Merge branch 'main' into gl/chore/use-base-model-for-llms

- add trust_remote_completion_status flag to A2AConfig, Adds configuration flag to control whether to trust A2A agent completion status. Resolves #3899
- update docs

.github/dependabot.yml

@@ -0,0 +1,11 @@
+# To get started with Dependabot version updates, you'll need to specify which 
+# package ecosystems to update and where the package manifests are located.
+# Please see the documentation for all configuration options:
+# https://docs.github.com/code-security/dependabot/dependabot-version-updates/configuration-options-for-the-dependabot.yml-file
+
+version: 2
+updates:
+  - package-ecosystem: uv # See documentation for possible values
+    directory: "/" # Location of package manifests
+    schedule:
+      interval: "weekly"

.github/workflows/docs-broken-links.yml

@@ -0,0 +1,35 @@
+name: Check Documentation Broken Links
+
+on:
+  pull_request:
+    paths:
+      - "docs/**"
+      - "docs.json"
+  push:
+    branches:
+      - main
+    paths:
+      - "docs/**"
+      - "docs.json"
+  workflow_dispatch:
+
+jobs:
+  check-links:
+    name: Check broken links
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up Node
+        uses: actions/setup-node@v4
+        with:
+          node-version: "latest"
+
+      - name: Install Mintlify CLI
+        run: npm i -g mintlify
+
+      - name: Run broken link checker
+        run: |
+          # Auto-answer the prompt with yes command
+          yes "" | mintlify broken-links || test $? -eq 141
+        working-directory: ./docs

docs/docs.json

@@ -313,7 +313,10 @@
                   "en/learn/multimodal-agents",
                   "en/learn/replay-tasks-from-latest-crew-kickoff",
                   "en/learn/sequential-process",
-                  "en/learn/using-annotations"
+                  "en/learn/using-annotations",
+                  "en/learn/execution-hooks",
+                  "en/learn/llm-hooks",
+                  "en/learn/tool-hooks"
                 ]
               },
               {
@@ -737,7 +740,10 @@
                   "pt-BR/learn/multimodal-agents",
                   "pt-BR/learn/replay-tasks-from-latest-crew-kickoff",
                   "pt-BR/learn/sequential-process",
-                  "pt-BR/learn/using-annotations"
+                  "pt-BR/learn/using-annotations",
+                  "pt-BR/learn/execution-hooks",
+                  "pt-BR/learn/llm-hooks",
+                  "pt-BR/learn/tool-hooks"
                 ]
               },
               {
@@ -1170,7 +1176,10 @@
                   "ko/learn/multimodal-agents",
                   "ko/learn/replay-tasks-from-latest-crew-kickoff",
                   "ko/learn/sequential-process",
-                  "ko/learn/using-annotations"
+                  "ko/learn/using-annotations",
+                  "ko/learn/execution-hooks",
+                  "ko/learn/llm-hooks",
+                  "ko/learn/tool-hooks"
                 ]
               },
               {

docs/en/concepts/knowledge.mdx

@@ -739,7 +739,7 @@ class KnowledgeMonitorListener(BaseEventListener):
 knowledge_monitor = KnowledgeMonitorListener()
 ```
 
-For more information on using events, see the [Event Listeners](https://docs.crewai.com/concepts/event-listener) documentation.
+For more information on using events, see the [Event Listeners](/en/concepts/event-listener) documentation.
 
 ### Custom Knowledge Sources
 

docs/en/concepts/llms.mdx

@@ -1035,7 +1035,7 @@ CrewAI supports streaming responses from LLMs, allowing your application to rece
     ```
 
     <Tip>
-      [Click here](https://docs.crewai.com/concepts/event-listener#event-listeners) for more details
+      [Click here](/en/concepts/event-listener#event-listeners) for more details
     </Tip>
   </Tab>
 
@@ -1212,7 +1212,7 @@ Learn how to get the most out of your LLM configuration:
     ```python
 import httpx
 from crewai import LLM
-from crewai.llms.hooks import BaseInterceptor
+from crewai.llm.hooks import BaseInterceptor
 
 class CustomInterceptor(BaseInterceptor[httpx.Request, httpx.Response]):
     """Custom interceptor to modify requests and responses."""

docs/en/concepts/tasks.mdx

@@ -60,6 +60,7 @@ crew = Crew(
 | **Output Pydantic** _(optional)_ | `output_pydantic` | `Optional[Type[BaseModel]]`   | A Pydantic model for task output.                                                                                    |
 | **Callback** _(optional)_        | `callback`        | `Optional[Any]`               | Function/object to be executed after task completion.                                                                |
 | **Guardrail** _(optional)_       | `guardrail`       | `Optional[Callable]`             | Function to validate task output before proceeding to next task.                                                  |
+| **Guardrails** _(optional)_       | `guardrails`       | `Optional[List[Callable] | List[str]]` | List of guardrails to validate task output before proceeding to next task.                                      |
 | **Guardrail Max Retries** _(optional)_ | `guardrail_max_retries` | `Optional[int]`     | Maximum number of retries when guardrail validation fails. Defaults to 3.                                         |
 
 <Note type="warning" title="Deprecated: max_retries">
@@ -223,6 +224,7 @@ By default, the `TaskOutput` will only include the `raw` output. A `TaskOutput`
 | **JSON Dict**     | `json_dict`     | `Optional[Dict[str, Any]]` | A dictionary representing the JSON output of the task.                                             |
 | **Agent**         | `agent`         | `str`                      | The agent that executed the task.                                                                  |
 | **Output Format** | `output_format` | `OutputFormat`             | The format of the task output, with options including RAW, JSON, and Pydantic. The default is RAW. |
+| **Messages**      | `messages`      | `list[LLMMessage]`         | The messages from the last task execution.                                                           |
 
 ### Task Methods and Properties
 
@@ -341,7 +343,11 @@ Task guardrails provide a way to validate and transform task outputs before they
 are passed to the next task. This feature helps ensure data quality and provides
 feedback to agents when their output doesn't meet specific criteria.
 
-Guardrails are implemented as Python functions that contain custom validation logic, giving you complete control over the validation process and ensuring reliable, deterministic results.
+CrewAI supports two types of guardrails:
+
+1. **Function-based guardrails**: Python functions with custom validation logic, giving you complete control over the validation process and ensuring reliable, deterministic results.
+
+2. **LLM-based guardrails**: String descriptions that use the agent's LLM to validate outputs based on natural language criteria. These are ideal for complex or subjective validation requirements.
 
 ### Function-Based Guardrails
 
@@ -355,12 +361,12 @@ def validate_blog_content(result: TaskOutput) -> Tuple[bool, Any]:
     """Validate blog content meets requirements."""
     try:
         # Check word count
-        word_count = len(result.split())
+        word_count = len(result.raw.split())
         if word_count > 200:
             return (False, "Blog content exceeds 200 words")
 
         # Additional validation logic here
-        return (True, result.strip())
+        return (True, result.raw.strip())
     except Exception as e:
         return (False, "Unexpected error during validation")
 
@@ -372,6 +378,147 @@ blog_task = Task(
 )
 ```
 
+### LLM-Based Guardrails (String Descriptions)
+
+Instead of writing custom validation functions, you can use string descriptions that leverage LLM-based validation. When you provide a string to the `guardrail` or `guardrails` parameter, CrewAI automatically creates an `LLMGuardrail` that uses the agent's LLM to validate the output based on your description.
+
+**Requirements**:
+- The task must have an `agent` assigned (the guardrail uses the agent's LLM)
+- Provide a clear, descriptive string explaining the validation criteria
+
+```python Code
+from crewai import Task
+
+# Single LLM-based guardrail
+blog_task = Task(
+    description="Write a blog post about AI",
+    expected_output="A blog post under 200 words",
+    agent=blog_agent,
+    guardrail="The blog post must be under 200 words and contain no technical jargon"
+)
+```
+
+LLM-based guardrails are particularly useful for:
+- **Complex validation logic** that's difficult to express programmatically
+- **Subjective criteria** like tone, style, or quality assessments
+- **Natural language requirements** that are easier to describe than code
+
+The LLM guardrail will:
+1. Analyze the task output against your description
+2. Return `(True, output)` if the output complies with the criteria
+3. Return `(False, feedback)` with specific feedback if validation fails
+
+**Example with detailed validation criteria**:
+
+```python Code
+research_task = Task(
+    description="Research the latest developments in quantum computing",
+    expected_output="A comprehensive research report",
+    agent=researcher_agent,
+    guardrail="""
+    The research report must:
+    - Be at least 1000 words long
+    - Include at least 5 credible sources
+    - Cover both technical and practical applications
+    - Be written in a professional, academic tone
+    - Avoid speculation or unverified claims
+    """
+)
+```
+
+### Multiple Guardrails
+
+You can apply multiple guardrails to a task using the `guardrails` parameter. Multiple guardrails are executed sequentially, with each guardrail receiving the output from the previous one. This allows you to chain validation and transformation steps.
+
+The `guardrails` parameter accepts:
+- A list of guardrail functions or string descriptions
+- A single guardrail function or string (same as `guardrail`)
+
+**Note**: If `guardrails` is provided, it takes precedence over `guardrail`. The `guardrail` parameter will be ignored when `guardrails` is set.
+
+```python Code
+from typing import Tuple, Any
+from crewai import TaskOutput, Task
+
+def validate_word_count(result: TaskOutput) -> Tuple[bool, Any]:
+    """Validate word count is within limits."""
+    word_count = len(result.raw.split())
+    if word_count < 100:
+        return (False, f"Content too short: {word_count} words. Need at least 100 words.")
+    if word_count > 500:
+        return (False, f"Content too long: {word_count} words. Maximum is 500 words.")
+    return (True, result.raw)
+
+def validate_no_profanity(result: TaskOutput) -> Tuple[bool, Any]:
+    """Check for inappropriate language."""
+    profanity_words = ["badword1", "badword2"]  # Example list
+    content_lower = result.raw.lower()
+    for word in profanity_words:
+        if word in content_lower:
+            return (False, f"Inappropriate language detected: {word}")
+    return (True, result.raw)
+
+def format_output(result: TaskOutput) -> Tuple[bool, Any]:
+    """Format and clean the output."""
+    formatted = result.raw.strip()
+    # Capitalize first letter
+    formatted = formatted[0].upper() + formatted[1:] if formatted else formatted
+    return (True, formatted)
+
+# Apply multiple guardrails sequentially
+blog_task = Task(
+    description="Write a blog post about AI",
+    expected_output="A well-formatted blog post between 100-500 words",
+    agent=blog_agent,
+    guardrails=[
+        validate_word_count,      # First: validate length
+        validate_no_profanity,    # Second: check content
+        format_output             # Third: format the result
+    ],
+    guardrail_max_retries=3
+)
+```
+
+In this example, the guardrails execute in order:
+1. `validate_word_count` checks the word count
+2. `validate_no_profanity` checks for inappropriate language (using the output from step 1)
+3. `format_output` formats the final result (using the output from step 2)
+
+If any guardrail fails, the error is sent back to the agent, and the task is retried up to `guardrail_max_retries` times.
+
+**Mixing function-based and LLM-based guardrails**:
+
+You can combine both function-based and string-based guardrails in the same list:
+
+```python Code
+from typing import Tuple, Any
+from crewai import TaskOutput, Task
+
+def validate_word_count(result: TaskOutput) -> Tuple[bool, Any]:
+    """Validate word count is within limits."""
+    word_count = len(result.raw.split())
+    if word_count < 100:
+        return (False, f"Content too short: {word_count} words. Need at least 100 words.")
+    if word_count > 500:
+        return (False, f"Content too long: {word_count} words. Maximum is 500 words.")
+    return (True, result.raw)
+
+# Mix function-based and LLM-based guardrails
+blog_task = Task(
+    description="Write a blog post about AI",
+    expected_output="A well-formatted blog post between 100-500 words",
+    agent=blog_agent,
+    guardrails=[
+        validate_word_count,  # Function-based: precise word count check
+        "The content must be engaging and suitable for a general audience",  # LLM-based: subjective quality check
+        "The writing style should be clear, concise, and free of technical jargon"  # LLM-based: style validation
+    ],
+    guardrail_max_retries=3
+)
+```
+
+This approach combines the precision of programmatic validation with the flexibility of LLM-based assessment for subjective criteria.
+
 ### Guardrail Function Requirements
 
 1. **Function Signature**:

docs/en/enterprise/features/marketplace.mdx

@@ -37,7 +37,7 @@ you can use them locally or refine them to your needs.
   <Card title="Tools & Integrations" href="/en/enterprise/features/tools-and-integrations" icon="wrench">
     Connect external apps and manage internal tools your agents can use.
   </Card>
-  <Card title="Tool Repository" href="/en/enterprise/features/tool-repository" icon="toolbox">
+  <Card title="Tool Repository" href="/en/enterprise/guides/tool-repository#tool-repository" icon="toolbox">
     Publish and install tools to enhance your crews' capabilities.
   </Card>
   <Card title="Agents Repository" href="/en/enterprise/features/agent-repositories" icon="people-group">

docs/en/enterprise/features/tools-and-integrations.mdx

@@ -241,7 +241,7 @@ Tools & Integrations is the central hub for connecting third‑party apps and ma
 ## Related
 
 <CardGroup cols={2}>
-  <Card title="Tool Repository" href="/en/enterprise/features/tool-repository" icon="toolbox">
+  <Card title="Tool Repository" href="/en/enterprise/guides/tool-repository#tool-repository" icon="toolbox">
     Create, publish, and version custom tools for your organization.
   </Card>
   <Card title="Webhook Automation" href="/en/enterprise/guides/webhook-automation" icon="bolt">

docs/en/enterprise/guides/tool-repository.mdx

@@ -21,7 +21,7 @@ The repository is not a version control system. Use Git to track code changes an
 Before using the Tool Repository, ensure you have:
 
