and tool_call.function.arguments:
                         tool_calls[call_id]["arguments"] += tool_call.function.arguments
 
+        self._track_token_usage_internal(usage_data)
+
         if tool_calls and available_functions:
             for call_data in tool_calls.values():
                 function_name = call_data["name"]
                 arguments = call_data["arguments"]
 
-                # Skip if function name is empty or arguments are empty
                 if not function_name or not arguments:
                     continue
 
-                # Check if function exists in available functions
                 if function_name not in available_functions:
                     logging.warning(
                         f"Function '{function_name}' not found in available functions"
@@ -608,8 +971,10 @@ class OpenAICompletion(BaseLLM):
         # Default context window size
         return int(8192 * CONTEXT_WINDOW_USAGE_RATIO)
 
-    def _extract_openai_token_usage(self, response: ChatCompletion) -> dict[str, Any]:
-        """Extract token usage from OpenAI ChatCompletion response."""
+    def _extract_openai_token_usage(
+        self, response: ChatCompletion | ChatCompletionChunk
+    ) -> dict[str, Any]:
+        """Extract token usage from OpenAI ChatCompletion or ChatCompletionChunk response."""
         if hasattr(response, "usage") and response.usage:
             usage = response.usage
             return {

lib/crewai/src/crewai/memory/contextual/contextual_memory.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import asyncio
 from typing import TYPE_CHECKING
 
 from crewai.memory import (
@@ -16,6 +17,8 @@ if TYPE_CHECKING:
 
 
 class ContextualMemory:
+    """Aggregates and retrieves context from multiple memory sources."""
+
     def __init__(
         self,
         stm: ShortTermMemory,
@@ -46,9 +49,14 @@ class ContextualMemory:
             self.exm.task = self.task
 
     def build_context_for_task(self, task: Task, context: str) -> str:
-        """
-        Automatically builds a minimal, highly relevant set of contextual information
-        for a given task.
+        """Build contextual information for a task synchronously.
+
+        Args:
+            task: The task to build context for.
+            context: Additional context string.
+
+        Returns:
+            Formatted context string from all memory sources.
         """
         query = f"{task.description} {context}".strip()
 
@@ -63,6 +71,31 @@ class ContextualMemory:
         ]
         return "\n".join(filter(None, context_parts))
 
+    async def abuild_context_for_task(self, task: Task, context: str) -> str:
+        """Build contextual information for a task asynchronously.
+
+        Args:
+            task: The task to build context for.
+            context: Additional context string.
+
+        Returns:
+            Formatted context string from all memory sources.
+        """
+        query = f"{task.description} {context}".strip()
+
+        if query == "":
+            return ""
+
+        # Fetch all contexts concurrently
+        results = await asyncio.gather(
+            self._afetch_ltm_context(task.description),
+            self._afetch_stm_context(query),
+            self._afetch_entity_context(query),
+            self._afetch_external_context(query),
+        )
+
+        return "\n".join(filter(None, results))
+
     def _fetch_stm_context(self, query: str) -> str:
         """
         Fetches recent relevant insights from STM related to the task's description and expected_output,
@@ -135,3 +168,87 @@ class ContextualMemory:
             f"- {result['content']}" for result in external_memories
         )
         return f"External memories:\n{formatted_memories}"
+
+    async def _afetch_stm_context(self, query: str) -> str:
+        """Fetch recent relevant insights from STM asynchronously.
+
+        Args:
+            query: The search query.
+
+        Returns:
+            Formatted insights as bullet points, or empty string if none found.
+        """
+        if self.stm is None:
+            return ""
+
+        stm_results = await self.stm.asearch(query)
+        formatted_results = "\n".join(
+            [f"- {result['content']}" for result in stm_results]
+        )
+        return f"Recent Insights:\n{formatted_results}" if stm_results else ""
+
+    async def _afetch_ltm_context(self, task: str) -> str | None:
+        """Fetch historical data from LTM asynchronously.
+
+        Args:
+            task: The task description to search for.
+
+        Returns:
+            Formatted historical data as bullet points, or None if none found.
+        """
+        if self.ltm is None:
+            return ""
+
+        ltm_results = await self.ltm.asearch(task, latest_n=2)
+        if not ltm_results:
+            return None
+
+        formatted_results = [
+            suggestion
+            for result in ltm_results
+            for suggestion in result["metadata"]["suggestions"]
+        ]
+        formatted_results = list(dict.fromkeys(formatted_results))
+        formatted_results = "\n".join([f"- {result}" for result in formatted_results])  # type: ignore # Incompatible types in assignment (expression has type "str", variable has type "list[str]")
+
+        return f"Historical Data:\n{formatted_results}" if ltm_results else ""
+
+    async def _afetch_entity_context(self, query: str) -> str:
+        """Fetch relevant entity information asynchronously.
+
+        Args:
+            query: The search query.
+
+        Returns:
+            Formatted entity information as bullet points, or empty string if none found.
+        """
+        if self.em is None:
+            return ""
+
+        em_results = await self.em.asearch(query)
+        formatted_results = "\n".join(
+            [f"- {result['content']}" for result in em_results]
+        )
+        return f"Entities:\n{formatted_results}" if em_results else ""
+
+    async def _afetch_external_context(self, query: str) -> str:
+        """Fetch relevant information from External Memory asynchronously.
+
+        Args:
+            query: The search query.
+
+        Returns:
+            Formatted information as bullet points, or empty string if none found.
+        """
+        if self.exm is None:
+            return ""
+
+        external_memories = await self.exm.asearch(query)
+
+        if not external_memories:
+            return ""
+
+        formatted_memories = "\n".join(
+            f"- {result['content']}" for result in external_memories
+        )
+        return f"External memories:\n{formatted_memories}"

lib/crewai/src/crewai/memory/entity/entity_memory.py

@@ -26,7 +26,13 @@ class EntityMemory(Memory):
 
     _memory_provider: str | None = PrivateAttr()
 
-    def __init__(self, crew=None, embedder_config=None, storage=None, path=None):
+    def __init__(
+        self,
+        crew: Any = None,
+        embedder_config: Any = None,
+        storage: Any = None,
+        path: str | None = None,
+    ) -> None:
         memory_provider = None
         if embedder_config and isinstance(embedder_config, dict):
             memory_provider = embedder_config.get("provider")
@@ -43,7 +49,7 @@ class EntityMemory(Memory):
                 if embedder_config and isinstance(embedder_config, dict)
                 else None
             )
-            storage = Mem0Storage(type="short_term", crew=crew, config=config)
+            storage = Mem0Storage(type="short_term", crew=crew, config=config)  # type: ignore[no-untyped-call]
         else:
             storage = (
                 storage
@@ -170,7 +176,17 @@ class EntityMemory(Memory):
         query: str,
         limit: int = 5,
         score_threshold: float = 0.6,
-    ):
+    ) -> list[Any]:
+        """Search entity memory for relevant entries.
+
+        Args:
+            query: The search query.
+            limit: Maximum number of results to return.
+            score_threshold: Minimum similarity score for results.
+
+        Returns:
+            List of matching memory entries.
+        """
         crewai_event_bus.emit(
             self,
             event=MemoryQueryStartedEvent(
@@ -217,6 +233,168 @@ class EntityMemory(Memory):
             )
             raise
 