 - A [CrewAI AMP](https://app.crewai.com) account
-- [CrewAI CLI](https://docs.crewai.com/concepts/cli#cli) installed
+- [CrewAI CLI](/en/concepts/cli#cli) installed
 - uv>=0.5.0 installed. Check out [how to upgrade](https://docs.astral.sh/uv/getting-started/installation/#upgrading-uv)
 - [Git](https://git-scm.com) installed and configured
 - Access permissions to publish or install tools in your CrewAI AMP organization
@@ -112,7 +112,7 @@ By default, tools are published as private. To make a tool public:
 crewai tool publish --public
 ```
 
-For more details on how to build tools, see [Creating your own tools](https://docs.crewai.com/concepts/tools#creating-your-own-tools).
+For more details on how to build tools, see [Creating your own tools](/en/concepts/tools#creating-your-own-tools).
 
 ## Updating Tools
 

docs/en/enterprise/resources/frequently-asked-questions.mdx

@@ -49,7 +49,7 @@ mode: "wide"
 
         To integrate human input into agent execution, set the `human_input` flag in the task definition. When enabled, the agent prompts the user for input before delivering its final answer. This input can provide extra context, clarify ambiguities, or validate the agent's output.
 
-        For detailed implementation guidance, see our [Human-in-the-Loop guide](/en/how-to/human-in-the-loop).
+        For detailed implementation guidance, see our [Human-in-the-Loop guide](/en/enterprise/guides/human-in-the-loop).
     </Accordion>
 
     <Accordion title="What advanced customization options are available for tailoring and enhancing agent behavior and capabilities in CrewAI?">
@@ -142,7 +142,7 @@ mode: "wide"
     <Accordion title="How can I create custom tools for my CrewAI agents?">
         You can create custom tools by subclassing the `BaseTool` class provided by CrewAI or by using the tool decorator. Subclassing involves defining a new class that inherits from `BaseTool`, specifying the name, description, and the `_run` method for operational logic. The tool decorator allows you to create a `Tool` object directly with the required attributes and a functional logic.
 
-        <Card href="https://docs.crewai.com/how-to/create-custom-tools" icon="code">CrewAI Tools Guide</Card>
+        <Card href="/en/learn/create-custom-tools" icon="code">CrewAI Tools Guide</Card>
     </Accordion>
 
     <Accordion title="How can you control the maximum number of requests per minute that the entire crew can perform?">

docs/en/learn/a2a-agent-delegation.mdx

@@ -83,6 +83,10 @@ The `A2AConfig` class accepts the following parameters:
   Whether to raise an error immediately if agent connection fails. When `False`, the agent continues with available agents and informs the LLM about unavailable ones.
 </ParamField>
 
+<ParamField path="trust_remote_completion_status" type="bool" default="False">
+  When `True`, returns the A2A agent's result directly when it signals completion. When `False`, allows the server agent to review the result and potentially continue the conversation.
+</ParamField>
+
 ## Authentication
 
 For A2A agents that require authentication, use one of the provided auth schemes:

docs/en/learn/execution-hooks.mdx

@@ -0,0 +1,522 @@
+---
+title: Execution Hooks Overview
+description: Understanding and using execution hooks in CrewAI for fine-grained control over agent operations
+mode: "wide"
+---
+
+Execution Hooks provide fine-grained control over the runtime behavior of your CrewAI agents. Unlike kickoff hooks that run before and after crew execution, execution hooks intercept specific operations during agent execution, allowing you to modify behavior, implement safety checks, and add comprehensive monitoring.
+
+## Types of Execution Hooks
+
+CrewAI provides two main categories of execution hooks:
+
+### 1. [LLM Call Hooks](/learn/llm-hooks)
+
+Control and monitor language model interactions:
+- **Before LLM Call**: Modify prompts, validate inputs, implement approval gates
+- **After LLM Call**: Transform responses, sanitize outputs, update conversation history
+
+**Use Cases:**
+- Iteration limiting
+- Cost tracking and token usage monitoring
+- Response sanitization and content filtering
+- Human-in-the-loop approval for LLM calls
+- Adding safety guidelines or context
+- Debug logging and request/response inspection
+
+[View LLM Hooks Documentation →](/learn/llm-hooks)
+
+### 2. [Tool Call Hooks](/learn/tool-hooks)
+
+Control and monitor tool execution:
+- **Before Tool Call**: Modify inputs, validate parameters, block dangerous operations
+- **After Tool Call**: Transform results, sanitize outputs, log execution details
+
+**Use Cases:**
+- Safety guardrails for destructive operations
+- Human approval for sensitive actions
+- Input validation and sanitization
+- Result caching and rate limiting
+- Tool usage analytics
+- Debug logging and monitoring
+
+[View Tool Hooks Documentation →](/learn/tool-hooks)
+
+## Hook Registration Methods
+
+### 1. Decorator-Based Hooks (Recommended)
+
+The cleanest and most Pythonic way to register hooks:
+
+```python
+from crewai.hooks import before_llm_call, after_llm_call, before_tool_call, after_tool_call
+
+@before_llm_call
+def limit_iterations(context):
+    """Prevent infinite loops by limiting iterations."""
+    if context.iterations > 10:
+        return False  # Block execution
+    return None
+
+@after_llm_call
+def sanitize_response(context):
+    """Remove sensitive data from LLM responses."""
+    if "API_KEY" in context.response:
+        return context.response.replace("API_KEY", "[REDACTED]")
+    return None
+
+@before_tool_call
+def block_dangerous_tools(context):
+    """Block destructive operations."""
+    if context.tool_name == "delete_database":
+        return False  # Block execution
+    return None
+
+@after_tool_call
+def log_tool_result(context):
+    """Log tool execution."""
+    print(f"Tool {context.tool_name} completed")
+    return None
+```
+
+### 2. Crew-Scoped Hooks
+
+Apply hooks only to specific crew instances:
+
+```python
+from crewai import CrewBase
+from crewai.project import crew
+from crewai.hooks import before_llm_call_crew, after_tool_call_crew
+
+@CrewBase
+class MyProjCrew:
+    @before_llm_call_crew
+    def validate_inputs(self, context):
+        # Only applies to this crew
+        print(f"LLM call in {self.__class__.__name__}")
+        return None
+
+    @after_tool_call_crew
+    def log_results(self, context):
+        # Crew-specific logging
+        print(f"Tool result: {context.tool_result[:50]}...")
+        return None
+
+    @crew
+    def crew(self) -> Crew:
+        return Crew(
+            agents=self.agents,
+            tasks=self.tasks,
+            process=Process.sequential
+        )
+```
+
+## Hook Execution Flow
+
+### LLM Call Flow
+
+```
+Agent needs to call LLM
+    ↓
+[Before LLM Call Hooks Execute]
+    ├→ Hook 1: Validate iteration count
+    ├→ Hook 2: Add safety context
+    └→ Hook 3: Log request
+    ↓
+If any hook returns False:
+    ├→ Block LLM call
+    └→ Raise ValueError
+    ↓
+If all hooks return True/None:
+    ├→ LLM call proceeds
+    └→ Response generated
+    ↓
+[After LLM Call Hooks Execute]
+    ├→ Hook 1: Sanitize response
+    ├→ Hook 2: Log response
+    └→ Hook 3: Update metrics
+    ↓
+Final response returned
+```
+
+### Tool Call Flow
+
+```
+Agent needs to execute tool
+    ↓
+[Before Tool Call Hooks Execute]
+    ├→ Hook 1: Check if tool is allowed
+    ├→ Hook 2: Validate inputs
+    └→ Hook 3: Request approval if needed
+    ↓
+If any hook returns False:
+    ├→ Block tool execution
+    └→ Return error message
+    ↓
+If all hooks return True/None:
+    ├→ Tool execution proceeds
+    └→ Result generated
+    ↓
+[After Tool Call Hooks Execute]
+    ├→ Hook 1: Sanitize result
+    ├→ Hook 2: Cache result
+    └→ Hook 3: Log metrics
+    ↓
+Final result returned
+```
+
+## Hook Context Objects
+
+### LLMCallHookContext
+
+Provides access to LLM execution state:
+
+```python
+class LLMCallHookContext:
+    executor: CrewAgentExecutor  # Full executor access
+    messages: list               # Mutable message list
+    agent: Agent                 # Current agent
+    task: Task                   # Current task
+    crew: Crew                   # Crew instance
+    llm: BaseLLM                 # LLM instance
+    iterations: int              # Current iteration
+    response: str | None         # LLM response (after hooks)
+```
+
+### ToolCallHookContext
+
+Provides access to tool execution state:
+
+```python
+class ToolCallHookContext:
+    tool_name: str               # Tool being called
+    tool_input: dict             # Mutable input parameters
+    tool: CrewStructuredTool     # Tool instance
+    agent: Agent | None          # Agent executing
+    task: Task | None            # Current task
+    crew: Crew | None            # Crew instance
+    tool_result: str | None      # Tool result (after hooks)
+```
+
+## Common Patterns
+
+### Safety and Validation
+
+```python
+@before_tool_call
+def safety_check(context):
+    """Block destructive operations."""
+    dangerous = ['delete_file', 'drop_table', 'system_shutdown']
+    if context.tool_name in dangerous:
+        print(f"🛑 Blocked: {context.tool_name}")
+        return False
+    return None
+
+@before_llm_call
+def iteration_limit(context):
+    """Prevent infinite loops."""
+    if context.iterations > 15:
+        print("⛔ Maximum iterations exceeded")
+        return False
+    return None
+```
+
+### Human-in-the-Loop
+
+```python
+@before_tool_call
+def require_approval(context):
+    """Require approval for sensitive operations."""
+    sensitive = ['send_email', 'make_payment', 'post_message']
+
+    if context.tool_name in sensitive:
+        response = context.request_human_input(
+            prompt=f"Approve {context.tool_name}?",
+            default_message="Type 'yes' to approve:"
+        )
+
+        if response.lower() != 'yes':
+            return False
+
+    return None
+```
+
+### Monitoring and Analytics
+
+```python
+from collections import defaultdict
+import time
+
+metrics = defaultdict(lambda: {'count': 0, 'total_time': 0})
+
+@before_tool_call
+def start_timer(context):
+    context.tool_input['_start'] = time.time()
+    return None
+
+@after_tool_call
+def track_metrics(context):
+    start = context.tool_input.get('_start', time.time())
+    duration = time.time() - start
+
+    metrics[context.tool_name]['count'] += 1
+    metrics[context.tool_name]['total_time'] += duration
+
+    return None
+
+# View metrics
+def print_metrics():
+    for tool, data in metrics.items():
+        avg = data['total_time'] / data['count']
+        print(f"{tool}: {data['count']} calls, {avg:.2f}s avg")
+```
+
+### Response Sanitization
+
+```python
+import re
+
+@after_llm_call
+def sanitize_llm_response(context):
+    """Remove sensitive data from LLM responses."""
+    if not context.response:
+        return None
+
+    result = context.response
+    result = re.sub(r'(api[_-]?key)["\']?\s*[:=]\s*["\']?[\w-]+',
+                   r'\1: [REDACTED]', result, flags=re.IGNORECASE)
+    return result
+
+@after_tool_call
+def sanitize_tool_result(context):
+    """Remove sensitive data from tool results."""
+    if not context.tool_result:
+        return None
+
+    result = context.tool_result
+    result = re.sub(r'\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Z|a-z]{2,}\b',
+                   '[EMAIL-REDACTED]', result)
+    return result
+```
+
+## Hook Management
+
+### Clearing All Hooks
+
+```python
+from crewai.hooks import clear_all_global_hooks
+
+# Clear all hooks at once
+result = clear_all_global_hooks()
+print(f"Cleared {result['total']} hooks")
+# Output: {'llm_hooks': (2, 1), 'tool_hooks': (1, 2), 'total': (3, 3)}
+```
+
+### Clearing Specific Hook Types
+
+```python
+from crewai.hooks import (
+    clear_before_llm_call_hooks,
+    clear_after_llm_call_hooks,
+    clear_before_tool_call_hooks,
+    clear_after_tool_call_hooks
+)
+
+# Clear specific types
+llm_before_count = clear_before_llm_call_hooks()
+tool_after_count = clear_after_tool_call_hooks()
+```
+
+### Unregistering Individual Hooks
+
+```python
+from crewai.hooks import (
+    unregister_before_llm_call_hook,
+    unregister_after_tool_call_hook
+)
+
+def my_hook(context):
+    ...
+
+# Register
+register_before_llm_call_hook(my_hook)
+
+# Later, unregister
+success = unregister_before_llm_call_hook(my_hook)
+print(f"Unregistered: {success}")
+```
+
+## Best Practices
+
+### 1. Keep Hooks Focused
+Each hook should have a single, clear responsibility:
+
+```python
+# ✅ Good - focused responsibility
+@before_tool_call
+def validate_file_path(context):
+    if context.tool_name == 'read_file':
+        if '..' in context.tool_input.get('path', ''):
+            return False
+    return None
+
+# ❌ Bad - too many responsibilities
+@before_tool_call
+def do_everything(context):
+    # Validation + logging + metrics + approval...
+    ...
+```
+
+### 2. Handle Errors Gracefully
+
+```python
+@before_llm_call
+def safe_hook(context):
+    try:
+        # Your logic
+        if some_condition:
+            return False
+    except Exception as e:
+        print(f"Hook error: {e}")
+        return None  # Allow execution despite error
+```
+
+### 3. Modify Context In-Place
+
+```python
+# ✅ Correct - modify in-place
+@before_llm_call
+def add_context(context):
+    context.messages.append({"role": "system", "content": "Be concise"})
+
+# ❌ Wrong - replaces reference
+@before_llm_call
+def wrong_approach(context):
+    context.messages = [{"role": "system", "content": "Be concise"}]
+```
+
+### 4. Use Type Hints
+
+```python
+from crewai.hooks import LLMCallHookContext, ToolCallHookContext
+
+def my_llm_hook(context: LLMCallHookContext) -> bool | None:
+    # IDE autocomplete and type checking
+    return None
+
+def my_tool_hook(context: ToolCallHookContext) -> str | None:
+    return None
+```
+
+### 5. Clean Up in Tests
+
+```python
+import pytest
+from crewai.hooks import clear_all_global_hooks
+
+@pytest.fixture(autouse=True)
+def clean_hooks():
+    """Reset hooks before each test."""
+    yield
+    clear_all_global_hooks()
+```
+
+## When to Use Which Hook
+
+### Use LLM Hooks When:
+- Implementing iteration limits
+- Adding context or safety guidelines to prompts
+- Tracking token usage and costs
+- Sanitizing or transforming responses
+- Implementing approval gates for LLM calls
+- Debugging prompt/response interactions
+
+### Use Tool Hooks When:
+- Blocking dangerous or destructive operations
+- Validating tool inputs before execution
+- Implementing approval gates for sensitive actions
+- Caching tool results
+- Tracking tool usage and performance
+- Sanitizing tool outputs
+- Rate limiting tool calls
+
+### Use Both When:
+Building comprehensive observability, safety, or approval systems that need to monitor all agent operations.
+
+## Alternative Registration Methods
+
+### Programmatic Registration (Advanced)
+
+For dynamic hook registration or when you need to register hooks programmatically:
+
+```python
+from crewai.hooks import (
+    register_before_llm_call_hook,
+    register_after_tool_call_hook
+)
+
+def my_hook(context):
+    return None
+
+# Register programmatically
+register_before_llm_call_hook(my_hook)
+
+# Useful for:
+# - Loading hooks from configuration
+# - Conditional hook registration
+# - Plugin systems
+```
+
+**Note:** For most use cases, decorators are cleaner and more maintainable.
+
+## Performance Considerations
+
+1. **Keep Hooks Fast**: Hooks execute on every call - avoid heavy computation
+2. **Cache When Possible**: Store expensive validations or lookups
+3. **Be Selective**: Use crew-scoped hooks when global hooks aren't needed
+4. **Monitor Hook Overhead**: Profile hook execution time in production
+5. **Lazy Import**: Import heavy dependencies only when needed
+
+## Debugging Hooks
+
+### Enable Debug Logging
+
+```python
+import logging
+
+logging.basicConfig(level=logging.DEBUG)
+logger = logging.getLogger(__name__)
+
+@before_llm_call
+def debug_hook(context):
+    logger.debug(f"LLM call: {context.agent.role}, iteration {context.iterations}")
+    return None
+```
+
+### Hook Execution Order
+
+Hooks execute in registration order. If a before hook returns `False`, subsequent hooks don't execute:
+
+```python
+# Register order matters!
+register_before_tool_call_hook(hook1)  # Executes first
+register_before_tool_call_hook(hook2)  # Executes second
+register_before_tool_call_hook(hook3)  # Executes third
+
+# If hook2 returns False:
+# - hook1 executed
+# - hook2 executed and returned False
+# - hook3 NOT executed
+# - Tool call blocked
+```
+
+## Related Documentation
+
+- [LLM Call Hooks →](/learn/llm-hooks) - Detailed LLM hook documentation
+- [Tool Call Hooks →](/learn/tool-hooks) - Detailed tool hook documentation
+- [Before and After Kickoff Hooks →](/learn/before-and-after-kickoff-hooks) - Crew lifecycle hooks
+- [Human-in-the-Loop →](/learn/human-in-the-loop) - Human input patterns
+
+## Conclusion
+
+Execution hooks provide powerful control over agent runtime behavior. Use them to implement safety guardrails, approval workflows, comprehensive monitoring, and custom business logic. Combined with proper error handling, type safety, and performance considerations, hooks enable production-ready, secure, and observable agent systems.

docs/en/learn/hierarchical-process.mdx

@@ -97,7 +97,7 @@ project_crew = Crew(
 ```
 
 <Tip>
-    For more details on creating and customizing a manager agent, check out the [Custom Manager Agent documentation](https://docs.crewai.com/how-to/custom-manager-agent#custom-manager-agent).
+    For more details on creating and customizing a manager agent, check out the [Custom Manager Agent documentation](/en/learn/custom-manager-agent).
 </Tip>
 
 

docs/en/learn/llm-hooks.mdx

@@ -0,0 +1,427 @@
+---
+title: LLM Call Hooks
+description: Learn how to use LLM call hooks to intercept, modify, and control language model interactions in CrewAI
+mode: "wide"
+---
+
+LLM Call Hooks provide fine-grained control over language model interactions during agent execution. These hooks allow you to intercept LLM calls, modify prompts, transform responses, implement approval gates, and add custom logging or monitoring.
+
+## Overview
+
+LLM hooks are executed at two critical points:
+- **Before LLM Call**: Modify messages, validate inputs, or block execution
+- **After LLM Call**: Transform responses, sanitize outputs, or modify conversation history
+
+## Hook Types
+
+### Before LLM Call Hooks
+
+Executed before every LLM call, these hooks can:
+- Inspect and modify messages sent to the LLM
+- Block LLM execution based on conditions
+- Implement rate limiting or approval gates
+- Add context or system messages
+- Log request details
+
+**Signature:**
+```python
+def before_hook(context: LLMCallHookContext) -> bool | None:
+    # Return False to block execution
+    # Return True or None to allow execution
+    ...
+```
+
+### After LLM Call Hooks
+
+Executed after every LLM call, these hooks can:
+- Modify or sanitize LLM responses
+- Add metadata or formatting
+- Log response details
+- Update conversation history
+- Implement content filtering
+
+**Signature:**
+```python
+def after_hook(context: LLMCallHookContext) -> str | None:
+    # Return modified response string
+    # Return None to keep original response
+    ...
+```
+
+## LLM Hook Context
+
+The `LLMCallHookContext` object provides comprehensive access to execution state:
+
+```python
+class LLMCallHookContext:
+    executor: CrewAgentExecutor  # Full executor reference
+    messages: list               # Mutable message list
+    agent: Agent                 # Current agent
+    task: Task                   # Current task
+    crew: Crew                   # Crew instance
+    llm: BaseLLM                 # LLM instance
+    iterations: int              # Current iteration count
+    response: str | None         # LLM response (after hooks only)
+```
+
+### Modifying Messages
+
+**Important:** Always modify messages in-place:
+
+```python
+# ✅ Correct - modify in-place
+def add_context(context: LLMCallHookContext) -> None:
+    context.messages.append({"role": "system", "content": "Be concise"})
+
+# ❌ Wrong - replaces list reference
+def wrong_approach(context: LLMCallHookContext) -> None:
+    context.messages = [{"role": "system", "content": "Be concise"}]
+```
+
+## Registration Methods
+
+### 1. Global Hook Registration
+
+Register hooks that apply to all LLM calls across all crews:
+
+```python
+from crewai.hooks import register_before_llm_call_hook, register_after_llm_call_hook
+
+def log_llm_call(context):
+    print(f"LLM call by {context.agent.role} at iteration {context.iterations}")
+    return None  # Allow execution
+
+register_before_llm_call_hook(log_llm_call)
+```
+
+### 2. Decorator-Based Registration
+
+Use decorators for cleaner syntax:
+
+```python
+from crewai.hooks import before_llm_call, after_llm_call
+
+@before_llm_call
+def validate_iteration_count(context):
+    if context.iterations > 10:
+        print("⚠️ Exceeded maximum iterations")
+        return False  # Block execution
+    return None
+
+@after_llm_call
+def sanitize_response(context):
+    if context.response and "API_KEY" in context.response:
+        return context.response.replace("API_KEY", "[REDACTED]")
+    return None
+```
+
+### 3. Crew-Scoped Hooks
+
+Register hooks for a specific crew instance:
+
+```python
+@CrewBase
+class MyProjCrew:
+    @before_llm_call_crew
+    def validate_inputs(self, context):
+        # Only applies to this crew
+        if context.iterations == 0:
+            print(f"Starting task: {context.task.description}")
+        return None
+
+    @after_llm_call_crew
+    def log_responses(self, context):
+        # Crew-specific response logging
+        print(f"Response length: {len(context.response)}")
+        return None
+
+    @crew
+    def crew(self) -> Crew:
+        return Crew(
+            agents=self.agents,
+            tasks=self.tasks,
+            process=Process.sequential,
+            verbose=True
+        )
+```
+
+## Common Use Cases
+
+### 1. Iteration Limiting
+
+```python
+@before_llm_call
+def limit_iterations(context: LLMCallHookContext) -> bool | None:
+    max_iterations = 15
+    if context.iterations > max_iterations:
+        print(f"⛔ Blocked: Exceeded {max_iterations} iterations")
+        return False  # Block execution
+    return None
+```
+
+### 2. Human Approval Gate
+
+```python
+@before_llm_call
+def require_approval(context: LLMCallHookContext) -> bool | None:
+    if context.iterations > 5:
+        response = context.request_human_input(
+            prompt=f"Iteration {context.iterations}: Approve LLM call?",
+            default_message="Press Enter to approve, or type 'no' to block:"
+        )
+        if response.lower() == "no":
+            print("🚫 LLM call blocked by user")
+            return False
+    return None
+```
+
+### 3. Adding System Context
+
+```python
+@before_llm_call
+def add_guardrails(context: LLMCallHookContext) -> None:
+    # Add safety guidelines to every LLM call
+    context.messages.append({
+        "role": "system",
+        "content": "Ensure responses are factual and cite sources when possible."
+    })
+    return None
+```
+
+### 4. Response Sanitization
+
+```python
+@after_llm_call
+def sanitize_sensitive_data(context: LLMCallHookContext) -> str | None:
+    if not context.response:
+        return None
+
+    # Remove sensitive patterns
+    import re
+    sanitized = context.response
+    sanitized = re.sub(r'\b\d{3}-\d{2}-\d{4}\b', '[SSN-REDACTED]', sanitized)
+    sanitized = re.sub(r'\b\d{4}[- ]?\d{4}[- ]?\d{4}[- ]?\d{4}\b', '[CARD-REDACTED]', sanitized)
+
+    return sanitized
+```
+
+### 5. Cost Tracking
+
+```python
+import tiktoken
+
+@before_llm_call
+def track_token_usage(context: LLMCallHookContext) -> None:
+    encoding = tiktoken.get_encoding("cl100k_base")
+    total_tokens = sum(
+        len(encoding.encode(msg.get("content", "")))
+        for msg in context.messages
+    )
+    print(f"📊 Input tokens: ~{total_tokens}")
+    return None
+
+@after_llm_call
+def track_response_tokens(context: LLMCallHookContext) -> None:
+    if context.response:
+        encoding = tiktoken.get_encoding("cl100k_base")
+        tokens = len(encoding.encode(context.response))
+        print(f"📊 Response tokens: ~{tokens}")
+    return None
+```
+
+### 6. Debug Logging
+
+```python
+@before_llm_call
+def debug_request(context: LLMCallHookContext) -> None:
+    print(f"""
+    🔍 LLM Call Debug:
+    - Agent: {context.agent.role}
+    - Task: {context.task.description[:50]}...
+    - Iteration: {context.iterations}
+    - Message Count: {len(context.messages)}
+    - Last Message: {context.messages[-1] if context.messages else 'None'}
+    """)
+    return None
+
+@after_llm_call
+def debug_response(context: LLMCallHookContext) -> None:
+    if context.response:
+        print(f"✅ Response Preview: {context.response[:100]}...")
+    return None
+```
+
+## Hook Management
+
+### Unregistering Hooks
+
+```python
+from crewai.hooks import (
+    unregister_before_llm_call_hook,
+    unregister_after_llm_call_hook
+)
+
+# Unregister specific hook
+def my_hook(context):
+    ...
+
+register_before_llm_call_hook(my_hook)
+# Later...
+unregister_before_llm_call_hook(my_hook)  # Returns True if found
+```
+
+### Clearing Hooks
+
+```python
+from crewai.hooks import (
+    clear_before_llm_call_hooks,
+    clear_after_llm_call_hooks,
+    clear_all_llm_call_hooks
+)
+
+# Clear specific hook type
+count = clear_before_llm_call_hooks()
+print(f"Cleared {count} before hooks")
+
+# Clear all LLM hooks
+before_count, after_count = clear_all_llm_call_hooks()
+print(f"Cleared {before_count} before and {after_count} after hooks")
+```
+
+### Listing Registered Hooks
+
+```python
+from crewai.hooks import (
+    get_before_llm_call_hooks,
+    get_after_llm_call_hooks
+)
+
+# Get current hooks
+before_hooks = get_before_llm_call_hooks()
+after_hooks = get_after_llm_call_hooks()
+
+print(f"Registered: {len(before_hooks)} before, {len(after_hooks)} after")
+```
+
+## Advanced Patterns
+
+### Conditional Hook Execution
+
+```python
+@before_llm_call
+def conditional_blocking(context: LLMCallHookContext) -> bool | None:
+    # Only block for specific agents
+    if context.agent.role == "researcher" and context.iterations > 10:
+        return False
+
+    # Only block for specific tasks
+    if "sensitive" in context.task.description.lower() and context.iterations > 5:
+        return False
+
+    return None
+```
+
+### Context-Aware Modifications
+
+```python
+@before_llm_call
+def adaptive_prompting(context: LLMCallHookContext) -> None:
+    # Add different context based on iteration
+    if context.iterations == 0:
+        context.messages.append({
+            "role": "system",
+            "content": "Start with a high-level overview."
+        })
+    elif context.iterations > 3:
+        context.messages.append({
+            "role": "system",
+            "content": "Focus on specific details and provide examples."
+        })
+    return None
+```
+
+### Chaining Hooks
+
+```python
+# Multiple hooks execute in registration order
+
+@before_llm_call
+def first_hook(context):
+    print("1. First hook executed")
+    return None
+
+@before_llm_call
+def second_hook(context):
+    print("2. Second hook executed")
+    return None
+
+@before_llm_call
+def blocking_hook(context):
+    if context.iterations > 10:
+        print("3. Blocking hook - execution stopped")
+        return False  # Subsequent hooks won't execute
+    print("3. Blocking hook - execution allowed")
+    return None
+```
+
+## Best Practices
+
+1. **Keep Hooks Focused**: Each hook should have a single responsibility
+2. **Avoid Heavy Computation**: Hooks execute on every LLM call
+3. **Handle Errors Gracefully**: Use try-except to prevent hook failures from breaking execution
+4. **Use Type Hints**: Leverage `LLMCallHookContext` for better IDE support
+5. **Document Hook Behavior**: Especially for blocking conditions
+6. **Test Hooks Independently**: Unit test hooks before using in production
+7. **Clear Hooks in Tests**: Use `clear_all_llm_call_hooks()` between test runs
+8. **Modify In-Place**: Always modify `context.messages` in-place, never replace
+
+## Error Handling
+
+```python
+@before_llm_call
+def safe_hook(context: LLMCallHookContext) -> bool | None:
+    try:
+        # Your hook logic
+        if some_condition:
+            return False
+    except Exception as e:
+        print(f"⚠️ Hook error: {e}")
+        # Decide: allow or block on error
+        return None  # Allow execution despite error
+```
+
+## Type Safety
+
+```python
+from crewai.hooks import LLMCallHookContext, BeforeLLMCallHookType, AfterLLMCallHookType
+
+# Explicit type annotations
+def my_before_hook(context: LLMCallHookContext) -> bool | None:
+    return None
+
+def my_after_hook(context: LLMCallHookContext) -> str | None:
+    return None
+
+# Type-safe registration
+register_before_llm_call_hook(my_before_hook)
+register_after_llm_call_hook(my_after_hook)
+```
+
+## Troubleshooting
+
+### Hook Not Executing
+- Verify hook is registered before crew execution
+- Check if previous hook returned `False` (blocks subsequent hooks)
+- Ensure hook signature matches expected type
+
+### Message Modifications Not Persisting
+- Use in-place modifications: `context.messages.append()`
+- Don't replace the list: `context.messages = []`
+
+### Response Modifications Not Working
+- Return the modified string from after hooks
+- Returning `None` keeps the original response
+
+## Conclusion
+
+LLM Call Hooks provide powerful capabilities for controlling and monitoring language model interactions in CrewAI. Use them to implement safety guardrails, approval gates, logging, cost tracking, and response sanitization. Combined with proper error handling and type safety, hooks enable robust and production-ready agent systems.