+    async def asave(
+        self,
+        value: EntityMemoryItem | list[EntityMemoryItem],
+        metadata: dict[str, Any] | None = None,
+    ) -> None:
+        """Save entity items asynchronously.
+
+        Args:
+            value: Single EntityMemoryItem or list of EntityMemoryItems to save.
+            metadata: Optional metadata dict (not used, for signature compatibility).
+        """
+        if not value:
+            return
+
+        items = value if isinstance(value, list) else [value]
+        is_batch = len(items) > 1
+
+        metadata = {"entity_count": len(items)} if is_batch else items[0].metadata
+        crewai_event_bus.emit(
+            self,
+            event=MemorySaveStartedEvent(
+                metadata=metadata,
+                source_type="entity_memory",
+                from_agent=self.agent,
+                from_task=self.task,
+            ),
+        )
+
+        start_time = time.time()
+        saved_count = 0
+        errors: list[str | None] = []
+
+        async def save_single_item(item: EntityMemoryItem) -> tuple[bool, str | None]:
+            """Save a single item asynchronously."""
+            try:
+                if self._memory_provider == "mem0":
+                    data = f"""
+                    Remember details about the following entity:
+                    Name: {item.name}
+                    Type: {item.type}
+                    Entity Description: {item.description}
+                    """
+                else:
+                    data = f"{item.name}({item.type}): {item.description}"
+
+                await super(EntityMemory, self).asave(data, item.metadata)
+                return True, None
+            except Exception as e:
+                return False, f"{item.name}: {e!s}"
+
+        try:
+            for item in items:
+                success, error = await save_single_item(item)
+                if success:
+                    saved_count += 1
+                else:
+                    errors.append(error)
+
+            if is_batch:
+                emit_value = f"Saved {saved_count} entities"
+                metadata = {"entity_count": saved_count, "errors": errors}
+            else:
+                emit_value = f"{items[0].name}({items[0].type}): {items[0].description}"
+                metadata = items[0].metadata
+
+            crewai_event_bus.emit(
+                self,
+                event=MemorySaveCompletedEvent(
+                    value=emit_value,
+                    metadata=metadata,
+                    save_time_ms=(time.time() - start_time) * 1000,
+                    source_type="entity_memory",
+                    from_agent=self.agent,
+                    from_task=self.task,
+                ),
+            )
+
+            if errors:
+                raise Exception(
+                    f"Partial save: {len(errors)} failed out of {len(items)}"
+                )
+
+        except Exception as e:
+            fail_metadata = (
+                {"entity_count": len(items), "saved": saved_count}
+                if is_batch
+                else items[0].metadata
+            )
+            crewai_event_bus.emit(
+                self,
+                event=MemorySaveFailedEvent(
+                    metadata=fail_metadata,
+                    error=str(e),
+                    source_type="entity_memory",
+                    from_agent=self.agent,
+                    from_task=self.task,
+                ),
+            )
+            raise
+
+    async def asearch(
+        self,
+        query: str,
+        limit: int = 5,
+        score_threshold: float = 0.6,
+    ) -> list[Any]:
+        """Search entity memory asynchronously.
+
+        Args:
+            query: The search query.
+            limit: Maximum number of results to return.
+            score_threshold: Minimum similarity score for results.
+
+        Returns:
+            List of matching memory entries.
+        """
+        crewai_event_bus.emit(
+            self,
+            event=MemoryQueryStartedEvent(
+                query=query,
+                limit=limit,
+                score_threshold=score_threshold,
+                source_type="entity_memory",
+                from_agent=self.agent,
+                from_task=self.task,
+            ),
+        )
+
+        start_time = time.time()
+        try:
+            results = await super().asearch(
+                query=query, limit=limit, score_threshold=score_threshold
+            )
+
+            crewai_event_bus.emit(
+                self,
+                event=MemoryQueryCompletedEvent(
+                    query=query,
+                    results=results,
+                    limit=limit,
+                    score_threshold=score_threshold,
+                    query_time_ms=(time.time() - start_time) * 1000,
+                    source_type="entity_memory",
+                    from_agent=self.agent,
+                    from_task=self.task,
+                ),
+            )
+
+            return results
+        except Exception as e:
+            crewai_event_bus.emit(
+                self,
+                event=MemoryQueryFailedEvent(
+                    query=query,
+                    limit=limit,
+                    score_threshold=score_threshold,
+                    error=str(e),
+                    source_type="entity_memory",
+                ),
+            )
+            raise
+
     def reset(self) -> None:
         try:
             self.storage.reset()

lib/crewai/src/crewai/memory/external/external_memory.py

@@ -30,7 +30,7 @@ class ExternalMemory(Memory):
     def _configure_mem0(crew: Any, config: dict[str, Any]) -> Mem0Storage:
         from crewai.memory.storage.mem0_storage import Mem0Storage
 
-        return Mem0Storage(type="external", crew=crew, config=config)
+        return Mem0Storage(type="external", crew=crew, config=config)  # type: ignore[no-untyped-call]
 
     @staticmethod
     def external_supported_storages() -> dict[str, Any]:
@@ -53,7 +53,10 @@ class ExternalMemory(Memory):
         if provider not in supported_storages:
             raise ValueError(f"Provider {provider} not supported")
 
-        return supported_storages[provider](crew, embedder_config.get("config", {}))
+        storage: Storage = supported_storages[provider](
+            crew, embedder_config.get("config", {})
+        )
+        return storage
 
     def save(
         self,
@@ -111,7 +114,17 @@ class ExternalMemory(Memory):
         query: str,
         limit: int = 5,
         score_threshold: float = 0.6,
-    ):
+    ) -> list[Any]:
+        """Search external memory for relevant entries.
+
+        Args:
+            query: The search query.
+            limit: Maximum number of results to return.
+            score_threshold: Minimum similarity score for results.
+
+        Returns:
+            List of matching memory entries.
+        """
         crewai_event_bus.emit(
             self,
             event=MemoryQueryStartedEvent(
@@ -158,6 +171,124 @@ class ExternalMemory(Memory):
             )
             raise
 
+    async def asave(
+        self,
+        value: Any,
+        metadata: dict[str, Any] | None = None,
+    ) -> None:
+        """Save a value to external memory asynchronously.
+
+        Args:
+            value: The value to save.
+            metadata: Optional metadata to associate with the value.
+        """
+        crewai_event_bus.emit(
+            self,
+            event=MemorySaveStartedEvent(
+                value=value,
+                metadata=metadata,
+                source_type="external_memory",
+                from_agent=self.agent,
+                from_task=self.task,
+            ),
+        )
+
+        start_time = time.time()
+        try:
+            item = ExternalMemoryItem(
+                value=value,
+                metadata=metadata,
+                agent=self.agent.role if self.agent else None,
+            )
+            await super().asave(value=item.value, metadata=item.metadata)
+
+            crewai_event_bus.emit(
+                self,
+                event=MemorySaveCompletedEvent(
+                    value=value,
+                    metadata=metadata,
+                    save_time_ms=(time.time() - start_time) * 1000,
+                    source_type="external_memory",
+                    from_agent=self.agent,
+                    from_task=self.task,
+                ),
+            )
+        except Exception as e:
+            crewai_event_bus.emit(
+                self,
+                event=MemorySaveFailedEvent(
+                    value=value,
+                    metadata=metadata,
+                    error=str(e),
+                    source_type="external_memory",
+                    from_agent=self.agent,
+                    from_task=self.task,
+                ),
+            )
+            raise
+
+    async def asearch(
+        self,
+        query: str,
+        limit: int = 5,
+        score_threshold: float = 0.6,
+    ) -> list[Any]:
+        """Search external memory asynchronously.
+
+        Args:
+            query: The search query.
+            limit: Maximum number of results to return.
+            score_threshold: Minimum similarity score for results.
+
+        Returns:
+            List of matching memory entries.
+        """
+        crewai_event_bus.emit(
+            self,
+            event=MemoryQueryStartedEvent(
+                query=query,
+                limit=limit,
+                score_threshold=score_threshold,
+                source_type="external_memory",
+                from_agent=self.agent,
+                from_task=self.task,
+            ),
+        )
+
+        start_time = time.time()
+        try:
+            results = await super().asearch(
+                query=query, limit=limit, score_threshold=score_threshold
+            )
+
+            crewai_event_bus.emit(
+                self,
+                event=MemoryQueryCompletedEvent(
+                    query=query,
+                    results=results,
+                    limit=limit,
+                    score_threshold=score_threshold,
+                    query_time_ms=(time.time() - start_time) * 1000,
+                    source_type="external_memory",
+                    from_agent=self.agent,
+                    from_task=self.task,
+                ),
+            )
+
+            return results
+        except Exception as e:
+            crewai_event_bus.emit(
+                self,
+                event=MemoryQueryFailedEvent(
+                    query=query,
+                    limit=limit,
+                    score_threshold=score_threshold,
+                    error=str(e),
+                    source_type="external_memory",
+                ),
+            )
+            raise
+
     def reset(self) -> None:
         self.storage.reset()
 

lib/crewai/src/crewai/memory/long_term/long_term_memory.py

@@ -24,7 +24,11 @@ class LongTermMemory(Memory):
     LongTermMemoryItem instances.
     """
 
-    def __init__(self, storage=None, path=None):
+    def __init__(
+        self,
+        storage: LTMSQLiteStorage | None = None,
+        path: str | None = None,
+    ) -> None:
         if not storage:
             storage = LTMSQLiteStorage(db_path=path) if path else LTMSQLiteStorage()
         super().__init__(storage=storage)
@@ -48,7 +52,7 @@ class LongTermMemory(Memory):
             metadata.update(
                 {"agent": item.agent, "expected_output": item.expected_output}
             )
-            self.storage.save(  # type: ignore # BUG?: Unexpected keyword argument "task_description","score","datetime" for "save" of "Storage"
+            self.storage.save(
                 task_description=item.task,
                 score=metadata["quality"],
                 metadata=metadata,
@@ -80,11 +84,20 @@ class LongTermMemory(Memory):
             )
             raise
 
-    def search(  # type: ignore # signature of "search" incompatible with supertype "Memory"
+    def search(  # type: ignore[override]
         self,
         task: str,
         latest_n: int = 3,
-    ) -> list[dict[str, Any]]:  # type: ignore # signature of "search" incompatible with supertype "Memory"
+    ) -> list[dict[str, Any]]:
+        """Search long-term memory for relevant entries.
+
+        Args:
+            task: The task description to search for.
+            latest_n: Maximum number of results to return.
+
+        Returns:
+            List of matching memory entries.
+        """
         crewai_event_bus.emit(
             self,
             event=MemoryQueryStartedEvent(
@@ -98,7 +111,7 @@ class LongTermMemory(Memory):
 
         start_time = time.time()
         try:
-            results = self.storage.load(task, latest_n)  # type: ignore # BUG?: "Storage" has no attribute "load"
+            results = self.storage.load(task, latest_n)
 
             crewai_event_bus.emit(
                 self,
@@ -113,7 +126,118 @@ class LongTermMemory(Memory):
                 ),
             )
 
-            return results
+            return results or []
+        except Exception as e:
+            crewai_event_bus.emit(
+                self,
+                event=MemoryQueryFailedEvent(
+                    query=task,
+                    limit=latest_n,
+                    error=str(e),
+                    source_type="long_term_memory",
+                ),
+            )
+            raise
+
+    async def asave(self, item: LongTermMemoryItem) -> None:  # type: ignore[override]
+        """Save an item to long-term memory asynchronously.
+
+        Args:
+            item: The LongTermMemoryItem to save.
+        """
+        crewai_event_bus.emit(
+            self,
+            event=MemorySaveStartedEvent(
+                value=item.task,
+                metadata=item.metadata,
+                agent_role=item.agent,
+                source_type="long_term_memory",
+                from_agent=self.agent,
+                from_task=self.task,
+            ),
+        )
+
+        start_time = time.time()
+        try:
+            metadata = item.metadata
+            metadata.update(
+                {"agent": item.agent, "expected_output": item.expected_output}
+            )
+            await self.storage.asave(
+                task_description=item.task,
+                score=metadata["quality"],
+                metadata=metadata,
+                datetime=item.datetime,
+            )
+
+            crewai_event_bus.emit(
+                self,
+                event=MemorySaveCompletedEvent(
+                    value=item.task,
+                    metadata=item.metadata,
+                    agent_role=item.agent,
+                    save_time_ms=(time.time() - start_time) * 1000,
+                    source_type="long_term_memory",
+                    from_agent=self.agent,
+                    from_task=self.task,
+                ),
+            )
+        except Exception as e:
+            crewai_event_bus.emit(
+                self,
+                event=MemorySaveFailedEvent(
+                    value=item.task,
+                    metadata=item.metadata,
+                    agent_role=item.agent,
+                    error=str(e),
+                    source_type="long_term_memory",
+                ),
+            )
+            raise
+
+    async def asearch(  # type: ignore[override]
+        self,
+        task: str,
+        latest_n: int = 3,
+    ) -> list[dict[str, Any]]:
+        """Search long-term memory asynchronously.
+
+        Args:
+            task: The task description to search for.
+            latest_n: Maximum number of results to return.
+
+        Returns:
+            List of matching memory entries.
+        """
+        crewai_event_bus.emit(
+            self,
+            event=MemoryQueryStartedEvent(
+                query=task,
+                limit=latest_n,
+                source_type="long_term_memory",
+                from_agent=self.agent,
+                from_task=self.task,
+            ),
+        )
+
+        start_time = time.time()
+        try:
+            results = await self.storage.aload(task, latest_n)
+
+            crewai_event_bus.emit(
+                self,
+                event=MemoryQueryCompletedEvent(
+                    query=task,
+                    results=results,
+                    limit=latest_n,
+                    query_time_ms=(time.time() - start_time) * 1000,
+                    source_type="long_term_memory",
+                    from_agent=self.agent,
+                    from_task=self.task,
+                ),
+            )
+
+            return results or []
         except Exception as e:
             crewai_event_bus.emit(
                 self,
@@ -127,4 +251,5 @@ class LongTermMemory(Memory):
             raise
 
     def reset(self) -> None:
+        """Reset long-term memory."""
         self.storage.reset()

lib/crewai/src/crewai/memory/memory.py

@@ -13,9 +13,7 @@ if TYPE_CHECKING:
 
 
 class Memory(BaseModel):
-    """
-    Base class for memory, now supporting agent tags and generic metadata.
-    """
+    """Base class for memory, supporting agent tags and generic metadata."""
 
     embedder_config: EmbedderConfig | dict[str, Any] | None = None
     crew: Any | None = None
@@ -52,20 +50,72 @@ class Memory(BaseModel):
         value: Any,
         metadata: dict[str, Any] | None = None,
     ) -> None:
-        metadata = metadata or {}
+        """Save a value to memory.
 
+        Args:
+            value: The value to save.
+            metadata: Optional metadata to associate with the value.
+        """
+        metadata = metadata or {}
         self.storage.save(value, metadata)
 
+    async def asave(
+        self,
+        value: Any,
+        metadata: dict[str, Any] | None = None,
+    ) -> None:
+        """Save a value to memory asynchronously.
+
+        Args:
+            value: The value to save.
+            metadata: Optional metadata to associate with the value.
+        """
+        metadata = metadata or {}
+        await self.storage.asave(value, metadata)
+
     def search(
         self,
         query: str,
         limit: int = 5,
         score_threshold: float = 0.6,
     ) -> list[Any]:
-        return self.storage.search(
+        """Search memory for relevant entries.
+
+        Args:
+            query: The search query.
+            limit: Maximum number of results to return.
+            score_threshold: Minimum similarity score for results.
+
+        Returns:
+            List of matching memory entries.
+        """
+        results: list[Any] = self.storage.search(
             query=query, limit=limit, score_threshold=score_threshold
         )
+        return results
+
+    async def asearch(
+        self,
+        query: str,
+        limit: int = 5,
+        score_threshold: float = 0.6,
+    ) -> list[Any]:
+        """Search memory for relevant entries asynchronously.
+
+        Args:
+            query: The search query.
+            limit: Maximum number of results to return.
+            score_threshold: Minimum similarity score for results.
+
+        Returns:
+            List of matching memory entries.
+        """
+        results: list[Any] = await self.storage.asearch(
+            query=query, limit=limit, score_threshold=score_threshold
+        )
+        return results
 
     def set_crew(self, crew: Any) -> Memory:
+        """Set the crew for this memory instance."""
         self.crew = crew
         return self

lib/crewai/src/crewai/memory/short_term/short_term_memory.py

@@ -30,7 +30,13 @@ class ShortTermMemory(Memory):
 
     _memory_provider: str | None = PrivateAttr()
 
-    def __init__(self, crew=None, embedder_config=None, storage=None, path=None):
+    def __init__(
+        self,
+        crew: Any = None,
+        embedder_config: Any = None,
+        storage: Any = None,
+        path: str | None = None,
+    ) -> None:
         memory_provider = None
         if embedder_config and isinstance(embedder_config, dict):
             memory_provider = embedder_config.get("provider")
@@ -47,7 +53,7 @@ class ShortTermMemory(Memory):
                 if embedder_config and isinstance(embedder_config, dict)
                 else None
             )
-            storage = Mem0Storage(type="short_term", crew=crew, config=config)
+            storage = Mem0Storage(type="short_term", crew=crew, config=config)  # type: ignore[no-untyped-call]
         else:
             storage = (
                 storage
@@ -123,7 +129,17 @@ class ShortTermMemory(Memory):
         query: str,
         limit: int = 5,
         score_threshold: float = 0.6,
-    ):
+    ) -> list[Any]:
+        """Search short-term memory for relevant entries.
+
+        Args:
+            query: The search query.
+            limit: Maximum number of results to return.
+            score_threshold: Minimum similarity score for results.
+
+        Returns:
+            List of matching memory entries.
+        """
         crewai_event_bus.emit(
             self,
             event=MemoryQueryStartedEvent(
@@ -140,7 +156,7 @@ class ShortTermMemory(Memory):
         try:
             results = self.storage.search(
                 query=query, limit=limit, score_threshold=score_threshold
-            )  # type: ignore # BUG? The reference is to the parent class, but the parent class does not have this parameters
+            )
 
             crewai_event_bus.emit(
                 self,
@@ -156,7 +172,130 @@ class ShortTermMemory(Memory):
                 ),
             )
 