docs/en/learn/tool-hooks.mdx

@@ -0,0 +1,600 @@
+---
+title: Tool Call Hooks
+description: Learn how to use tool call hooks to intercept, modify, and control tool execution in CrewAI
+mode: "wide"
+---
+
+Tool Call Hooks provide fine-grained control over tool execution during agent operations. These hooks allow you to intercept tool calls, modify inputs, transform outputs, implement safety checks, and add comprehensive logging or monitoring.
+
+## Overview
+
+Tool hooks are executed at two critical points:
+- **Before Tool Call**: Modify inputs, validate parameters, or block execution
+- **After Tool Call**: Transform results, sanitize outputs, or log execution details
+
+## Hook Types
+
+### Before Tool Call Hooks
+
+Executed before every tool execution, these hooks can:
+- Inspect and modify tool inputs
+- Block tool execution based on conditions
+- Implement approval gates for dangerous operations
+- Validate parameters
+- Log tool invocations
+
+**Signature:**
+```python
+def before_hook(context: ToolCallHookContext) -> bool | None:
+    # Return False to block execution
+    # Return True or None to allow execution
+    ...
+```
+
+### After Tool Call Hooks
+
+Executed after every tool execution, these hooks can:
+- Modify or sanitize tool results
+- Add metadata or formatting
+- Log execution results
+- Implement result validation
+- Transform output formats
+
+**Signature:**
+```python
+def after_hook(context: ToolCallHookContext) -> str | None:
+    # Return modified result string
+    # Return None to keep original result
+    ...
+```
+
+## Tool Hook Context
+
+The `ToolCallHookContext` object provides comprehensive access to tool execution state:
+
+```python
+class ToolCallHookContext:
+    tool_name: str                    # Name of the tool being called
+    tool_input: dict[str, Any]        # Mutable tool input parameters
+    tool: CrewStructuredTool          # Tool instance reference
+    agent: Agent | BaseAgent | None   # Agent executing the tool
+    task: Task | None                 # Current task
+    crew: Crew | None                 # Crew instance
+    tool_result: str | None           # Tool result (after hooks only)
+```
+
+### Modifying Tool Inputs
+
+**Important:** Always modify tool inputs in-place:
+
+```python
+# ✅ Correct - modify in-place
+def sanitize_input(context: ToolCallHookContext) -> None:
+    context.tool_input['query'] = context.tool_input['query'].lower()
+
+# ❌ Wrong - replaces dict reference
+def wrong_approach(context: ToolCallHookContext) -> None:
+    context.tool_input = {'query': 'new query'}
+```
+
+## Registration Methods
+
+### 1. Global Hook Registration
+
+Register hooks that apply to all tool calls across all crews:
+
+```python
+from crewai.hooks import register_before_tool_call_hook, register_after_tool_call_hook
+
+def log_tool_call(context):
+    print(f"Tool: {context.tool_name}")
+    print(f"Input: {context.tool_input}")
+    return None  # Allow execution
+
+register_before_tool_call_hook(log_tool_call)
+```
+
+### 2. Decorator-Based Registration
+
+Use decorators for cleaner syntax:
+
+```python
+from crewai.hooks import before_tool_call, after_tool_call
+
+@before_tool_call
+def block_dangerous_tools(context):
+    dangerous_tools = ['delete_database', 'drop_table', 'rm_rf']
+    if context.tool_name in dangerous_tools:
+        print(f"⛔ Blocked dangerous tool: {context.tool_name}")
+        return False  # Block execution
+    return None
+
+@after_tool_call
+def sanitize_results(context):
+    if context.tool_result and "password" in context.tool_result.lower():
+        return context.tool_result.replace("password", "[REDACTED]")
+    return None
+```
+
+### 3. Crew-Scoped Hooks
+
+Register hooks for a specific crew instance:
+
+```python
+@CrewBase
+class MyProjCrew:
+    @before_tool_call_crew
+    def validate_tool_inputs(self, context):
+        # Only applies to this crew
+        if context.tool_name == "web_search":
+            if not context.tool_input.get('query'):
+                print("❌ Invalid search query")
+                return False
+        return None
+
+    @after_tool_call_crew
+    def log_tool_results(self, context):
+        # Crew-specific tool logging
+        print(f"✅ {context.tool_name} completed")
+        return None
+
+    @crew
+    def crew(self) -> Crew:
+        return Crew(
+            agents=self.agents,
+            tasks=self.tasks,
+            process=Process.sequential,
+            verbose=True
+        )
+```
+
+## Common Use Cases
+
+### 1. Safety Guardrails
+
+```python
+@before_tool_call
+def safety_check(context: ToolCallHookContext) -> bool | None:
+    # Block tools that could cause harm
+    destructive_tools = [
+        'delete_file',
+        'drop_table',
+        'remove_user',
+        'system_shutdown'
+    ]
+
+    if context.tool_name in destructive_tools:
+        print(f"🛑 Blocked destructive tool: {context.tool_name}")
+        return False
+
+    # Warn on sensitive operations
+    sensitive_tools = ['send_email', 'post_to_social_media', 'charge_payment']
+    if context.tool_name in sensitive_tools:
+        print(f"⚠️  Executing sensitive tool: {context.tool_name}")
+
+    return None
+```
+
+### 2. Human Approval Gate
+
+```python
+@before_tool_call
+def require_approval_for_actions(context: ToolCallHookContext) -> bool | None:
+    approval_required = [
+        'send_email',
+        'make_purchase',
+        'delete_file',
+        'post_message'
+    ]
+
+    if context.tool_name in approval_required:
+        response = context.request_human_input(
+            prompt=f"Approve {context.tool_name}?",
+            default_message=f"Input: {context.tool_input}\nType 'yes' to approve:"
+        )
+
+        if response.lower() != 'yes':
+            print(f"❌ Tool execution denied: {context.tool_name}")
+            return False
+
+    return None
+```
+
+### 3. Input Validation and Sanitization
+
+```python
+@before_tool_call
+def validate_and_sanitize_inputs(context: ToolCallHookContext) -> bool | None:
+    # Validate search queries
+    if context.tool_name == 'web_search':
+        query = context.tool_input.get('query', '')
+        if len(query) < 3:
+            print("❌ Search query too short")
+            return False
+
+        # Sanitize query
+        context.tool_input['query'] = query.strip().lower()
+
+    # Validate file paths
+    if context.tool_name == 'read_file':
+        path = context.tool_input.get('path', '')
+        if '..' in path or path.startswith('/'):
+            print("❌ Invalid file path")
+            return False
+
+    return None
+```
+
+### 4. Result Sanitization
+
+```python
+@after_tool_call
+def sanitize_sensitive_data(context: ToolCallHookContext) -> str | None:
+    if not context.tool_result:
+        return None
+
+    import re
+    result = context.tool_result
+
+    # Remove API keys
+    result = re.sub(
+        r'(api[_-]?key|token)["\']?\s*[:=]\s*["\']?[\w-]+',
+        r'\1: [REDACTED]',
+        result,
+        flags=re.IGNORECASE
+    )
+
+    # Remove email addresses
+    result = re.sub(
+        r'\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Z|a-z]{2,}\b',
+        '[EMAIL-REDACTED]',
+        result
+    )
+
+    # Remove credit card numbers
+    result = re.sub(
+        r'\b\d{4}[- ]?\d{4}[- ]?\d{4}[- ]?\d{4}\b',
+        '[CARD-REDACTED]',
+        result
+    )
+
+    return result
+```
+
+### 5. Tool Usage Analytics
+
+```python
+import time
+from collections import defaultdict
+
+tool_stats = defaultdict(lambda: {'count': 0, 'total_time': 0, 'failures': 0})
+
+@before_tool_call
+def start_timer(context: ToolCallHookContext) -> None:
+    context.tool_input['_start_time'] = time.time()
+    return None
+
+@after_tool_call
+def track_tool_usage(context: ToolCallHookContext) -> None:
+    start_time = context.tool_input.get('_start_time', time.time())
+    duration = time.time() - start_time
+
+    tool_stats[context.tool_name]['count'] += 1
+    tool_stats[context.tool_name]['total_time'] += duration
+
+    if not context.tool_result or 'error' in context.tool_result.lower():
+        tool_stats[context.tool_name]['failures'] += 1
+
+    print(f"""
+    📊 Tool Stats for {context.tool_name}:
+    - Executions: {tool_stats[context.tool_name]['count']}
+    - Avg Time: {tool_stats[context.tool_name]['total_time'] / tool_stats[context.tool_name]['count']:.2f}s
+    - Failures: {tool_stats[context.tool_name]['failures']}
+    """)
+
+    return None
+```
+
+### 6. Rate Limiting
+
+```python
+from collections import defaultdict
+from datetime import datetime, timedelta
+
+tool_call_history = defaultdict(list)
+
+@before_tool_call
+def rate_limit_tools(context: ToolCallHookContext) -> bool | None:
+    tool_name = context.tool_name
+    now = datetime.now()
+
+    # Clean old entries (older than 1 minute)
+    tool_call_history[tool_name] = [
+        call_time for call_time in tool_call_history[tool_name]
+        if now - call_time < timedelta(minutes=1)
+    ]
+
+    # Check rate limit (max 10 calls per minute)
+    if len(tool_call_history[tool_name]) >= 10:
+        print(f"🚫 Rate limit exceeded for {tool_name}")
+        return False
+
+    # Record this call
+    tool_call_history[tool_name].append(now)
+    return None
+```
+
+### 7. Caching Tool Results
+
+```python
+import hashlib
+import json
+
+tool_cache = {}
+
+def cache_key(tool_name: str, tool_input: dict) -> str:
+    """Generate cache key from tool name and input."""
+    input_str = json.dumps(tool_input, sort_keys=True)
+    return hashlib.md5(f"{tool_name}:{input_str}".encode()).hexdigest()
+
+@before_tool_call
+def check_cache(context: ToolCallHookContext) -> bool | None:
+    key = cache_key(context.tool_name, context.tool_input)
+    if key in tool_cache:
+        print(f"💾 Cache hit for {context.tool_name}")
+        # Note: Can't return cached result from before hook
+        # Would need to implement this differently
+    return None
+
+@after_tool_call
+def cache_result(context: ToolCallHookContext) -> None:
+    if context.tool_result:
+        key = cache_key(context.tool_name, context.tool_input)
+        tool_cache[key] = context.tool_result
+        print(f"💾 Cached result for {context.tool_name}")
+    return None
+```
+
+### 8. Debug Logging
+
+```python
+@before_tool_call
+def debug_tool_call(context: ToolCallHookContext) -> None:
+    print(f"""
+    🔍 Tool Call Debug:
+    - Tool: {context.tool_name}
+    - Agent: {context.agent.role if context.agent else 'Unknown'}
+    - Task: {context.task.description[:50] if context.task else 'Unknown'}...
+    - Input: {context.tool_input}
+    """)
+    return None
+
+@after_tool_call
+def debug_tool_result(context: ToolCallHookContext) -> None:
+    if context.tool_result:
+        result_preview = context.tool_result[:200]
+        print(f"✅ Result Preview: {result_preview}...")
+    else:
+        print("⚠️  No result returned")
+    return None
+```
+
+## Hook Management
+
+### Unregistering Hooks
+
+```python
+from crewai.hooks import (
+    unregister_before_tool_call_hook,
+    unregister_after_tool_call_hook
+)
+
+# Unregister specific hook
+def my_hook(context):
+    ...
+
+register_before_tool_call_hook(my_hook)
+# Later...
+success = unregister_before_tool_call_hook(my_hook)
+print(f"Unregistered: {success}")
+```
+
+### Clearing Hooks
+
+```python
+from crewai.hooks import (
+    clear_before_tool_call_hooks,
+    clear_after_tool_call_hooks,
+    clear_all_tool_call_hooks
+)
+
+# Clear specific hook type
+count = clear_before_tool_call_hooks()
+print(f"Cleared {count} before hooks")
+
+# Clear all tool hooks
+before_count, after_count = clear_all_tool_call_hooks()
+print(f"Cleared {before_count} before and {after_count} after hooks")
+```
+
+### Listing Registered Hooks
+
+```python
+from crewai.hooks import (
+    get_before_tool_call_hooks,
+    get_after_tool_call_hooks
+)
+
+# Get current hooks
+before_hooks = get_before_tool_call_hooks()
+after_hooks = get_after_tool_call_hooks()
+
+print(f"Registered: {len(before_hooks)} before, {len(after_hooks)} after")
+```
+
+## Advanced Patterns
+
+### Conditional Hook Execution
+
+```python
+@before_tool_call
+def conditional_blocking(context: ToolCallHookContext) -> bool | None:
+    # Only block for specific agents
+    if context.agent and context.agent.role == "junior_agent":
+        if context.tool_name in ['delete_file', 'send_email']:
+            print(f"❌ Junior agents cannot use {context.tool_name}")
+            return False
+
+    # Only block during specific tasks
+    if context.task and "sensitive" in context.task.description.lower():
+        if context.tool_name == 'web_search':
+            print("❌ Web search blocked for sensitive tasks")
+            return False
+
+    return None
+```
+
+### Context-Aware Input Modification
+
+```python
+@before_tool_call
+def enhance_tool_inputs(context: ToolCallHookContext) -> None:
+    # Add context based on agent role
+    if context.agent and context.agent.role == "researcher":
+        if context.tool_name == 'web_search':
+            # Add domain restrictions for researchers
+            context.tool_input['domains'] = ['edu', 'gov', 'org']
+
+    # Add context based on task
+    if context.task and "urgent" in context.task.description.lower():
+        if context.tool_name == 'send_email':
+            context.tool_input['priority'] = 'high'
+
+    return None
+```
+
+### Tool Chain Monitoring
+
+```python
+tool_call_chain = []
+
+@before_tool_call
+def track_tool_chain(context: ToolCallHookContext) -> None:
+    tool_call_chain.append({
+        'tool': context.tool_name,
+        'timestamp': time.time(),
+        'agent': context.agent.role if context.agent else 'Unknown'
+    })
+
+    # Detect potential infinite loops
+    recent_calls = tool_call_chain[-5:]
+    if len(recent_calls) == 5 and all(c['tool'] == context.tool_name for c in recent_calls):
+        print(f"⚠️  Warning: {context.tool_name} called 5 times in a row")
+
+    return None
+```
+
+## Best Practices
+
+1. **Keep Hooks Focused**: Each hook should have a single responsibility
+2. **Avoid Heavy Computation**: Hooks execute on every tool call
+3. **Handle Errors Gracefully**: Use try-except to prevent hook failures
+4. **Use Type Hints**: Leverage `ToolCallHookContext` for better IDE support
+5. **Document Blocking Conditions**: Make it clear when/why tools are blocked
+6. **Test Hooks Independently**: Unit test hooks before using in production
+7. **Clear Hooks in Tests**: Use `clear_all_tool_call_hooks()` between test runs
+8. **Modify In-Place**: Always modify `context.tool_input` in-place, never replace
+9. **Log Important Decisions**: Especially when blocking tool execution
+10. **Consider Performance**: Cache expensive validations when possible
+
+## Error Handling
+
+```python
+@before_tool_call
+def safe_validation(context: ToolCallHookContext) -> bool | None:
+    try:
+        # Your validation logic
+        if not validate_input(context.tool_input):
+            return False
+    except Exception as e:
+        print(f"⚠️ Hook error: {e}")
+        # Decide: allow or block on error
+        return None  # Allow execution despite error
+```
+
+## Type Safety
+
+```python
+from crewai.hooks import ToolCallHookContext, BeforeToolCallHookType, AfterToolCallHookType
+
+# Explicit type annotations
+def my_before_hook(context: ToolCallHookContext) -> bool | None:
+    return None
+
+def my_after_hook(context: ToolCallHookContext) -> str | None:
+    return None
+
+# Type-safe registration
+register_before_tool_call_hook(my_before_hook)
+register_after_tool_call_hook(my_after_hook)
+```
+
+## Integration with Existing Tools
+
+### Wrapping Existing Validation
+
+```python
+def existing_validator(tool_name: str, inputs: dict) -> bool:
+    """Your existing validation function."""
+    # Your validation logic
+    return True
+
+@before_tool_call
+def integrate_validator(context: ToolCallHookContext) -> bool | None:
+    if not existing_validator(context.tool_name, context.tool_input):
+        print(f"❌ Validation failed for {context.tool_name}")
+        return False
+    return None
+```
+
+### Logging to External Systems
+
+```python
+import logging
+
+logger = logging.getLogger(__name__)
+
+@before_tool_call
+def log_to_external_system(context: ToolCallHookContext) -> None:
+    logger.info(f"Tool call: {context.tool_name}", extra={
+        'tool_name': context.tool_name,
+        'tool_input': context.tool_input,
+        'agent': context.agent.role if context.agent else None
+    })
+    return None
+```
+
+## Troubleshooting
+
+### Hook Not Executing
+- Verify hook is registered before crew execution
+- Check if previous hook returned `False` (blocks execution and subsequent hooks)
+- Ensure hook signature matches expected type
+
+### Input Modifications Not Working
+- Use in-place modifications: `context.tool_input['key'] = value`
+- Don't replace the dict: `context.tool_input = {}`
+
+### Result Modifications Not Working
+- Return the modified string from after hooks
+- Returning `None` keeps the original result
+- Ensure the tool actually returned a result
+
+### Tool Blocked Unexpectedly
+- Check all before hooks for blocking conditions
+- Verify hook execution order
+- Add debug logging to identify which hook is blocking
+
+## Conclusion
+
+Tool Call Hooks provide powerful capabilities for controlling and monitoring tool execution in CrewAI. Use them to implement safety guardrails, approval gates, input validation, result sanitization, logging, and analytics. Combined with proper error handling and type safety, hooks enable secure and production-ready agent systems with comprehensive observability.

docs/en/observability/portkey.mdx

@@ -733,9 +733,7 @@ Here's a basic configuration to route requests to OpenAI, specifically using GPT
     - Collect relevant metadata to filter logs
     - Enforce access permissions
 
-    Create API keys through:
-    - [Portkey App](https://app.portkey.ai/)
-    - [API Key Management API](/en/api-reference/admin-api/control-plane/api-keys/create-api-key)
+    Create API keys through the [Portkey App](https://app.portkey.ai/)
 
     Example using Python SDK:
     ```python
@@ -758,7 +756,7 @@ Here's a basic configuration to route requests to OpenAI, specifically using GPT
     )
     ```
 
-    For detailed key management instructions, see our [API Keys documentation](/en/api-reference/admin-api/control-plane/api-keys/create-api-key).
+    For detailed key management instructions, see the [Portkey documentation](https://portkey.ai/docs).
   </Accordion>
 
   <Accordion title="Step 4: Deploy & Monitor">

docs/en/tools/cloud-storage/overview.mdx

@@ -18,7 +18,7 @@ These tools enable your agents to interact with cloud services, access cloud sto
     Write and upload files to Amazon S3 storage.
   </Card>
 
-  <Card title="Bedrock Invoke Agent" icon="aws" href="/en/tools/cloud-storage/bedrockinvokeagenttool">
+  <Card title="Bedrock Invoke Agent" icon="aws" href="/en/tools/integration/bedrockinvokeagenttool">
     Invoke Amazon Bedrock agents for AI-powered tasks.
   </Card>
 

docs/ko/changelog.mdx

@@ -632,11 +632,11 @@ mode: "wide"
 
   ## 기여
 
-  기여를 원하시면, [기여 가이드](CONTRIBUTING.md)를 참조하세요.
+  기여를 원하시면, [기여 가이드](https://github.com/crewAIInc/crewAI/blob/main/CONTRIBUTING.md)를 참조하세요.
 
   ## 라이센스
 
-  이 프로젝트는 MIT 라이센스 하에 배포됩니다. 자세한 내용은 [LICENSE](LICENSE) 파일을 확인하세요.
+  이 프로젝트는 MIT 라이센스 하에 배포됩니다. 자세한 내용은 [LICENSE](https://github.com/crewAIInc/crewAI/blob/main/LICENSE) 파일을 확인하세요.
 </Update>
 
 <Update label="2025년 5월 22일">

docs/ko/concepts/knowledge.mdx

@@ -706,7 +706,7 @@ class KnowledgeMonitorListener(BaseEventListener):
 knowledge_monitor = KnowledgeMonitorListener()
 ```
 
-이벤트 사용에 대한 자세한 내용은 [이벤트 리스너](https://docs.crewai.com/concepts/event-listener) 문서를 참고하세요.
+이벤트 사용에 대한 자세한 내용은 [이벤트 리스너](/ko/concepts/event-listener) 문서를 참고하세요.
 
 ### 맞춤형 지식 소스
 

docs/ko/concepts/llms.mdx

@@ -748,7 +748,7 @@ CrewAI는 LLM의 스트리밍 응답을 지원하여, 애플리케이션이 출
     ```
 
     <Tip>
-      [자세한 내용은 여기를 클릭하세요](https://docs.crewai.com/concepts/event-listener#event-listeners)
+      [자세한 내용은 여기를 클릭하세요](/ko/concepts/event-listener#event-listeners)
     </Tip>
   </Tab>
 

docs/ko/enterprise/features/marketplace.mdx

@@ -36,7 +36,7 @@ mode: "wide"
   <Card title="도구 & 통합" href="/ko/enterprise/features/tools-and-integrations" icon="wrench">
     에이전트가 사용할 외부 앱 연결 및 내부 도구 관리.
   </Card>
-  <Card title="도구 저장소" href="/ko/enterprise/features/tool-repository" icon="toolbox">
+  <Card title="도구 저장소" href="/ko/enterprise/guides/tool-repository" icon="toolbox">
     크루 기능을 확장할 수 있도록 도구를 게시하고 설치.
   </Card>
   <Card title="에이전트 저장소" href="/ko/enterprise/features/agent-repositories" icon="people-group">

docs/ko/enterprise/features/tools-and-integrations.mdx

@@ -231,7 +231,7 @@ mode: "wide"
 ## 관련 문서
 
 <CardGroup cols={2}>
-  <Card title="도구 저장소" href="/ko/enterprise/features/tool-repository" icon="toolbox">
+  <Card title="도구 저장소" href="/ko/enterprise/guides/tool-repository" icon="toolbox">
     크루 기능을 확장할 수 있도록 도구를 게시하고 설치하세요.
   </Card>
   <Card title="Webhook 자동화" href="/ko/enterprise/guides/webhook-automation" icon="bolt">

docs/ko/enterprise/guides/tool-repository.mdx

@@ -21,7 +21,7 @@ Tool Repository는 CrewAI 도구를 위한 패키지 관리자입니다. 사용
 Tool Repository를 사용하기 전에 다음이 준비되어 있어야 합니다:
 
 - [CrewAI AMP](https://app.crewai.com) 계정
-- [CrewAI CLI](https://docs.crewai.com/concepts/cli#cli) 설치됨
+- [CrewAI CLI](/ko/concepts/cli#cli) 설치됨
 - uv>=0.5.0 이 설치되어 있어야 합니다. [업그레이드 방법](https://docs.astral.sh/uv/getting-started/installation/#upgrading-uv)을 참고하세요.
 - [Git](https://git-scm.com) 설치 및 구성 완료
 - CrewAI AMP 조직에서 도구를 게시하거나 설치할 수 있는 액세스 권한
@@ -66,7 +66,7 @@ crewai tool publish
 crewai tool publish --public
 ```
 
-도구 빌드에 대한 자세한 내용은 [나만의 도구 만들기](https://docs.crewai.com/concepts/tools#creating-your-own-tools)를 참고하세요.
+도구 빌드에 대한 자세한 내용은 [나만의 도구 만들기](/ko/concepts/tools#creating-your-own-tools)를 참고하세요.
 
 ## 도구 업데이트
 

docs/ko/enterprise/resources/frequently-asked-questions.mdx

@@ -49,7 +49,7 @@ mode: "wide"
 
         에이전트 실행에 인간 입력을 통합하려면 작업 정의에서 `human_input` 플래그를 설정하세요. 활성화하면, 에이전트가 최종 답변을 제공하기 전에 사용자에게 입력을 요청합니다. 이 입력은 추가 맥락을 제공하거나, 애매함을 해소하거나, 에이전트의 출력을 검증해야 할 때 활용될 수 있습니다.
 
-        자세한 구현 방법은 [Human-in-the-Loop 가이드](/ko/how-to/human-in-the-loop)를 참고해 주세요.
+        자세한 구현 방법은 [Human-in-the-Loop 가이드](/ko/enterprise/guides/human-in-the-loop)를 참고해 주세요.
     </Accordion>
 
     <Accordion title="CrewAI에서 에이전트의 행동과 역량을 맞춤화하고 향상시키기 위한 고급 커스터마이징 옵션에는 어떤 것이 있나요?">
@@ -142,7 +142,7 @@ mode: "wide"
     <Accordion title="CrewAI 에이전트를 위한 커스텀 도구는 어떻게 만들 수 있습니까?">
         CrewAI에서 제공하는 `BaseTool` 클래스를 상속받아 커스텀 도구를 직접 만들거나, tool 데코레이터를 활용할 수 있습니다. 상속 방식은 `BaseTool`을 상속하는 새로운 클래스를 정의해 이름, 설명, 그리고 실제 논리를 처리하는 `_run` 메서드를 작성합니다. tool 데코레이터를 사용하면 필수 속성과 운영 로직만 정의해 바로 `Tool` 객체를 만들 수 있습니다.
 
-        <Card href="https://docs.crewai.com/how-to/create-custom-tools" icon="code">CrewAI 도구 가이드</Card>
+        <Card href="/ko/learn/create-custom-tools" icon="code">CrewAI 도구 가이드</Card>
     </Accordion>
 
     <Accordion title="전체 crew가 수행할 수 있는 분당 최대 요청 수는 어떻게 제한할 수 있나요?">

docs/ko/learn/execution-hooks.mdx

@@ -0,0 +1,379 @@
+---
+title: 실행 훅 개요
+description: 에이전트 작업에 대한 세밀한 제어를 위한 CrewAI 실행 훅 이해 및 사용
+mode: "wide"
+---
+
+실행 훅(Execution Hooks)은 CrewAI 에이전트의 런타임 동작을 세밀하게 제어할 수 있게 해줍니다. 크루 실행 전후에 실행되는 킥오프 훅과 달리, 실행 훅은 에이전트 실행 중 특정 작업을 가로채서 동작을 수정하고, 안전성 검사를 구현하며, 포괄적인 모니터링을 추가할 수 있습니다.
+
+## 실행 훅의 유형
+
+CrewAI는 두 가지 주요 범주의 실행 훅을 제공합니다:
+
+### 1. [LLM 호출 훅](/learn/llm-hooks)
+
+언어 모델 상호작용을 제어하고 모니터링합니다:
+- **LLM 호출 전**: 프롬프트 수정, 입력 검증, 승인 게이트 구현
+- **LLM 호출 후**: 응답 변환, 출력 정제, 대화 기록 업데이트
+
+**사용 사례:**
+- 반복 제한
+- 비용 추적 및 토큰 사용량 모니터링
+- 응답 정제 및 콘텐츠 필터링
+- LLM 호출에 대한 사람의 승인
+- 안전 가이드라인 또는 컨텍스트 추가
+- 디버그 로깅 및 요청/응답 검사
+
+[LLM 훅 문서 보기 →](/learn/llm-hooks)
+
+### 2. [도구 호출 훅](/learn/tool-hooks)
+
+도구 실행을 제어하고 모니터링합니다:
+- **도구 호출 전**: 입력 수정, 매개변수 검증, 위험한 작업 차단
+- **도구 호출 후**: 결과 변환, 출력 정제, 실행 세부사항 로깅
+
+**사용 사례:**
+- 파괴적인 작업에 대한 안전 가드레일
+- 민감한 작업에 대한 사람의 승인
+- 입력 검증 및 정제
+- 결과 캐싱 및 속도 제한
+- 도구 사용 분석
+- 디버그 로깅 및 모니터링
+
+[도구 훅 문서 보기 →](/learn/tool-hooks)
+
+## 훅 등록 방법
+
+### 1. 데코레이터 기반 훅 (권장)
+
+훅을 등록하는 가장 깔끔하고 파이썬스러운 방법:
+
+```python
+from crewai.hooks import before_llm_call, after_llm_call, before_tool_call, after_tool_call
+
+@before_llm_call
+def limit_iterations(context):
+    """반복 횟수를 제한하여 무한 루프를 방지합니다."""
+    if context.iterations > 10:
+        return False  # 실행 차단
+    return None
+
+@after_llm_call
+def sanitize_response(context):
+    """LLM 응답에서 민감한 데이터를 제거합니다."""
+    if "API_KEY" in context.response:
+        return context.response.replace("API_KEY", "[수정됨]")
+    return None
+
+@before_tool_call
+def block_dangerous_tools(context):
+    """파괴적인 작업을 차단합니다."""
+    if context.tool_name == "delete_database":
+        return False  # 실행 차단
+    return None
+
+@after_tool_call
+def log_tool_result(context):
+    """도구 실행을 로깅합니다."""
+    print(f"도구 {context.tool_name} 완료")
+    return None
+```
+
+### 2. 크루 범위 훅
+
+특정 크루 인스턴스에만 훅을 적용합니다:
+
+```python
+from crewai import CrewBase
+from crewai.project import crew
+from crewai.hooks import before_llm_call_crew, after_tool_call_crew
+
+@CrewBase
+class MyProjCrew:
+    @before_llm_call_crew
+    def validate_inputs(self, context):
+        # 이 크루에만 적용됩니다
+        print(f"{self.__class__.__name__}에서 LLM 호출")
+        return None
+
+    @after_tool_call_crew
+    def log_results(self, context):
+        # 크루별 로깅
+        print(f"도구 결과: {context.tool_result[:50]}...")
+        return None
+
+    @crew
+    def crew(self) -> Crew:
+        return Crew(
+            agents=self.agents,
+            tasks=self.tasks,
+            process=Process.sequential
+        )
+```
+
+## 훅 실행 흐름
+
+### LLM 호출 흐름
+
+```
+에이전트가 LLM을 호출해야 함
+    ↓
+[LLM 호출 전 훅 실행]
+    ├→ 훅 1: 반복 횟수 검증
+    ├→ 훅 2: 안전 컨텍스트 추가
+    └→ 훅 3: 요청 로깅
+    ↓
+훅이 False를 반환하는 경우:
+    ├→ LLM 호출 차단
+    └→ ValueError 발생
+    ↓
+모든 훅이 True/None을 반환하는 경우:
+    ├→ LLM 호출 진행
+    └→ 응답 생성
+    ↓
+[LLM 호출 후 훅 실행]
+    ├→ 훅 1: 응답 정제
+    ├→ 훅 2: 응답 로깅
+    └→ 훅 3: 메트릭 업데이트
+    ↓
+최종 응답 반환
+```
+
+### 도구 호출 흐름
+
+```
+에이전트가 도구를 실행해야 함
+    ↓
+[도구 호출 전 훅 실행]
+    ├→ 훅 1: 도구 허용 여부 확인
+    ├→ 훅 2: 입력 검증
+    └→ 훅 3: 필요시 승인 요청
+    ↓
+훅이 False를 반환하는 경우:
+    ├→ 도구 실행 차단
+    └→ 오류 메시지 반환
+    ↓
+모든 훅이 True/None을 반환하는 경우:
+    ├→ 도구 실행 진행
+    └→ 결과 생성
+    ↓
+[도구 호출 후 훅 실행]
+    ├→ 훅 1: 결과 정제
+    ├→ 훅 2: 결과 캐싱
+    └→ 훅 3: 메트릭 로깅
+    ↓
+최종 결과 반환
+```
+
+## 훅 컨텍스트 객체
+
+### LLMCallHookContext
+
+LLM 실행 상태에 대한 액세스를 제공합니다:
+
+```python
+class LLMCallHookContext:
+    executor: CrewAgentExecutor  # 전체 실행자 액세스
+    messages: list               # 변경 가능한 메시지 목록
+    agent: Agent                 # 현재 에이전트
+    task: Task                   # 현재 작업
+    crew: Crew                   # 크루 인스턴스
+    llm: BaseLLM                 # LLM 인스턴스
+    iterations: int              # 현재 반복 횟수
+    response: str | None         # LLM 응답 (후 훅용)
+```
+
+### ToolCallHookContext
+
+도구 실행 상태에 대한 액세스를 제공합니다:
+
+```python
+class ToolCallHookContext:
+    tool_name: str               # 호출되는 도구
+    tool_input: dict             # 변경 가능한 입력 매개변수
+    tool: CrewStructuredTool     # 도구 인스턴스
+    agent: Agent | None          # 실행 중인 에이전트
+    task: Task | None            # 현재 작업
+    crew: Crew | None            # 크루 인스턴스
+    tool_result: str | None      # 도구 결과 (후 훅용)
+```
+
+## 일반적인 패턴
+
+### 안전 및 검증
+
+```python
+@before_tool_call
+def safety_check(context):
+    """파괴적인 작업을 차단합니다."""
+    dangerous = ['delete_file', 'drop_table', 'system_shutdown']
+    if context.tool_name in dangerous:
+        print(f"🛑 차단됨: {context.tool_name}")
+        return False
+    return None
+
+@before_llm_call
+def iteration_limit(context):
+    """무한 루프를 방지합니다."""
+    if context.iterations > 15:
+        print("⛔ 최대 반복 횟수 초과")
+        return False
+    return None
+```
+
+### 사람의 개입
+
+```python
+@before_tool_call
+def require_approval(context):
+    """민감한 작업에 대한 승인을 요구합니다."""
+    sensitive = ['send_email', 'make_payment', 'post_message']
+
+    if context.tool_name in sensitive:
+        response = context.request_human_input(
+            prompt=f"{context.tool_name} 승인하시겠습니까?",
+            default_message="승인하려면 'yes'를 입력하세요:"
+        )
+
+        if response.lower() != 'yes':
+            return False
+
+    return None
+```
+
+### 모니터링 및 분석
+
+```python
+from collections import defaultdict
+import time
+
+metrics = defaultdict(lambda: {'count': 0, 'total_time': 0})
+
+@before_tool_call
+def start_timer(context):
+    context.tool_input['_start'] = time.time()
+    return None
+
+@after_tool_call
+def track_metrics(context):
+    start = context.tool_input.get('_start', time.time())
+    duration = time.time() - start
+
+    metrics[context.tool_name]['count'] += 1
+    metrics[context.tool_name]['total_time'] += duration
+
+    return None
+```
+
+## 훅 관리
+
+### 모든 훅 지우기
+
+```python
+from crewai.hooks import clear_all_global_hooks
+
+# 모든 훅을 한 번에 지웁니다
+result = clear_all_global_hooks()
+print(f"{result['total']} 훅이 지워졌습니다")
+```
+
+### 특정 훅 유형 지우기
+
+```python
+from crewai.hooks import (
+    clear_before_llm_call_hooks,
+    clear_after_llm_call_hooks,
+    clear_before_tool_call_hooks,
+    clear_after_tool_call_hooks
+)
+
+# 특정 유형 지우기
+llm_before_count = clear_before_llm_call_hooks()
+tool_after_count = clear_after_tool_call_hooks()
+```
+
+## 모범 사례
+
+### 1. 훅을 집중적으로 유지
+각 훅은 단일하고 명확한 책임을 가져야 합니다.
+
+### 2. 오류를 우아하게 처리
+```python
+@before_llm_call
+def safe_hook(context):
+    try:
+        if some_condition:
+            return False
+    except Exception as e:
+        print(f"훅 오류: {e}")
+        return None  # 오류에도 불구하고 실행 허용
+```
+
+### 3. 컨텍스트를 제자리에서 수정
+```python
+# ✅ 올바름 - 제자리에서 수정
+@before_llm_call
+def add_context(context):
+    context.messages.append({"role": "system", "content": "간결하게"})
+
+# ❌ 잘못됨 - 참조를 교체
+@before_llm_call
+def wrong_approach(context):
+    context.messages = [{"role": "system", "content": "간결하게"}]
+```
+
+### 4. 타입 힌트 사용
+```python
+from crewai.hooks import LLMCallHookContext, ToolCallHookContext
+
+def my_llm_hook(context: LLMCallHookContext) -> bool | None:
+    return None
+
+def my_tool_hook(context: ToolCallHookContext) -> str | None:
+    return None
+```
+
+### 5. 테스트에서 정리
+```python
+import pytest
+from crewai.hooks import clear_all_global_hooks
+
+@pytest.fixture(autouse=True)
+def clean_hooks():
+    """각 테스트 전에 훅을 재설정합니다."""
+    yield
+    clear_all_global_hooks()
+```
+
+## 어떤 훅을 사용해야 할까요
+
+### LLM 훅을 사용하는 경우:
+- 반복 제한 구현
+- 프롬프트에 컨텍스트 또는 안전 가이드라인 추가
+- 토큰 사용량 및 비용 추적
+- 응답 정제 또는 변환
+- LLM 호출에 대한 승인 게이트 구현
+- 프롬프트/응답 상호작용 디버깅
+
+### 도구 훅을 사용하는 경우:
+- 위험하거나 파괴적인 작업 차단
+- 실행 전 도구 입력 검증
+- 민감한 작업에 대한 승인 게이트 구현
+- 도구 결과 캐싱
+- 도구 사용 및 성능 추적
+- 도구 출력 정제
+- 도구 호출 속도 제한
+
+### 둘 다 사용하는 경우:
+모든 에이전트 작업을 모니터링해야 하는 포괄적인 관찰성, 안전 또는 승인 시스템을 구축하는 경우.
+
+## 관련 문서
+
+- [LLM 호출 훅 →](/learn/llm-hooks) - 상세한 LLM 훅 문서
+- [도구 호출 훅 →](/learn/tool-hooks) - 상세한 도구 훅 문서
+- [킥오프 전후 훅 →](/learn/before-and-after-kickoff-hooks) - 크루 생명주기 훅
+- [사람의 개입 →](/learn/human-in-the-loop) - 사람 입력 패턴
+
+## 결론
+
+실행 훅은 에이전트 런타임 동작에 대한 강력한 제어를 제공합니다. 이를 사용하여 안전 가드레일, 승인 워크플로우, 포괄적인 모니터링 및 사용자 정의 비즈니스 로직을 구현하세요. 적절한 오류 처리, 타입 안전성 및 성능 고려사항과 결합하면, 훅을 통해 프로덕션 준비가 된 안전하고 관찰 가능한 에이전트 시스템을 구축할 수 있습니다.