-            return results
+            return list(results)
+        except Exception as e:
+            crewai_event_bus.emit(
+                self,
+                event=MemoryQueryFailedEvent(
+                    query=query,
+                    limit=limit,
+                    score_threshold=score_threshold,
+                    error=str(e),
+                    source_type="short_term_memory",
+                ),
+            )
+            raise
+
+    async def asave(
+        self,
+        value: Any,
+        metadata: dict[str, Any] | None = None,
+    ) -> None:
+        """Save a value to short-term memory asynchronously.
+
+        Args:
+            value: The value to save.
+            metadata: Optional metadata to associate with the value.
+        """
+        crewai_event_bus.emit(
+            self,
+            event=MemorySaveStartedEvent(
+                value=value,
+                metadata=metadata,
+                source_type="short_term_memory",
+                from_agent=self.agent,
+                from_task=self.task,
+            ),
+        )
+
+        start_time = time.time()
+        try:
+            item = ShortTermMemoryItem(
+                data=value,
+                metadata=metadata,
+                agent=self.agent.role if self.agent else None,
+            )
+            if self._memory_provider == "mem0":
+                item.data = (
+                    f"Remember the following insights from Agent run: {item.data}"
+                )
+
+            await super().asave(value=item.data, metadata=item.metadata)
+
+            crewai_event_bus.emit(
+                self,
+                event=MemorySaveCompletedEvent(
+                    value=value,
+                    metadata=metadata,
+                    save_time_ms=(time.time() - start_time) * 1000,
+                    source_type="short_term_memory",
+                    from_agent=self.agent,
+                    from_task=self.task,
+                ),
+            )
+        except Exception as e:
+            crewai_event_bus.emit(
+                self,
+                event=MemorySaveFailedEvent(
+                    value=value,
+                    metadata=metadata,
+                    error=str(e),
+                    source_type="short_term_memory",
+                    from_agent=self.agent,
+                    from_task=self.task,
+                ),
+            )
+            raise
+
+    async def asearch(
+        self,
+        query: str,
+        limit: int = 5,
+        score_threshold: float = 0.6,
+    ) -> list[Any]:
+        """Search short-term memory asynchronously.
+
+        Args:
+            query: The search query.
+            limit: Maximum number of results to return.
+            score_threshold: Minimum similarity score for results.
+
+        Returns:
+            List of matching memory entries.
+        """
+        crewai_event_bus.emit(
+            self,
+            event=MemoryQueryStartedEvent(
+                query=query,
+                limit=limit,
+                score_threshold=score_threshold,
+                source_type="short_term_memory",
+                from_agent=self.agent,
+                from_task=self.task,
+            ),
+        )
+
+        start_time = time.time()
+        try:
+            results = await self.storage.asearch(
+                query=query, limit=limit, score_threshold=score_threshold
+            )
+
+            crewai_event_bus.emit(
+                self,
+                event=MemoryQueryCompletedEvent(
+                    query=query,
+                    results=results,
+                    limit=limit,
+                    score_threshold=score_threshold,
+                    query_time_ms=(time.time() - start_time) * 1000,
+                    source_type="short_term_memory",
+                    from_agent=self.agent,
+                    from_task=self.task,
+                ),
+            )
+
+            return list(results)
         except Exception as e:
             crewai_event_bus.emit(
                 self,

lib/crewai/src/crewai/memory/storage/ltm_sqlite_storage.py

@@ -3,29 +3,30 @@ from pathlib import Path
 import sqlite3
 from typing import Any
 
+import aiosqlite
+
 from crewai.utilities import Printer
 from crewai.utilities.paths import db_storage_path
 
 
 class LTMSQLiteStorage:
-    """
-    An updated SQLite storage class for LTM data storage.
-    """
+    """SQLite storage class for long-term memory data."""
 
     def __init__(self, db_path: str | None = None) -> None:
+        """Initialize the SQLite storage.
+
+        Args:
+            db_path: Optional path to the database file.
+        """
         if db_path is None:
-            # Get the parent directory of the default db path and create our db file there
             db_path = str(Path(db_storage_path()) / "long_term_memory_storage.db")
         self.db_path = db_path
         self._printer: Printer = Printer()
-        # Ensure parent directory exists
         Path(self.db_path).parent.mkdir(parents=True, exist_ok=True)
         self._initialize_db()
 
-    def _initialize_db(self):
-        """
-        Initializes the SQLite database and creates LTM table
-        """
+    def _initialize_db(self) -> None:
+        """Initialize the SQLite database and create LTM table."""
         try:
             with sqlite3.connect(self.db_path) as conn:
                 cursor = conn.cursor()
@@ -106,9 +107,7 @@ class LTMSQLiteStorage:
             )
         return None
 
-    def reset(
-        self,
-    ) -> None:
+    def reset(self) -> None:
         """Resets the LTM table with error handling."""
         try:
             with sqlite3.connect(self.db_path) as conn:
@@ -121,4 +120,87 @@ class LTMSQLiteStorage:
                 content=f"MEMORY ERROR: An error occurred while deleting all rows in LTM: {e}",
                 color="red",
             )
-        return
+
+    async def asave(
+        self,
+        task_description: str,
+        metadata: dict[str, Any],
+        datetime: str,
+        score: int | float,
+    ) -> None:
+        """Save data to the LTM table asynchronously.
+
+        Args:
+            task_description: Description of the task.
+            metadata: Metadata associated with the memory.
+            datetime: Timestamp of the memory.
+            score: Quality score of the memory.
+        """
+        try:
+            async with aiosqlite.connect(self.db_path) as conn:
+                await conn.execute(
+                    """
+                    INSERT INTO long_term_memories (task_description, metadata, datetime, score)
+                    VALUES (?, ?, ?, ?)
+                    """,
+                    (task_description, json.dumps(metadata), datetime, score),
+                )
+                await conn.commit()
+        except aiosqlite.Error as e:
+            self._printer.print(
+                content=f"MEMORY ERROR: An error occurred while saving to LTM: {e}",
+                color="red",
+            )
+
+    async def aload(
+        self, task_description: str, latest_n: int
+    ) -> list[dict[str, Any]] | None:
+        """Query the LTM table by task description asynchronously.
+
+        Args:
+            task_description: Description of the task to search for.
+            latest_n: Maximum number of results to return.
+
+        Returns:
+            List of matching memory entries or None if error occurs.
+        """
+        try:
+            async with aiosqlite.connect(self.db_path) as conn:
+                cursor = await conn.execute(
+                    f"""
+                    SELECT metadata, datetime, score
+                    FROM long_term_memories
+                    WHERE task_description = ?
+                    ORDER BY datetime DESC, score ASC
+                    LIMIT {latest_n}
+                    """,  # nosec # noqa: S608
+                    (task_description,),
+                )
+                rows = await cursor.fetchall()
+                if rows:
+                    return [
+                        {
+                            "metadata": json.loads(row[0]),
+                            "datetime": row[1],
+                            "score": row[2],
+                        }
+                        for row in rows
+                    ]
+        except aiosqlite.Error as e:
+            self._printer.print(
+                content=f"MEMORY ERROR: An error occurred while querying LTM: {e}",
+                color="red",
+            )
+        return None
+
+    async def areset(self) -> None:
+        """Reset the LTM table asynchronously."""
+        try:
+            async with aiosqlite.connect(self.db_path) as conn:
+                await conn.execute("DELETE FROM long_term_memories")
+                await conn.commit()
+        except aiosqlite.Error as e:
+            self._printer.print(
+                content=f"MEMORY ERROR: An error occurred while deleting all rows in LTM: {e}",
+                color="red",
+            )

lib/crewai/src/crewai/memory/storage/rag_storage.py

@@ -129,6 +129,12 @@ class RAGStorage(BaseRAGStorage):
         return f"{base_path}/{file_name}"
 
     def save(self, value: Any, metadata: dict[str, Any]) -> None:
+        """Save a value to storage.
+
+        Args:
+            value: The value to save.
+            metadata: Metadata to associate with the value.
+        """
         try:
             client = self._get_client()
             collection_name = (
@@ -167,6 +173,51 @@ class RAGStorage(BaseRAGStorage):
                 f"Error during {self.type} save: {e!s}\n{traceback.format_exc()}"
             )
 
+    async def asave(self, value: Any, metadata: dict[str, Any]) -> None:
+        """Save a value to storage asynchronously.
+
+        Args:
+            value: The value to save.
+            metadata: Metadata to associate with the value.
+        """
+        try:
+            client = self._get_client()
+            collection_name = (
+                f"memory_{self.type}_{self.agents}"
+                if self.agents
+                else f"memory_{self.type}"
+            )
+            await client.aget_or_create_collection(collection_name=collection_name)
+
+            document: BaseRecord = {"content": value}
+            if metadata:
+                document["metadata"] = metadata
+
+            batch_size = None
+            if (
+                self.embedder_config
+                and isinstance(self.embedder_config, dict)
+                and "config" in self.embedder_config
+            ):
+                nested_config = self.embedder_config["config"]
+                if isinstance(nested_config, dict):
+                    batch_size = nested_config.get("batch_size")
+
+            if batch_size is not None:
+                await client.aadd_documents(
+                    collection_name=collection_name,
+                    documents=[document],
+                    batch_size=cast(int, batch_size),
+                )
+            else:
+                await client.aadd_documents(
+                    collection_name=collection_name, documents=[document]
+                )
+        except Exception as e:
+            logging.error(
+                f"Error during {self.type} async save: {e!s}\n{traceback.format_exc()}"
+            )
+
     def search(
         self,
         query: str,
@@ -174,6 +225,17 @@ class RAGStorage(BaseRAGStorage):
         filter: dict[str, Any] | None = None,
         score_threshold: float = 0.6,
     ) -> list[Any]:
+        """Search for matching entries in storage.
+
+        Args:
+            query: The search query.
+            limit: Maximum number of results to return.
+            filter: Optional metadata filter.
+            score_threshold: Minimum similarity score for results.
+
+        Returns:
+            List of matching entries.
+        """
         try:
             client = self._get_client()
             collection_name = (
@@ -194,6 +256,44 @@ class RAGStorage(BaseRAGStorage):
             )
             return []
 