docs/ko/learn/hierarchical-process.mdx

@@ -95,7 +95,7 @@ project_crew = Crew(
 ```
 
 <Tip>
-    매니저 에이전트 생성 및 맞춤화에 대한 자세한 내용은 [커스텀 매니저 에이전트 문서](https://docs.crewai.com/how-to/custom-manager-agent#custom-manager-agent)를 참고하세요.
+    매니저 에이전트 생성 및 맞춤화에 대한 자세한 내용은 [커스텀 매니저 에이전트 문서](/ko/learn/custom-manager-agent)를 참고하세요.
 </Tip>
 
 ### 워크플로우 실행

docs/ko/learn/llm-hooks.mdx

@@ -0,0 +1,412 @@
+---
+title: LLM 호출 훅
+description: CrewAI에서 언어 모델 상호작용을 가로채고, 수정하고, 제어하는 LLM 호출 훅 사용 방법 배우기
+mode: "wide"
+---
+
+LLM 호출 훅(LLM Call Hooks)은 에이전트 실행 중 언어 모델 상호작용에 대한 세밀한 제어를 제공합니다. 이러한 훅을 사용하면 LLM 호출을 가로채고, 프롬프트를 수정하고, 응답을 변환하고, 승인 게이트를 구현하고, 사용자 정의 로깅 또는 모니터링을 추가할 수 있습니다.
+
+## 개요
+
+LLM 훅은 두 가지 중요한 시점에 실행됩니다:
+- **LLM 호출 전**: 메시지 수정, 입력 검증 또는 실행 차단
+- **LLM 호출 후**: 응답 변환, 출력 정제 또는 대화 기록 수정
+
+## 훅 타입
+
+### LLM 호출 전 훅
+
+모든 LLM 호출 전에 실행되며, 다음을 수행할 수 있습니다:
+- LLM에 전송되는 메시지 검사 및 수정
+- 조건에 따라 LLM 실행 차단
+- 속도 제한 또는 승인 게이트 구현
+- 컨텍스트 또는 시스템 메시지 추가
+- 요청 세부사항 로깅
+
+**시그니처:**
+```python
+def before_hook(context: LLMCallHookContext) -> bool | None:
+    # 실행을 차단하려면 False 반환
+    # 실행을 허용하려면 True 또는 None 반환
+    ...
+```
+
+### LLM 호출 후 훅
+
+모든 LLM 호출 후에 실행되며, 다음을 수행할 수 있습니다:
+- LLM 응답 수정 또는 정제
+- 메타데이터 또는 서식 추가
+- 응답 세부사항 로깅
+- 대화 기록 업데이트
+- 콘텐츠 필터링 구현
+
+**시그니처:**
+```python
+def after_hook(context: LLMCallHookContext) -> str | None:
+    # 수정된 응답 문자열 반환
+    # 원본 응답을 유지하려면 None 반환
+    ...
+```
+
+## LLM 훅 컨텍스트
+
+`LLMCallHookContext` 객체는 실행 상태에 대한 포괄적인 액세스를 제공합니다:
+
+```python
+class LLMCallHookContext:
+    executor: CrewAgentExecutor  # 전체 실행자 참조
+    messages: list               # 변경 가능한 메시지 목록
+    agent: Agent                 # 현재 에이전트
+    task: Task                   # 현재 작업
+    crew: Crew                   # 크루 인스턴스
+    llm: BaseLLM                 # LLM 인스턴스
+    iterations: int              # 현재 반복 횟수
+    response: str | None         # LLM 응답 (후 훅용)
+```
+
+### 메시지 수정
+
+**중요:** 항상 메시지를 제자리에서 수정하세요:
+
+```python
+# ✅ 올바름 - 제자리에서 수정
+def add_context(context: LLMCallHookContext) -> None:
+    context.messages.append({"role": "system", "content": "간결하게 작성하세요"})
+
+# ❌ 잘못됨 - 리스트 참조를 교체
+def wrong_approach(context: LLMCallHookContext) -> None:
+    context.messages = [{"role": "system", "content": "간결하게 작성하세요"}]
+```
+
+## 등록 방법
+
+### 1. 데코레이터 기반 등록 (권장)
+
+더 깔끔한 구문을 위해 데코레이터를 사용합니다:
+
+```python
+from crewai.hooks import before_llm_call, after_llm_call
+
+@before_llm_call
+def validate_iteration_count(context):
+    """반복 횟수를 검증합니다."""
+    if context.iterations > 10:
+        print("⚠️ 최대 반복 횟수 초과")
+        return False  # 실행 차단
+    return None
+
+@after_llm_call
+def sanitize_response(context):
+    """민감한 데이터를 제거합니다."""
+    if context.response and "API_KEY" in context.response:
+        return context.response.replace("API_KEY", "[수정됨]")
+    return None
+```
+
+### 2. 크루 범위 훅
+
+특정 크루 인스턴스에 대한 훅을 등록합니다:
+
+```python
+from crewai import CrewBase
+from crewai.project import crew
+from crewai.hooks import before_llm_call_crew, after_llm_call_crew
+
+@CrewBase
+class MyProjCrew:
+    @before_llm_call_crew
+    def validate_inputs(self, context):
+        # 이 크루에만 적용됩니다
+        if context.iterations == 0:
+            print(f"작업 시작: {context.task.description}")
+        return None
+    
+    @after_llm_call_crew
+    def log_responses(self, context):
+        # 크루별 응답 로깅
+        print(f"응답 길이: {len(context.response)}")
+        return None
+    
+    @crew
+    def crew(self) -> Crew:
+        return Crew(
+            agents=self.agents,
+            tasks=self.tasks,
+            process=Process.sequential,
+            verbose=True
+        )
+```
+
+## 일반적인 사용 사례
+
+### 1. 반복 제한
+
+```python
+@before_llm_call
+def limit_iterations(context: LLMCallHookContext) -> bool | None:
+    """무한 루프를 방지하기 위해 반복을 제한합니다."""
+    max_iterations = 15
+    if context.iterations > max_iterations:
+        print(f"⛔ 차단됨: {max_iterations}회 반복 초과")
+        return False  # 실행 차단
+    return None
+```
+
+### 2. 사람의 승인 게이트
+
+```python
+@before_llm_call
+def require_approval(context: LLMCallHookContext) -> bool | None:
+    """특정 반복 후 승인을 요구합니다."""
+    if context.iterations > 5:
+        response = context.request_human_input(
+            prompt=f"반복 {context.iterations}: LLM 호출을 승인하시겠습니까?",
+            default_message="승인하려면 Enter를 누르고, 차단하려면 'no'를 입력하세요:"
+        )
+        if response.lower() == "no":
+            print("🚫 사용자에 의해 LLM 호출이 차단되었습니다")
+            return False
+    return None
+```
+
+### 3. 시스템 컨텍스트 추가
+
+```python
+@before_llm_call
+def add_guardrails(context: LLMCallHookContext) -> None:
+    """모든 LLM 호출에 안전 가이드라인을 추가합니다."""
+    context.messages.append({
+        "role": "system",
+        "content": "응답이 사실에 기반하고 가능한 경우 출처를 인용하도록 하세요."
+    })
+    return None
+```
+
+### 4. 응답 정제
+
+```python
+@after_llm_call
+def sanitize_sensitive_data(context: LLMCallHookContext) -> str | None:
+    """민감한 데이터 패턴을 제거합니다."""
+    if not context.response:
+        return None
+    
+    import re
+    sanitized = context.response
+    sanitized = re.sub(r'\b\d{3}-\d{2}-\d{4}\b', '[주민번호-수정됨]', sanitized)
+    sanitized = re.sub(r'\b\d{4}[- ]?\d{4}[- ]?\d{4}[- ]?\d{4}\b', '[카드번호-수정됨]', sanitized)
+    
+    return sanitized
+```
+
+### 5. 비용 추적
+
+```python
+import tiktoken
+
+@before_llm_call
+def track_token_usage(context: LLMCallHookContext) -> None:
+    """입력 토큰을 추적합니다."""
+    encoding = tiktoken.get_encoding("cl100k_base")
+    total_tokens = sum(
+        len(encoding.encode(msg.get("content", ""))) 
+        for msg in context.messages
+    )
+    print(f"📊 입력 토큰: ~{total_tokens}")
+    return None
+
+@after_llm_call
+def track_response_tokens(context: LLMCallHookContext) -> None:
+    """응답 토큰을 추적합니다."""
+    if context.response:
+        encoding = tiktoken.get_encoding("cl100k_base")
+        tokens = len(encoding.encode(context.response))
+        print(f"📊 응답 토큰: ~{tokens}")
+    return None
+```
+
+### 6. 디버그 로깅
+
+```python
+@before_llm_call
+def debug_request(context: LLMCallHookContext) -> None:
+    """LLM 요청을 디버그합니다."""
+    print(f"""
+    🔍 LLM 호출 디버그:
+    - 에이전트: {context.agent.role}
+    - 작업: {context.task.description[:50]}...
+    - 반복: {context.iterations}
+    - 메시지 수: {len(context.messages)}
+    - 마지막 메시지: {context.messages[-1] if context.messages else 'None'}
+    """)
+    return None
+
+@after_llm_call
+def debug_response(context: LLMCallHookContext) -> None:
+    """LLM 응답을 디버그합니다."""
+    if context.response:
+        print(f"✅ 응답 미리보기: {context.response[:100]}...")
+    return None
+```
+
+## 훅 관리
+
+### 훅 등록 해제
+
+```python
+from crewai.hooks import (
+    unregister_before_llm_call_hook,
+    unregister_after_llm_call_hook
+)
+
+# 특정 훅 등록 해제
+def my_hook(context):
+    ...
+
+register_before_llm_call_hook(my_hook)
+# 나중에...
+unregister_before_llm_call_hook(my_hook)  # 찾으면 True 반환
+```
+
+### 훅 지우기
+
+```python
+from crewai.hooks import (
+    clear_before_llm_call_hooks,
+    clear_after_llm_call_hooks,
+    clear_all_llm_call_hooks
+)
+
+# 특정 훅 타입 지우기
+count = clear_before_llm_call_hooks()
+print(f"{count}개의 전(before) 훅이 지워졌습니다")
+
+# 모든 LLM 훅 지우기
+before_count, after_count = clear_all_llm_call_hooks()
+print(f"{before_count}개의 전(before) 훅과 {after_count}개의 후(after) 훅이 지워졌습니다")
+```
+
+## 고급 패턴
+
+### 조건부 훅 실행
+
+```python
+@before_llm_call
+def conditional_blocking(context: LLMCallHookContext) -> bool | None:
+    """특정 조건에서만 차단합니다."""
+    # 특정 에이전트에 대해서만 차단
+    if context.agent.role == "researcher" and context.iterations > 10:
+        return False
+    
+    # 특정 작업에 대해서만 차단
+    if "민감한" in context.task.description.lower() and context.iterations > 5:
+        return False
+    
+    return None
+```
+
+### 컨텍스트 인식 수정
+
+```python
+@before_llm_call
+def adaptive_prompting(context: LLMCallHookContext) -> None:
+    """반복에 따라 다른 컨텍스트를 추가합니다."""
+    if context.iterations == 0:
+        context.messages.append({
+            "role": "system",
+            "content": "높은 수준의 개요부터 시작하세요."
+        })
+    elif context.iterations > 3:
+        context.messages.append({
+            "role": "system",
+            "content": "구체적인 세부사항에 집중하고 예제를 제공하세요."
+        })
+    return None
+```
+
+### 훅 체이닝
+
+```python
+# 여러 훅은 등록 순서대로 실행됩니다
+
+@before_llm_call
+def first_hook(context):
+    print("1. 첫 번째 훅 실행됨")
+    return None
+
+@before_llm_call
+def second_hook(context):
+    print("2. 두 번째 훅 실행됨")
+    return None
+
+@before_llm_call
+def blocking_hook(context):
+    if context.iterations > 10:
+        print("3. 차단 훅 - 실행 중지")
+        return False  # 후속 훅은 실행되지 않습니다
+    print("3. 차단 훅 - 실행 허용")
+    return None
+```
+
+## 모범 사례
+
+1. **훅을 집중적으로 유지**: 각 훅은 단일 책임을 가져야 합니다
+2. **무거운 계산 피하기**: 훅은 모든 LLM 호출마다 실행됩니다
+3. **오류를 우아하게 처리**: try-except를 사용하여 훅 실패로 인한 실행 중단 방지
+4. **타입 힌트 사용**: 더 나은 IDE 지원을 위해 `LLMCallHookContext` 활용
+5. **훅 동작 문서화**: 특히 차단 조건에 대해
+6. **훅을 독립적으로 테스트**: 프로덕션에서 사용하기 전에 단위 테스트
+7. **테스트에서 훅 지우기**: 테스트 실행 간 `clear_all_llm_call_hooks()` 사용
+8. **제자리에서 수정**: 항상 `context.messages`를 제자리에서 수정하고 교체하지 마세요
+
+## 오류 처리
+
+```python
+@before_llm_call
+def safe_hook(context: LLMCallHookContext) -> bool | None:
+    try:
+        # 훅 로직
+        if some_condition:
+            return False
+    except Exception as e:
+        print(f"⚠️ 훅 오류: {e}")
+        # 결정: 오류 발생 시 허용 또는 차단
+        return None  # 오류에도 불구하고 실행 허용
+```
+
+## 타입 안전성
+
+```python
+from crewai.hooks import LLMCallHookContext, BeforeLLMCallHookType, AfterLLMCallHookType
+
+# 명시적 타입 주석
+def my_before_hook(context: LLMCallHookContext) -> bool | None:
+    return None
+
+def my_after_hook(context: LLMCallHookContext) -> str | None:
+    return None
+
+# 타입 안전 등록
+register_before_llm_call_hook(my_before_hook)
+register_after_llm_call_hook(my_after_hook)
+```
+
+## 문제 해결
+
+### 훅이 실행되지 않음
+- 크루 실행 전에 훅이 등록되었는지 확인
+- 이전 훅이 `False`를 반환했는지 확인 (후속 훅 차단)
+- 훅 시그니처가 예상 타입과 일치하는지 확인
+
+### 메시지 수정이 지속되지 않음
+- 제자리 수정 사용: `context.messages.append()`
+- 리스트를 교체하지 마세요: `context.messages = []`
+
+### 응답 수정이 작동하지 않음
+- 후 훅에서 수정된 문자열을 반환
+- `None`을 반환하면 원본 응답이 유지됩니다
+
+## 결론
+
+LLM 호출 훅은 CrewAI에서 언어 모델 상호작용을 제어하고 모니터링하는 강력한 기능을 제공합니다. 이를 사용하여 안전 가드레일, 승인 게이트, 로깅, 비용 추적 및 응답 정제를 구현하세요. 적절한 오류 처리 및 타입 안전성과 결합하면, 훅을 통해 강력하고 프로덕션 준비가 된 에이전트 시스템을 구축할 수 있습니다.
+

docs/ko/learn/tool-hooks.mdx

@@ -0,0 +1,498 @@
+---
+title: 도구 호출 훅
+description: CrewAI에서 도구 실행을 가로채고, 수정하고, 제어하는 도구 호출 훅 사용 방법 배우기
+mode: "wide"
+---
+
+도구 호출 훅(Tool Call Hooks)은 에이전트 작업 중 도구 실행에 대한 세밀한 제어를 제공합니다. 이러한 훅을 사용하면 도구 호출을 가로채고, 입력을 수정하고, 출력을 변환하고, 안전 검사를 구현하고, 포괄적인 로깅 또는 모니터링을 추가할 수 있습니다.
+
+## 개요
+
+도구 훅은 두 가지 중요한 시점에 실행됩니다:
+- **도구 호출 전**: 입력 수정, 매개변수 검증 또는 실행 차단
+- **도구 호출 후**: 결과 변환, 출력 정제 또는 실행 세부사항 로깅
+
+## 훅 타입
+
+### 도구 호출 전 훅
+
+모든 도구 실행 전에 실행되며, 다음을 수행할 수 있습니다:
+- 도구 입력 검사 및 수정
+- 조건에 따라 도구 실행 차단
+- 위험한 작업에 대한 승인 게이트 구현
+- 매개변수 검증
+- 도구 호출 로깅
+
+**시그니처:**
+```python
+def before_hook(context: ToolCallHookContext) -> bool | None:
+    # 실행을 차단하려면 False 반환
+    # 실행을 허용하려면 True 또는 None 반환
+    ...
+```
+
+### 도구 호출 후 훅
+
+모든 도구 실행 후에 실행되며, 다음을 수행할 수 있습니다:
+- 도구 결과 수정 또는 정제
+- 메타데이터 또는 서식 추가
+- 실행 결과 로깅
+- 결과 검증 구현
+- 출력 형식 변환
+
+**시그니처:**
+```python
+def after_hook(context: ToolCallHookContext) -> str | None:
+    # 수정된 결과 문자열 반환
+    # 원본 결과를 유지하려면 None 반환
+    ...
+```
+
+## 도구 훅 컨텍스트
+
+`ToolCallHookContext` 객체는 도구 실행 상태에 대한 포괄적인 액세스를 제공합니다:
+
+```python
+class ToolCallHookContext:
+    tool_name: str                    # 호출되는 도구의 이름
+    tool_input: dict[str, Any]        # 변경 가능한 도구 입력 매개변수
+    tool: CrewStructuredTool          # 도구 인스턴스 참조
+    agent: Agent | BaseAgent | None   # 도구를 실행하는 에이전트
+    task: Task | None                 # 현재 작업
+    crew: Crew | None                 # 크루 인스턴스
+    tool_result: str | None           # 도구 결과 (후 훅용)
+```
+
+### 도구 입력 수정
+
+**중요:** 항상 도구 입력을 제자리에서 수정하세요:
+
+```python
+# ✅ 올바름 - 제자리에서 수정
+def sanitize_input(context: ToolCallHookContext) -> None:
+    context.tool_input['query'] = context.tool_input['query'].lower()
+
+# ❌ 잘못됨 - 딕셔너리 참조를 교체
+def wrong_approach(context: ToolCallHookContext) -> None:
+    context.tool_input = {'query': 'new query'}
+```
+
+## 등록 방법
+
+### 1. 데코레이터 기반 등록 (권장)
+
+더 깔끔한 구문을 위해 데코레이터를 사용합니다:
+
+```python
+from crewai.hooks import before_tool_call, after_tool_call
+
+@before_tool_call
+def block_dangerous_tools(context):
+    """위험한 도구를 차단합니다."""
+    dangerous_tools = ['delete_database', 'drop_table', 'rm_rf']
+    if context.tool_name in dangerous_tools:
+        print(f"⛔ 위험한 도구 차단됨: {context.tool_name}")
+        return False  # 실행 차단
+    return None
+
+@after_tool_call
+def sanitize_results(context):
+    """결과를 정제합니다."""
+    if context.tool_result and "password" in context.tool_result.lower():
+        return context.tool_result.replace("password", "[수정됨]")
+    return None
+```
+
+### 2. 크루 범위 훅
+
+특정 크루 인스턴스에 대한 훅을 등록합니다:
+
+```python
+from crewai import CrewBase
+from crewai.project import crew
+from crewai.hooks import before_tool_call_crew, after_tool_call_crew
+
+@CrewBase
+class MyProjCrew:
+    @before_tool_call_crew
+    def validate_tool_inputs(self, context):
+        # 이 크루에만 적용됩니다
+        if context.tool_name == "web_search":
+            if not context.tool_input.get('query'):
+                print("❌ 잘못된 검색 쿼리")
+                return False
+        return None
+    
+    @after_tool_call_crew
+    def log_tool_results(self, context):
+        # 크루별 도구 로깅
+        print(f"✅ {context.tool_name} 완료됨")
+        return None
+    
+    @crew
+    def crew(self) -> Crew:
+        return Crew(
+            agents=self.agents,
+            tasks=self.tasks,
+            process=Process.sequential,
+            verbose=True
+        )
+```
+
+## 일반적인 사용 사례
+
+### 1. 안전 가드레일
+
+```python
+@before_tool_call
+def safety_check(context: ToolCallHookContext) -> bool | None:
+    """해를 끼칠 수 있는 도구를 차단합니다."""
+    destructive_tools = [
+        'delete_file',
+        'drop_table',
+        'remove_user',
+        'system_shutdown'
+    ]
+    
+    if context.tool_name in destructive_tools:
+        print(f"🛑 파괴적인 도구 차단됨: {context.tool_name}")
+        return False
+    
+    # 민감한 작업에 대해 경고
+    sensitive_tools = ['send_email', 'post_to_social_media', 'charge_payment']
+    if context.tool_name in sensitive_tools:
+        print(f"⚠️  민감한 도구 실행 중: {context.tool_name}")
+    
+    return None
+```
+
+### 2. 사람의 승인 게이트
+
+```python
+@before_tool_call
+def require_approval_for_actions(context: ToolCallHookContext) -> bool | None:
+    """특정 작업에 대한 승인을 요구합니다."""
+    approval_required = [
+        'send_email',
+        'make_purchase',
+        'delete_file',
+        'post_message'
+    ]
+    
+    if context.tool_name in approval_required:
+        response = context.request_human_input(
+            prompt=f"{context.tool_name}을(를) 승인하시겠습니까?",
+            default_message=f"입력: {context.tool_input}\n승인하려면 'yes'를 입력하세요:"
+        )
+        
+        if response.lower() != 'yes':
+            print(f"❌ 도구 실행 거부됨: {context.tool_name}")
+            return False
+    
+    return None
+```
+
+### 3. 입력 검증 및 정제
+
+```python
+@before_tool_call
+def validate_and_sanitize_inputs(context: ToolCallHookContext) -> bool | None:
+    """입력을 검증하고 정제합니다."""
+    # 검색 쿼리 검증
+    if context.tool_name == 'web_search':
+        query = context.tool_input.get('query', '')
+        if len(query) < 3:
+            print("❌ 검색 쿼리가 너무 짧습니다")
+            return False
+        
+        # 쿼리 정제
+        context.tool_input['query'] = query.strip().lower()
+    
+    # 파일 경로 검증
+    if context.tool_name == 'read_file':
+        path = context.tool_input.get('path', '')
+        if '..' in path or path.startswith('/'):
+            print("❌ 잘못된 파일 경로")
+            return False
+    
+    return None
+```
+
+### 4. 결과 정제
+
+```python
+@after_tool_call
+def sanitize_sensitive_data(context: ToolCallHookContext) -> str | None:
+    """민감한 데이터를 정제합니다."""
+    if not context.tool_result:
+        return None
+    
+    import re
+    result = context.tool_result
+    
+    # API 키 제거
+    result = re.sub(
+        r'(api[_-]?key|token)["\']?\s*[:=]\s*["\']?[\w-]+',
+        r'\1: [수정됨]',
+        result,
+        flags=re.IGNORECASE
+    )
+    
+    # 이메일 주소 제거
+    result = re.sub(
+        r'\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Z|a-z]{2,}\b',
+        '[이메일-수정됨]',
+        result
+    )
+    
+    # 신용카드 번호 제거
+    result = re.sub(
+        r'\b\d{4}[- ]?\d{4}[- ]?\d{4}[- ]?\d{4}\b',
+        '[카드-수정됨]',
+        result
+    )
+    
+    return result
+```
+
+### 5. 도구 사용 분석
+
+```python
+import time
+from collections import defaultdict
+
+tool_stats = defaultdict(lambda: {'count': 0, 'total_time': 0, 'failures': 0})
+
+@before_tool_call
+def start_timer(context: ToolCallHookContext) -> None:
+    context.tool_input['_start_time'] = time.time()
+    return None
+
+@after_tool_call
+def track_tool_usage(context: ToolCallHookContext) -> None:
+    start_time = context.tool_input.get('_start_time', time.time())
+    duration = time.time() - start_time
+    
+    tool_stats[context.tool_name]['count'] += 1
+    tool_stats[context.tool_name]['total_time'] += duration
+    
+    if not context.tool_result or 'error' in context.tool_result.lower():
+        tool_stats[context.tool_name]['failures'] += 1
+    
+    print(f"""
+    📊 {context.tool_name} 도구 통계:
+    - 실행 횟수: {tool_stats[context.tool_name]['count']}
+    - 평균 시간: {tool_stats[context.tool_name]['total_time'] / tool_stats[context.tool_name]['count']:.2f}초
+    - 실패: {tool_stats[context.tool_name]['failures']}
+    """)
+    
+    return None
+```
+
+### 6. 속도 제한
+
+```python
+from collections import defaultdict
+from datetime import datetime, timedelta
+
+tool_call_history = defaultdict(list)
+
+@before_tool_call
+def rate_limit_tools(context: ToolCallHookContext) -> bool | None:
+    """도구 호출 속도를 제한합니다."""
+    tool_name = context.tool_name
+    now = datetime.now()
+    
+    # 오래된 항목 정리 (1분 이상 된 것)
+    tool_call_history[tool_name] = [
+        call_time for call_time in tool_call_history[tool_name]
+        if now - call_time < timedelta(minutes=1)
+    ]
+    
+    # 속도 제한 확인 (분당 최대 10회 호출)
+    if len(tool_call_history[tool_name]) >= 10:
+        print(f"🚫 {tool_name}에 대한 속도 제한 초과")
+        return False
+    
+    # 이 호출 기록
+    tool_call_history[tool_name].append(now)
+    return None
+```
+
+### 7. 디버그 로깅
+
+```python
+@before_tool_call
+def debug_tool_call(context: ToolCallHookContext) -> None:
+    """도구 호출을 디버그합니다."""
+    print(f"""
+    🔍 도구 호출 디버그:
+    - 도구: {context.tool_name}
+    - 에이전트: {context.agent.role if context.agent else '알 수 없음'}
+    - 작업: {context.task.description[:50] if context.task else '알 수 없음'}...
+    - 입력: {context.tool_input}
+    """)
+    return None
+
+@after_tool_call
+def debug_tool_result(context: ToolCallHookContext) -> None:
+    """도구 결과를 디버그합니다."""
+    if context.tool_result:
+        result_preview = context.tool_result[:200]
+        print(f"✅ 결과 미리보기: {result_preview}...")
+    else:
+        print("⚠️  반환된 결과 없음")
+    return None
+```
+
+## 훅 관리
+
+### 훅 등록 해제
+
+```python
+from crewai.hooks import (
+    unregister_before_tool_call_hook,
+    unregister_after_tool_call_hook
+)
+
+# 특정 훅 등록 해제
+def my_hook(context):
+    ...
+
+register_before_tool_call_hook(my_hook)
+# 나중에...
+success = unregister_before_tool_call_hook(my_hook)
+print(f"등록 해제됨: {success}")
+```
+
+### 훅 지우기
+
+```python
+from crewai.hooks import (
+    clear_before_tool_call_hooks,
+    clear_after_tool_call_hooks,
+    clear_all_tool_call_hooks
+)
+
+# 특정 훅 타입 지우기
+count = clear_before_tool_call_hooks()
+print(f"{count}개의 전(before) 훅이 지워졌습니다")
+
+# 모든 도구 훅 지우기
+before_count, after_count = clear_all_tool_call_hooks()
+print(f"{before_count}개의 전(before) 훅과 {after_count}개의 후(after) 훅이 지워졌습니다")
+```
+
+## 고급 패턴
+
+### 조건부 훅 실행
+
+```python
+@before_tool_call
+def conditional_blocking(context: ToolCallHookContext) -> bool | None:
+    """특정 조건에서만 차단합니다."""
+    # 특정 에이전트에 대해서만 차단
+    if context.agent and context.agent.role == "junior_agent":
+        if context.tool_name in ['delete_file', 'send_email']:
+            print(f"❌ 주니어 에이전트는 {context.tool_name}을(를) 사용할 수 없습니다")
+            return False
+    
+    # 특정 작업 중에만 차단
+    if context.task and "민감한" in context.task.description.lower():
+        if context.tool_name == 'web_search':
+            print("❌ 민감한 작업에서는 웹 검색이 차단됩니다")
+            return False
+    
+    return None
+```
+
+### 컨텍스트 인식 입력 수정
+
+```python
+@before_tool_call
+def enhance_tool_inputs(context: ToolCallHookContext) -> None:
+    """에이전트 역할에 따라 컨텍스트를 추가합니다."""
+    # 에이전트 역할에 따라 컨텍스트 추가
+    if context.agent and context.agent.role == "researcher":
+        if context.tool_name == 'web_search':
+            # 연구원에 대한 도메인 제한 추가
+            context.tool_input['domains'] = ['edu', 'gov', 'org']
+    
+    # 작업에 따라 컨텍스트 추가
+    if context.task and "긴급" in context.task.description.lower():
+        if context.tool_name == 'send_email':
+            context.tool_input['priority'] = 'high'
+    
+    return None
+```
+
+## 모범 사례
+
+1. **훅을 집중적으로 유지**: 각 훅은 단일 책임을 가져야 합니다
+2. **무거운 계산 피하기**: 훅은 모든 도구 호출마다 실행됩니다
+3. **오류를 우아하게 처리**: try-except를 사용하여 훅 실패 방지
+4. **타입 힌트 사용**: 더 나은 IDE 지원을 위해 `ToolCallHookContext` 활용
+5. **차단 조건 문서화**: 도구가 차단되는 시기/이유를 명확히 하세요
+6. **훅을 독립적으로 테스트**: 프로덕션에서 사용하기 전에 단위 테스트
+7. **테스트에서 훅 지우기**: 테스트 실행 간 `clear_all_tool_call_hooks()` 사용
+8. **제자리에서 수정**: 항상 `context.tool_input`을 제자리에서 수정하고 교체하지 마세요
+9. **중요한 결정 로깅**: 특히 도구 실행을 차단할 때
+10. **성능 고려**: 가능한 경우 비용이 많이 드는 검증을 캐시
+
+## 오류 처리
+
+```python
+@before_tool_call
+def safe_validation(context: ToolCallHookContext) -> bool | None:
+    try:
+        # 검증 로직
+        if not validate_input(context.tool_input):
+            return False
+    except Exception as e:
+        print(f"⚠️ 훅 오류: {e}")
+        # 결정: 오류 발생 시 허용 또는 차단
+        return None  # 오류에도 불구하고 실행 허용
+```
+
+## 타입 안전성
+
+```python
+from crewai.hooks import ToolCallHookContext, BeforeToolCallHookType, AfterToolCallHookType
+
+# 명시적 타입 주석
+def my_before_hook(context: ToolCallHookContext) -> bool | None:
+    return None
+
+def my_after_hook(context: ToolCallHookContext) -> str | None:
+    return None
+
+# 타입 안전 등록
+register_before_tool_call_hook(my_before_hook)
+register_after_tool_call_hook(my_after_hook)
+```
+
+## 문제 해결
+
+### 훅이 실행되지 않음
+- 크루 실행 전에 훅이 등록되었는지 확인
+- 이전 훅이 `False`를 반환했는지 확인 (실행 및 후속 훅 차단)
+- 훅 시그니처가 예상 타입과 일치하는지 확인
+
+### 입력 수정이 작동하지 않음
+- 제자리 수정 사용: `context.tool_input['key'] = value`
+- 딕셔너리를 교체하지 마세요: `context.tool_input = {}`
+
+### 결과 수정이 작동하지 않음
+- 후 훅에서 수정된 문자열을 반환
+- `None`을 반환하면 원본 결과가 유지됩니다
+- 도구가 실제로 결과를 반환했는지 확인
+
+### 도구가 예기치 않게 차단됨
+- 차단 조건에 대한 모든 전(before) 훅 확인
+- 훅 실행 순서 확인
+- 어떤 훅이 차단하는지 식별하기 위해 디버그 로깅 추가
+
+## 결론
+
+도구 호출 훅은 CrewAI에서 도구 실행을 제어하고 모니터링하는 강력한 기능을 제공합니다. 이를 사용하여 안전 가드레일, 승인 게이트, 입력 검증, 결과 정제, 로깅 및 분석을 구현하세요. 적절한 오류 처리 및 타입 안전성과 결합하면, 훅을 통해 포괄적인 관찰성을 갖춘 안전하고 프로덕션 준비가 된 에이전트 시스템을 구축할 수 있습니다.
+

docs/ko/observability/portkey.mdx

@@ -730,9 +730,7 @@ Portkey 대시보드에서 [구성 페이지](https://app.portkey.ai/configs)에
 - 로그를 필터링하기 위한 관련 메타데이터 수집
 - 액세스 권한 적용
 
-API 키 생성 방법:
-- [Portkey App](https://app.portkey.ai/)
-- [API Key Management API](/ko/api-reference/admin-api/control-plane/api-keys/create-api-key)
+[Portkey App](https://app.portkey.ai/)를 통해 API 키를 생성하세요
 
 Python SDK를 사용한 예시:
 ```python
@@ -755,7 +753,7 @@ api_key = portkey.api_keys.create(
 )
 ```
 
-자세한 키 관리 방법은 [API 키 문서](/ko/api-reference/admin-api/control-plane/api-keys/create-api-key)를 참조하세요.
+자세한 키 관리 방법은 [Portkey 문서](https://portkey.ai/docs)를 참조하세요.
 </Accordion>
 
 <Accordion title="4단계: 배포 및 모니터링">

docs/ko/tools/cloud-storage/overview.mdx

@@ -18,7 +18,7 @@ mode: "wide"
     파일을 Amazon S3 스토리지에 작성하고 업로드합니다.
   </Card>
 
-  <Card title="Bedrock Invoke Agent" icon="aws" href="/ko/tools/cloud-storage/bedrockinvokeagenttool">
+  <Card title="Bedrock Invoke Agent" icon="aws" href="/ko/tools/integration/bedrockinvokeagenttool">
     AI 기반 작업을 위해 Amazon Bedrock 에이전트를 호출합니다.
   </Card>
 

docs/ko/tools/tool-integrations/overview.mdx

@@ -11,7 +11,7 @@ mode: "wide"
   <Card
     title="Bedrock Invoke Agent Tool"
     icon="cloud"
-    href="/en/tools/tool-integrations/bedrockinvokeagenttool"
+    href="/ko/tools/integration/bedrockinvokeagenttool"
     color="#0891B2"
   >
     Invoke Amazon Bedrock Agents from CrewAI to orchestrate actions across AWS services.
@@ -20,7 +20,7 @@ mode: "wide"
   <Card
     title="CrewAI Automation Tool"
     icon="bolt"
-    href="/en/tools/tool-integrations/crewaiautomationtool"
+    href="/ko/tools/integration/crewaiautomationtool"
     color="#7C3AED"
   >
     Automate deployment and operations by integrating CrewAI with external platforms and workflows.

docs/pt-BR/concepts/knowledge.mdx

@@ -704,7 +704,7 @@ class KnowledgeMonitorListener(BaseEventListener):
 knowledge_monitor = KnowledgeMonitorListener()
 ```
 