+    async def asearch(
+        self,
+        query: str,
+        limit: int = 5,
+        filter: dict[str, Any] | None = None,
+        score_threshold: float = 0.6,
+    ) -> list[Any]:
+        """Search for matching entries in storage asynchronously.
+
+        Args:
+            query: The search query.
+            limit: Maximum number of results to return.
+            filter: Optional metadata filter.
+            score_threshold: Minimum similarity score for results.
+
+        Returns:
+            List of matching entries.
+        """
+        try:
+            client = self._get_client()
+            collection_name = (
+                f"memory_{self.type}_{self.agents}"
+                if self.agents
+                else f"memory_{self.type}"
+            )
+            return await client.asearch(
+                collection_name=collection_name,
+                query=query,
+                limit=limit,
+                metadata_filter=filter,
+                score_threshold=score_threshold,
+            )
+        except Exception as e:
+            logging.error(
+                f"Error during {self.type} async search: {e!s}\n{traceback.format_exc()}"
+            )
+            return []
+
     def reset(self) -> None:
         try:
             client = self._get_client()

lib/crewai/src/crewai/rag/embeddings/providers/huggingface/huggingface_provider.py

@@ -1,21 +1,35 @@
 """HuggingFace embeddings provider."""
 
 from chromadb.utils.embedding_functions.huggingface_embedding_function import (
-    HuggingFaceEmbeddingServer,
+    HuggingFaceEmbeddingFunction,
 )
 from pydantic import AliasChoices, Field
 
 from crewai.rag.core.base_embeddings_provider import BaseEmbeddingsProvider
 
 
-class HuggingFaceProvider(BaseEmbeddingsProvider[HuggingFaceEmbeddingServer]):
-    """HuggingFace embeddings provider."""
+class HuggingFaceProvider(BaseEmbeddingsProvider[HuggingFaceEmbeddingFunction]):
+    """HuggingFace embeddings provider for the HuggingFace Inference API."""
 
-    embedding_callable: type[HuggingFaceEmbeddingServer] = Field(
-        default=HuggingFaceEmbeddingServer,
+    embedding_callable: type[HuggingFaceEmbeddingFunction] = Field(
+        default=HuggingFaceEmbeddingFunction,
         description="HuggingFace embedding function class",
     )
-    url: str = Field(
-        description="HuggingFace API URL",
-        validation_alias=AliasChoices("EMBEDDINGS_HUGGINGFACE_URL", "HUGGINGFACE_URL"),
+    api_key: str | None = Field(
+        default=None,
+        description="HuggingFace API key",
+        validation_alias=AliasChoices(
+            "EMBEDDINGS_HUGGINGFACE_API_KEY",
+            "HUGGINGFACE_API_KEY",
+            "HF_TOKEN",
+        ),
+    )
+    model_name: str = Field(
+        default="sentence-transformers/all-MiniLM-L6-v2",
+        description="Model name to use for embeddings",
+        validation_alias=AliasChoices(
+            "EMBEDDINGS_HUGGINGFACE_MODEL_NAME",
+            "HUGGINGFACE_MODEL_NAME",
+            "model",
+        ),
     )

lib/crewai/src/crewai/rag/embeddings/providers/huggingface/types.py

@@ -1,6 +1,6 @@
 """Type definitions for HuggingFace embedding providers."""
 
-from typing import Literal
+from typing import Annotated, Literal
 
 from typing_extensions import Required, TypedDict
 
@@ -8,7 +8,11 @@ from typing_extensions import Required, TypedDict
 class HuggingFaceProviderConfig(TypedDict, total=False):
     """Configuration for HuggingFace provider."""
 
-    url: str
+    api_key: str
+    model: Annotated[
+        str, "sentence-transformers/all-MiniLM-L6-v2"
+    ]  # alias for model_name for backward compat
+    model_name: Annotated[str, "sentence-transformers/all-MiniLM-L6-v2"]
 
 
 class HuggingFaceProviderSpec(TypedDict, total=False):

lib/crewai/src/crewai/task.py

@@ -494,8 +494,112 @@ class Task(BaseModel):
         future: Future[TaskOutput],
     ) -> None:
         """Execute the task asynchronously with context handling."""
-        result = self._execute_core(agent, context, tools)
-        future.set_result(result)
+        try:
+          result = self._execute_core(agent, context, tools)
+          future.set_result(result)
+        except Exception as e:
+          future.set_exception(e)
+
+    async def aexecute_sync(
+        self,
+        agent: BaseAgent | None = None,
+        context: str | None = None,
+        tools: list[BaseTool] | None = None,
+    ) -> TaskOutput:
+        """Execute the task asynchronously using native async/await."""
+        return await self._aexecute_core(agent, context, tools)
+
+    async def _aexecute_core(
+        self,
+        agent: BaseAgent | None,
+        context: str | None,
+        tools: list[Any] | None,
+    ) -> TaskOutput:
+        """Run the core execution logic of the task asynchronously."""
+        try:
+            agent = agent or self.agent
+            self.agent = agent
+            if not agent:
+                raise Exception(
+                    f"The task '{self.description}' has no agent assigned, therefore it can't be executed directly and should be executed in a Crew using a specific process that support that, like hierarchical."
+                )
+
+            self.start_time = datetime.datetime.now()
+
+            self.prompt_context = context
+            tools = tools or self.tools or []
+
+            self.processed_by_agents.add(agent.role)
+            crewai_event_bus.emit(self, TaskStartedEvent(context=context, task=self))  # type: ignore[no-untyped-call]
+            result = await agent.aexecute_task(
+                task=self,
+                context=context,
+                tools=tools,
+            )
+
+            if not self._guardrails and not self._guardrail:
+                pydantic_output, json_output = self._export_output(result)
+            else:
+                pydantic_output, json_output = None, None
+
+            task_output = TaskOutput(
+                name=self.name or self.description,
+                description=self.description,
+                expected_output=self.expected_output,
+                raw=result,
+                pydantic=pydantic_output,
+                json_dict=json_output,
+                agent=agent.role,
+                output_format=self._get_output_format(),
+                messages=agent.last_messages,  # type: ignore[attr-defined]
+            )
+
+            if self._guardrails:
+                for idx, guardrail in enumerate(self._guardrails):
+                    task_output = await self._ainvoke_guardrail_function(
+                        task_output=task_output,
+                        agent=agent,
+                        tools=tools,
+                        guardrail=guardrail,
+                        guardrail_index=idx,
+                    )
+
+            if self._guardrail:
+                task_output = await self._ainvoke_guardrail_function(
+                    task_output=task_output,
+                    agent=agent,
+                    tools=tools,
+                    guardrail=self._guardrail,
+                )
+
+            self.output = task_output
+            self.end_time = datetime.datetime.now()
+
+            if self.callback:
+                self.callback(self.output)
+
+            crew = self.agent.crew  # type: ignore[union-attr]
+            if crew and crew.task_callback and crew.task_callback != self.callback:
+                crew.task_callback(self.output)
+
+            if self.output_file:
+                content = (
+                    json_output
+                    if json_output
+                    else (
+                        pydantic_output.model_dump_json() if pydantic_output else result
+                    )
+                )
+                self._save_file(content)
+            crewai_event_bus.emit(
+                self,
+                TaskCompletedEvent(output=task_output, task=self),  # type: ignore[no-untyped-call]
+            )
+            return task_output
+        except Exception as e:
+            self.end_time = datetime.datetime.now()
+            crewai_event_bus.emit(self, TaskFailedEvent(error=str(e), task=self))  # type: ignore[no-untyped-call]
+            raise e  # Re-raise the exception after emitting the event
 
     def _execute_core(
         self,
@@ -539,7 +643,7 @@ class Task(BaseModel):
                 json_dict=json_output,
                 agent=agent.role,
                 output_format=self._get_output_format(),
-                messages=agent.last_messages,
+                messages=agent.last_messages,  # type: ignore[attr-defined]
             )
 
             if self._guardrails:
@@ -950,7 +1054,103 @@ Follow these guidelines:
                 json_dict=json_output,
                 agent=agent.role,
                 output_format=self._get_output_format(),
-                messages=agent.last_messages,
+                messages=agent.last_messages,  # type: ignore[attr-defined]
+            )
+
+        return task_output
+
+    async def _ainvoke_guardrail_function(
+        self,
+        task_output: TaskOutput,
+        agent: BaseAgent,
+        tools: list[BaseTool],
+        guardrail: GuardrailCallable | None,
+        guardrail_index: int | None = None,
+    ) -> TaskOutput:
+        """Invoke the guardrail function asynchronously."""
+        if not guardrail:
+            return task_output
+
+        if guardrail_index is not None:
+            current_retry_count = self._guardrail_retry_counts.get(guardrail_index, 0)
+        else:
+            current_retry_count = self.retry_count
+
+        max_attempts = self.guardrail_max_retries + 1
+
+        for attempt in range(max_attempts):
+            guardrail_result = process_guardrail(
+                output=task_output,
+                guardrail=guardrail,
+                retry_count=current_retry_count,
+                event_source=self,
+                from_task=self,
+                from_agent=agent,
+            )
+
+            if guardrail_result.success:
+                if guardrail_result.result is None:
+                    raise Exception(
+                        "Task guardrail returned None as result. This is not allowed."
+                    )
+
+                if isinstance(guardrail_result.result, str):
+                    task_output.raw = guardrail_result.result
+                    pydantic_output, json_output = self._export_output(
+                        guardrail_result.result
+                    )
+                    task_output.pydantic = pydantic_output
+                    task_output.json_dict = json_output
+                elif isinstance(guardrail_result.result, TaskOutput):
+                    task_output = guardrail_result.result
+
+                return task_output
+
+            if attempt >= self.guardrail_max_retries:
+                guardrail_name = (
+                    f"guardrail {guardrail_index}"
+                    if guardrail_index is not None
+                    else "guardrail"
+                )
+                raise Exception(
+                    f"Task failed {guardrail_name} validation after {self.guardrail_max_retries} retries. "
+                    f"Last error: {guardrail_result.error}"
+                )
+
+            if guardrail_index is not None:
+                current_retry_count += 1
+                self._guardrail_retry_counts[guardrail_index] = current_retry_count
+            else:
+                self.retry_count += 1
+                current_retry_count = self.retry_count
+
+            context = self.i18n.errors("validation_error").format(
+                guardrail_result_error=guardrail_result.error,
+                task_output=task_output.raw,
+            )
+            printer = Printer()
+            printer.print(
+                content=f"Guardrail {guardrail_index if guardrail_index is not None else ''} blocked (attempt {attempt + 1}/{max_attempts}), retrying due to: {guardrail_result.error}\n",
+                color="yellow",
+            )
+
+            result = await agent.aexecute_task(
+                task=self,
+                context=context,
+                tools=tools,
+            )
+
+            pydantic_output, json_output = self._export_output(result)
+            task_output = TaskOutput(
+                name=self.name or self.description,
+                description=self.description,
+                expected_output=self.expected_output,
+                raw=result,
+                pydantic=pydantic_output,
+                json_dict=json_output,
+                agent=agent.role,
+                output_format=self._get_output_format(),
+                messages=agent.last_messages,  # type: ignore[attr-defined]
             )
 
         return task_output

lib/crewai/src/crewai/telemetry/telemetry.py

@@ -9,12 +9,14 @@ data is collected. Users can opt-in to share more complete data using the
 from __future__ import annotations
 
 import asyncio
+import atexit
 from collections.abc import Callable
 from importlib.metadata import version
 import json
 import logging
 import os
 import platform
+import signal
 import threading
 from typing import TYPE_CHECKING, Any
 
@@ -31,6 +33,14 @@ from opentelemetry.sdk.trace.export import (
 from opentelemetry.trace import Span
 from typing_extensions import Self
 
+from crewai.events.event_bus import crewai_event_bus
+from crewai.events.types.system_events import (
+    SigContEvent,
+    SigHupEvent,
+    SigIntEvent,
+    SigTStpEvent,
+    SigTermEvent,
+)
 from crewai.telemetry.constants import (
     CREWAI_TELEMETRY_BASE_URL,
     CREWAI_TELEMETRY_SERVICE_NAME,
@@ -121,6 +131,7 @@ class Telemetry:
             )
 
             self.provider.add_span_processor(processor)
+            self._register_shutdown_handlers()
             self.ready = True
         except Exception as e:
             if isinstance(
@@ -155,6 +166,74 @@ class Telemetry:
                 self.ready = False
                 self.trace_set = False
 
+    def _register_shutdown_handlers(self) -> None:
+        """Register handlers for graceful shutdown on process exit and signals."""
+        atexit.register(self._shutdown)
+
+        self._original_handlers: dict[int, Any] = {}
+
+        self._register_signal_handler(signal.SIGTERM, SigTermEvent, shutdown=True)
+        self._register_signal_handler(signal.SIGINT, SigIntEvent, shutdown=True)
+        if hasattr(signal, "SIGHUP"):
+            self._register_signal_handler(signal.SIGHUP, SigHupEvent, shutdown=False)
+        if hasattr(signal, "SIGTSTP"):
+            self._register_signal_handler(signal.SIGTSTP, SigTStpEvent, shutdown=False)
+        if hasattr(signal, "SIGCONT"):
+            self._register_signal_handler(signal.SIGCONT, SigContEvent, shutdown=False)
+
+    def _register_signal_handler(
+        self,
+        sig: signal.Signals,
+        event_class: type,
+        shutdown: bool = False,
+    ) -> None:
+        """Register a signal handler that emits an event.
+
+        Args:
+            sig: The signal to handle.
+            event_class: The event class to instantiate and emit.
+            shutdown: Whether to trigger shutdown on this signal.
+        """
+        try:
+            original_handler = signal.getsignal(sig)
+            self._original_handlers[sig] = original_handler
+
+            def handler(signum: int, frame: Any) -> None:
+                crewai_event_bus.emit(self, event_class())
+
+                if shutdown:
+                    self._shutdown()
+
+                if original_handler not in (signal.SIG_DFL, signal.SIG_IGN, None):
+                    if callable(original_handler):
+                        original_handler(signum, frame)
+                elif shutdown:
+                    raise SystemExit(0)
+
+            signal.signal(sig, handler)
+        except ValueError as e:
+            logger.warning(
+                f"Cannot register {sig.name} handler: not running in main thread",
+                exc_info=e,
+            )
+        except OSError as e:
+            logger.warning(f"Cannot register {sig.name} handler: {e}", exc_info=e)
+
+    def _shutdown(self) -> None:
+        """Flush and shutdown the telemetry provider on process exit.
+
+        Uses a short timeout to avoid blocking process shutdown.
+        """
+        if not self.ready:
+            return
+
+        try:
+            self.provider.force_flush(timeout_millis=5000)
+            self.provider.shutdown()
+            self.ready = False
+        except Exception as e:
+            logger.debug(f"Telemetry shutdown failed: {e}")
+
     def _safe_telemetry_operation(
         self, operation: Callable[[], Span | None]
     ) -> Span | None:
@@ -316,9 +395,7 @@ class Telemetry:
                 self._add_attribute(span, "platform_system", platform.system())
                 self._add_attribute(span, "platform_version", platform.version())
                 self._add_attribute(span, "cpus", os.cpu_count())
-                self._add_attribute(
-                    span, "crew_inputs", json.dumps(inputs) if inputs else None
-                )
+                self._add_attribute(span, "crew_inputs", json.dumps(inputs or {}))
             else:
                 self._add_attribute(
                     span,
@@ -631,9 +708,7 @@ class Telemetry:
             self._add_attribute(span, "model_name", model_name)
 
             if crew.share_crew:
-                self._add_attribute(
-                    span, "inputs", json.dumps(inputs) if inputs else None
-                )
+                self._add_attribute(span, "inputs", json.dumps(inputs or {}))
 
             close_span(span)
 
@@ -738,9 +813,7 @@ class Telemetry:
             add_crew_attributes(
                 span, crew, self._add_attribute, include_fingerprint=False
             )
-            self._add_attribute(
-                span, "crew_inputs", json.dumps(inputs) if inputs else None
-            )
+            self._add_attribute(span, "crew_inputs", json.dumps(inputs or {}))
             self._add_attribute(
                 span,
                 "crew_agents",

lib/crewai/src/crewai/tools/base_tool.py

@@ -2,9 +2,16 @@ from __future__ import annotations
 
 from abc import ABC, abstractmethod
 import asyncio
-from collections.abc import Callable
-from inspect import signature
-from typing import Any, cast, get_args, get_origin
+from collections.abc import Awaitable, Callable
+from inspect import Parameter, signature
+import json
+from typing import (
+    Any,
+    Generic,
+    ParamSpec,
+    TypeVar,
+    overload,
+)
 
 from pydantic import (
     BaseModel,
@@ -14,13 +21,28 @@ from pydantic import (
     create_model,
     field_validator,
 )
+from typing_extensions import TypeIs
 
 from crewai.tools.structured_tool import CrewStructuredTool
 from crewai.utilities.printer import Printer
+from crewai.utilities.pydantic_schema_utils import generate_model_description
 
 
 _printer = Printer()
 
+P = ParamSpec("P")
+R = TypeVar("R", covariant=True)
+
+
+def _is_async_callable(func: Callable[..., Any]) -> bool:
+    """Check if a callable is async."""
+    return asyncio.iscoroutinefunction(func)
+
+
+def _is_awaitable(value: R | Awaitable[R]) -> TypeIs[Awaitable[R]]:
+    """Type narrowing check for awaitable values."""
+    return asyncio.iscoroutine(value) or asyncio.isfuture(value)
+
 
 class EnvVar(BaseModel):
     name: str
@@ -55,7 +77,7 @@ class BaseTool(BaseModel, ABC):
         default=False, description="Flag to check if the description has been updated."
     )
 
-    cache_function: Callable = Field(
+    cache_function: Callable[..., bool] = Field(
         default=lambda _args=None, _result=None: True,
         description="Function that will be used to determine if the tool should be cached, should return a boolean. If None, the tool will be cached.",
     )
@@ -80,20 +102,40 @@ class BaseTool(BaseModel, ABC):
         if v != cls._ArgsSchemaPlaceholder:
             return v
 
-        return cast(
-            type[PydanticBaseModel],
-            type(
-                f"{cls.__name__}Schema",
-                (PydanticBaseModel,),
-                {
-                    "__annotations__": {
-                        k: v
-                        for k, v in cls._run.__annotations__.items()
-                        if k != "return"
-                    },
-                },
-            ),
-        )
+        run_sig = signature(cls._run)
+        fields: dict[str, Any] = {}
+
+        for param_name, param in run_sig.parameters.items():
+            if param_name in ("self", "return"):
+                continue
+            if param.kind in (Parameter.VAR_POSITIONAL, Parameter.VAR_KEYWORD):
+                continue
+
+            annotation = param.annotation if param.annotation != param.empty else Any
+
+            if param.default is param.empty:
+                fields[param_name] = (annotation, ...)
+            else:
+                fields[param_name] = (annotation, param.default)
+
+        if not fields:
+            arun_sig = signature(cls._arun)
+            for param_name, param in arun_sig.parameters.items():
+                if param_name in ("self", "return"):
+                    continue
+                if param.kind in (Parameter.VAR_POSITIONAL, Parameter.VAR_KEYWORD):
+                    continue
+
+                annotation = (
+                    param.annotation if param.annotation != param.empty else Any
+                )
+
+                if param.default is param.empty:
+                    fields[param_name] = (annotation, ...)
+                else:
+                    fields[param_name] = (annotation, param.default)
+
+        return create_model(f"{cls.__name__}Schema", **fields)
 
     @field_validator("max_usage_count", mode="before")
     @classmethod
@@ -123,6 +165,35 @@ class BaseTool(BaseModel, ABC):
 
         return result
 
+    async def arun(
+        self,
+        *args: Any,
+        **kwargs: Any,
+    ) -> Any:
+        """Execute the tool asynchronously.
+
+        Args:
+            *args: Positional arguments to pass to the tool.
+            **kwargs: Keyword arguments to pass to the tool.
+
+        Returns:
+            The result of the tool execution.
+        """
+        result = await self._arun(*args, **kwargs)
+        self.current_usage_count += 1
+        return result
+
+    async def _arun(
+        self,
+        *args: Any,
+        **kwargs: Any,
+    ) -> Any:
+        """Async implementation of the tool. Override for async support."""
+        raise NotImplementedError(
+            f"{self.__class__.__name__} does not implement _arun. "
+            "Override _arun for async support or use run() for sync execution."
+        )
+
     def reset_usage_count(self) -> None:
         """Reset the current usage count to zero."""
         self.current_usage_count = 0
@@ -133,7 +204,17 @@ class BaseTool(BaseModel, ABC):
         *args: Any,
         **kwargs: Any,
     ) -> Any:
-        """Here goes the actual implementation of the tool."""
+        """Sync implementation of the tool.
+
+        Subclasses must implement this method for synchronous execution.
+
+        Args:
+            *args: Positional arguments for the tool.
+            **kwargs: Keyword arguments for the tool.
+
+        Returns:
+            The result of the tool execution.
+        """
 
     def to_structured_tool(self) -> CrewStructuredTool:
         """Convert this tool to a CrewStructuredTool instance."""
@@ -164,24 +245,23 @@ class BaseTool(BaseModel, ABC):
         args_schema = getattr(tool, "args_schema", None)
 
         if args_schema is None:
-            # Infer args_schema from the function signature if not provided
             func_signature = signature(tool.func)
-            annotations = func_signature.parameters
-            args_fields: dict[str, Any] = {}
-            for name, param in annotations.items():
-                if name != "self":
-                    param_annotation = (
-                        param.annotation if param.annotation != param.empty else Any
-                    )
-                    field_info = Field(
-                        default=...,
-                        description="",
-                    )
-                    args_fields[name] = (param_annotation, field_info)
-            if args_fields:
-                args_schema = create_model(f"{tool.name}Input", **args_fields)
+            fields: dict[str, Any] = {}
+            for name, param in func_signature.parameters.items():
+                if name == "self":
+                    continue
+                if param.kind in (Parameter.VAR_POSITIONAL, Parameter.VAR_KEYWORD):
+                    continue
+                param_annotation = (
+                    param.annotation if param.annotation != param.empty else Any
+                )
+                if param.default is param.empty:
+                    fields[name] = (param_annotation, ...)
+                else:
+                    fields[name] = (param_annotation, param.default)
+            if fields:
+                args_schema = create_model(f"{tool.name}Input", **fields)
             else:
-                # Create a default schema with no fields if no parameters are found
                 args_schema = create_model(
                     f"{tool.name}Input", __base__=PydanticBaseModel
                 )
@@ -195,65 +275,118 @@ class BaseTool(BaseModel, ABC):
 
     def _set_args_schema(self) -> None:
         if self.args_schema is None:
-            class_name = f"{self.__class__.__name__}Schema"
-            self.args_schema = cast(
-                type[PydanticBaseModel],
-                type(
-                    class_name,
-                    (PydanticBaseModel,),
-                    {
-                        "__annotations__": {
-                            k: v
-                            for k, v in self._run.__annotations__.items()
-                            if k != "return"
-                        },
-                    },
-                ),
+            run_sig = signature(self._run)
+            fields: dict[str, Any] = {}
+
+            for param_name, param in run_sig.parameters.items():
+                if param_name in ("self", "return"):
+                    continue
+                if param.kind in (Parameter.VAR_POSITIONAL, Parameter.VAR_KEYWORD):
+                    continue
+
+                annotation = (
+                    param.annotation if param.annotation != param.empty else Any
+                )
+
+                if param.default is param.empty:
+                    fields[param_name] = (annotation, ...)
+                else:
+                    fields[param_name] = (annotation, param.default)
+
+            self.args_schema = create_model(
+                f"{self.__class__.__name__}Schema", **fields
             )
 
     def _generate_description(self) -> None:
-        args_schema = {
-            name: {
-                "description": field.description,
-                "type": BaseTool._get_arg_annotations(field.annotation),
-            }
-            for name, field in self.args_schema.model_fields.items()
-        }
-
-        self.description = f"Tool Name: {self.name}\nTool Arguments: {args_schema}\nTool Description: {self.description}"
-
-    @staticmethod
-    def _get_arg_annotations(annotation: type[Any] | None) -> str:
-        if annotation is None:
-            return "None"
-
-        origin = get_origin(annotation)
-        args = get_args(annotation)
-
-        if origin is None:
-            return (
-                annotation.__name__
-                if hasattr(annotation, "__name__")
-                else str(annotation)
-            )
-
-        if args:
-            args_str = ", ".join(BaseTool._get_arg_annotations(arg) for arg in args)
-            return f"{origin.__name__}[{args_str}]"
-
-        return origin.__name__
+        """Generate the tool description with a JSON schema for arguments."""
+        schema = generate_model_description(self.args_schema)
+        args_json = json.dumps(schema["json_schema"]["schema"], indent=2)
+        self.description = (
+            f"Tool Name: {self.name}\n"
+            f"Tool Arguments: {args_json}\n"
+            f"Tool Description: {self.description}"
+        )
 
 
-class Tool(BaseTool):
-    """The function that will be executed when the tool is called."""
+class Tool(BaseTool, Generic[P, R]):
+    """Tool that wraps a callable function.
 
-    func: Callable
 
-    def _run(self, *args: Any, **kwargs: Any) -> Any:
-        return self.func(*args, **kwargs)
+    Type Parameters:
+        P: ParamSpec capturing the function's parameters.
+        R: The return type of the function.
+    """
+
+    func: Callable[P, R | Awaitable[R]]
+
+    def run(self, *args: P.args, **kwargs: P.kwargs) -> R:
+        """Executes the tool synchronously.
+
+        Args:
+            *args: Positional arguments for the tool.
+            **kwargs: Keyword arguments for the tool.
+
+        Returns:
+            The result of the tool execution.
+        """
+        _printer.print(f"Using Tool: {self.name}", color="cyan")
+        result = self.func(*args, **kwargs)
+
+        if asyncio.iscoroutine(result):
+            result = asyncio.run(result)
+
+        self.current_usage_count += 1
+        return result  # type: ignore[return-value]
+
+    def _run(self, *args: P.args, **kwargs: P.kwargs) -> R:
+        """Executes the wrapped function.
+
+        Args:
+            *args: Positional arguments for the function.
+            **kwargs: Keyword arguments for the function.
+
+        Returns:
+            The result of the function execution.
+        """
+        return self.func(*args, **kwargs)  # type: ignore[return-value]
+
+    async def arun(self, *args: P.args, **kwargs: P.kwargs) -> R:
+        """Executes the tool asynchronously.
+
+        Args:
+            *args: Positional arguments for the tool.
+            **kwargs: Keyword arguments for the tool.
+
+        Returns:
+            The result of the tool execution.
+        """
+        result = await self._arun(*args, **kwargs)
+        self.current_usage_count += 1
+        return result
+
+    async def _arun(self, *args: P.args, **kwargs: P.kwargs) -> R:
+        """Executes the wrapped function asynchronously.
+
+        Args:
+            *args: Positional arguments for the function.
+            **kwargs: Keyword arguments for the function.
+
+        Returns:
+            The result of the async function execution.
+
+        Raises:
+            NotImplementedError: If the wrapped function is not async.
+        """
+        result = self.func(*args, **kwargs)
+        if _is_awaitable(result):
+            return await result
+        raise NotImplementedError(
+            f"{self.name} does not have an async function. "
+            "Use run() for sync execution or provide an async function."
+        )
 
     @classmethod
-    def from_langchain(cls, tool: Any) -> Tool:
+    def from_langchain(cls, tool: Any) -> Tool[..., Any]:
         """Create a Tool instance from a CrewStructuredTool.
 
         This method takes a CrewStructuredTool object and converts it into a
@@ -261,10 +394,10 @@ class Tool(BaseTool):
         attribute and infers the argument schema if not explicitly provided.
 
         Args:
-            tool (Any): The CrewStructuredTool object to be converted.
+            tool: The CrewStructuredTool object to be converted.
 
         Returns:
-            Tool: A new Tool instance created from the provided CrewStructuredTool.
+            A new Tool instance created from the provided CrewStructuredTool.
 
         Raises:
             ValueError: If the provided tool does not have a callable 'func' attribute.
@@ -275,24 +408,23 @@ class Tool(BaseTool):
         args_schema = getattr(tool, "args_schema", None)
 
         if args_schema is None:
-            # Infer args_schema from the function signature if not provided
             func_signature = signature(tool.func)
-            annotations = func_signature.parameters
-            args_fields: dict[str, Any] = {}
-            for name, param in annotations.items():
-                if name != "self":
-                    param_annotation = (
-                        param.annotation if param.annotation != param.empty else Any
-                    )
-                    field_info = Field(
-                        default=...,
-                        description="",
-                    )
-                    args_fields[name] = (param_annotation, field_info)
-            if args_fields:
-                args_schema = create_model(f"{tool.name}Input", **args_fields)
+            fields: dict[str, Any] = {}
+            for name, param in func_signature.parameters.items():
+                if name == "self":
+                    continue
+                if param.kind in (Parameter.VAR_POSITIONAL, Parameter.VAR_KEYWORD):
+                    continue
+                param_annotation = (
+                    param.annotation if param.annotation != param.empty else Any
+                )
+                if param.default is param.empty:
+                    fields[name] = (param_annotation, ...)
+                else:
+                    fields[name] = (param_annotation, param.default)
+            if fields:
+                args_schema = create_model(f"{tool.name}Input", **fields)
             else:
-                # Create a default schema with no fields if no parameters are found
                 args_schema = create_model(
                     f"{tool.name}Input", __base__=PydanticBaseModel
                 )
@@ -308,41 +440,92 @@ class Tool(BaseTool):
 def to_langchain(
     tools: list[BaseTool | CrewStructuredTool],
 ) -> list[CrewStructuredTool]:
+    """Convert a list of tools to CrewStructuredTool instances."""
     return [t.to_structured_tool() if isinstance(t, BaseTool) else t for t in tools]
 
 
+P2 = ParamSpec("P2")
+R2 = TypeVar("R2")
+
+
+@overload
+def tool(func: Callable[P2, R2], /) -> Tool[P2, R2]: ...
+
+
+@overload
 def tool(
-    *args, result_as_answer: bool = False, max_usage_count: int | None = None
-) -> Callable:
-    """
-    Decorator to create a tool from a function.
+    name: str,
+    /,
+    *,
+    result_as_answer: bool = ...,
+    max_usage_count: int | None = ...,
+) -> Callable[[Callable[P2, R2]], Tool[P2, R2]]: ...
+
+
+@overload
+def tool(
+    *,
+    result_as_answer: bool = ...,
+    max_usage_count: int | None = ...,
+) -> Callable[[Callable[P2, R2]], Tool[P2, R2]]: ...
+
+
+def tool(
+    *args: Callable[P2, R2] | str,
+    result_as_answer: bool = False,
+    max_usage_count: int | None = None,
+) -> Tool[P2, R2] | Callable[[Callable[P2, R2]], Tool[P2, R2]]:
+    """Decorator to create a Tool from a function.
+
+    Can be used in three ways:
+    1. @tool - decorator without arguments, uses function name
+    2. @tool("name") - decorator with custom name
+    3. @tool(result_as_answer=True) - decorator with options
 
     Args:
-        *args: Positional arguments, either the function to decorate or the tool name.
-        result_as_answer: Flag to indicate if the tool result should be used as the final agent answer.
-        max_usage_count: Maximum number of times this tool can be used. None means unlimited usage.
+        *args: Either the function to decorate or a custom tool name.
+        result_as_answer: If True, the tool result becomes the final agent answer.
+        max_usage_count: Maximum times this tool can be used. None means unlimited.
+
+    Returns:
+        A Tool instance.
+
+    Example:
+        @tool
+        def greet(name: str) -> str:
+            '''Greet someone.'''
+            return f"Hello, {name}!"
+
+        result = greet.run("World")
     """
 
-    def _make_with_name(tool_name: str) -> Callable:
-        def _make_tool(f: Callable) -> BaseTool:
+    def _make_with_name(tool_name: str) -> Callable[[Callable[P2, R2]], Tool[P2, R2]]:
+        def _make_tool(f: Callable[P2, R2]) -> Tool[P2, R2]:
             if f.__doc__ is None:
                 raise ValueError("Function must have a docstring")
             if f.__annotations__ is None:
                 raise ValueError("Function must have type annotations")
 
+            func_sig = signature(f)
+            fields: dict[str, Any] = {}
+
+            for param_name, param in func_sig.parameters.items():
+                if param_name == "return":
+                    continue
+                if param.kind in (Parameter.VAR_POSITIONAL, Parameter.VAR_KEYWORD):
+                    continue
+
+                annotation = (
+                    param.annotation if param.annotation != param.empty else Any
+                )
+
+                if param.default is param.empty:
+                    fields[param_name] = (annotation, ...)
+                else:
+                    fields[param_name] = (annotation, param.default)
+
             class_name = "".join(tool_name.split()).title()
-            args_schema = cast(
-                type[PydanticBaseModel],
-                type(
-                    class_name,
-                    (PydanticBaseModel,),
-                    {
-                        "__annotations__": {
-                            k: v for k, v in f.__annotations__.items() if k != "return"
-                        },
-                    },
-                ),
-            )
+            args_schema = create_model(class_name, **fields)
 
             return Tool(
                 name=tool_name,
@@ -360,4 +543,10 @@ def tool(
         return _make_with_name(args[0].__name__)(args[0])
     if len(args) == 1 and isinstance(args[0], str):
         return _make_with_name(args[0])
+    if len(args) == 0:
+
+        def decorator(f: Callable[P2, R2]) -> Tool[P2, R2]:
+            return _make_with_name(f.__name__)(f)
+
+        return decorator
     raise ValueError("Invalid arguments")