-Para mais informações sobre como usar eventos, consulte a documentação [Event Listeners](https://docs.crewai.com/concepts/event-listener).
+Para mais informações sobre como usar eventos, consulte a documentação [Event Listeners](/pt-BR/concepts/event-listener).
 
 ### Fontes de Knowledge Personalizadas
 

docs/pt-BR/concepts/llms.mdx

@@ -725,7 +725,7 @@ O CrewAI suporta respostas em streaming de LLMs, permitindo que sua aplicação
     ```
 
     <Tip>
-      [Clique aqui](https://docs.crewai.com/concepts/event-listener#event-listeners) para mais detalhes 
+      [Clique aqui](/pt-BR/concepts/event-listener#event-listeners) para mais detalhes
     </Tip>
   </Tab>
 </Tabs>

docs/pt-BR/enterprise/features/marketplace.mdx

@@ -36,7 +36,7 @@ Você também pode baixar templates diretamente do marketplace clicando em `Down
   <Card title="Ferramentas & Integrações" href="/pt-BR/enterprise/features/tools-and-integrations" icon="wrench">
     Conecte apps externos e gerencie ferramentas internas que seus agentes podem usar.
   </Card>
-  <Card title="Repositório de Ferramentas" href="/pt-BR/enterprise/features/tool-repository" icon="toolbox">
+  <Card title="Repositório de Ferramentas" href="/pt-BR/enterprise/guides/tool-repository" icon="toolbox">
     Publique e instale ferramentas para ampliar as capacidades dos seus crews.
   </Card>
   <Card title="Repositório de Agentes" href="/pt-BR/enterprise/features/agent-repositories" icon="people-group">

docs/pt-BR/enterprise/features/tools-and-integrations.mdx

@@ -231,7 +231,7 @@ Ferramentas & Integrações é o hub central para conectar aplicações de terce
 ## Relacionados
 
 <CardGroup cols={2}>
-  <Card title="Repositório de Ferramentas" href="/pt-BR/enterprise/features/tool-repository" icon="toolbox">
+  <Card title="Repositório de Ferramentas" href="/pt-BR/enterprise/guides/tool-repository" icon="toolbox">
     Publique e instale ferramentas para ampliar as capacidades dos seus crews.
   </Card>
   <Card title="Automação com Webhook" href="/pt-BR/enterprise/guides/webhook-automation" icon="bolt">

docs/pt-BR/enterprise/guides/tool-repository.mdx

@@ -21,7 +21,7 @@ O repositório não é um sistema de controle de versões. Use Git para rastrear
 Antes de usar o Repositório de Ferramentas, certifique-se de que você possui:
 
 - Uma conta [CrewAI AMP](https://app.crewai.com)
-- [CrewAI CLI](https://docs.crewai.com/concepts/cli#cli) instalada
+- [CrewAI CLI](/pt-BR/concepts/cli#cli) instalada
 - uv>=0.5.0 instalado. Veja [como atualizar](https://docs.astral.sh/uv/getting-started/installation/#upgrading-uv)
 - [Git](https://git-scm.com) instalado e configurado
 - Permissões de acesso para publicar ou instalar ferramentas em sua organização CrewAI AMP
@@ -66,7 +66,7 @@ Por padrão, as ferramentas são publicadas como privadas. Para tornar uma ferra
 crewai tool publish --public
 ```
 
-Para mais detalhes sobre como construir ferramentas, acesse [Criando suas próprias ferramentas](https://docs.crewai.com/concepts/tools#creating-your-own-tools).
+Para mais detalhes sobre como construir ferramentas, acesse [Criando suas próprias ferramentas](/pt-BR/concepts/tools#creating-your-own-tools).
 
 ## Atualizando ferramentas
 

docs/pt-BR/enterprise/resources/frequently-asked-questions.mdx

@@ -49,7 +49,7 @@ mode: "wide"
 
         Para integrar a entrada humana na execução do agente, defina a flag `human_input` na definição da tarefa. Quando habilitada, o agente solicitará a entrada do usuário antes de entregar sua resposta final. Essa entrada pode fornecer contexto extra, esclarecer ambiguidades ou validar a saída do agente.
 
-        Para orientações detalhadas de implementação, veja nosso [guia Human-in-the-Loop](/pt-BR/how-to/human-in-the-loop).
+        Para orientações detalhadas de implementação, veja nosso [guia Human-in-the-Loop](/pt-BR/enterprise/guides/human-in-the-loop).
     </Accordion>
 
     <Accordion title="Quais opções avançadas de customização estão disponíveis para aprimorar e personalizar o comportamento e as capacidades dos agentes na CrewAI?">
@@ -142,7 +142,7 @@ mode: "wide"
     <Accordion title="Como posso criar ferramentas personalizadas para meus agentes CrewAI?">
         Você pode criar ferramentas personalizadas herdando da classe `BaseTool` fornecida pela CrewAI ou usando o decorador de ferramenta. Herdar envolve definir uma nova classe que herda de `BaseTool`, especificando o nome, a descrição e o método `_run` para a lógica operacional. O decorador de ferramenta permite criar um objeto `Tool` diretamente com os atributos necessários e uma lógica funcional.
 
-        <Card href="https://docs.crewai.com/how-to/create-custom-tools" icon="code">CrewAI Tools Guide</Card>
+        <Card href="/pt-BR/learn/create-custom-tools" icon="code">CrewAI Tools Guide</Card>
     </Accordion>
 
     <Accordion title="Como controlar o número máximo de solicitações por minuto que toda a crew pode realizar?">

docs/pt-BR/learn/execution-hooks.mdx

@@ -0,0 +1,379 @@
+---
+title: Visão Geral dos Hooks de Execução
+description: Entendendo e usando hooks de execução no CrewAI para controle fino sobre operações de agentes
+mode: "wide"
+---
+
+Os Hooks de Execução fornecem controle fino sobre o comportamento em tempo de execução dos seus agentes CrewAI. Diferentemente dos hooks de kickoff que são executados antes e depois da execução da crew, os hooks de execução interceptam operações específicas durante a execução do agente, permitindo que você modifique comportamentos, implemente verificações de segurança e adicione monitoramento abrangente.
+
+## Tipos de Hooks de Execução
+
+O CrewAI fornece duas categorias principais de hooks de execução:
+
+### 1. [Hooks de Chamada LLM](/learn/llm-hooks)
+
+Controle e monitore interações com o modelo de linguagem:
+- **Antes da Chamada LLM**: Modifique prompts, valide entradas, implemente gates de aprovação
+- **Depois da Chamada LLM**: Transforme respostas, sanitize saídas, atualize histórico de conversação
+
+**Casos de Uso:**
+- Limitação de iterações
+- Rastreamento de custos e monitoramento de uso de tokens
+- Sanitização de respostas e filtragem de conteúdo
+- Aprovação humana para chamadas LLM
+- Adição de diretrizes de segurança ou contexto
+- Logging de debug e inspeção de requisição/resposta
+
+[Ver Documentação de Hooks LLM →](/learn/llm-hooks)
+
+### 2. [Hooks de Chamada de Ferramenta](/learn/tool-hooks)
+
+Controle e monitore execução de ferramentas:
+- **Antes da Chamada de Ferramenta**: Modifique entradas, valide parâmetros, bloqueie operações perigosas
+- **Depois da Chamada de Ferramenta**: Transforme resultados, sanitize saídas, registre detalhes de execução
+
+**Casos de Uso:**
+- Guardrails de segurança para operações destrutivas
+- Aprovação humana para ações sensíveis
+- Validação e sanitização de entrada
+- Cache de resultados e limitação de taxa
+- Análise de uso de ferramentas
+- Logging de debug e monitoramento
+
+[Ver Documentação de Hooks de Ferramenta →](/learn/tool-hooks)
+
+## Métodos de Registro
+
+### 1. Hooks Baseados em Decoradores (Recomendado)
+
+A maneira mais limpa e pythônica de registrar hooks:
+
+```python
+from crewai.hooks import before_llm_call, after_llm_call, before_tool_call, after_tool_call
+
+@before_llm_call
+def limit_iterations(context):
+    """Previne loops infinitos limitando iterações."""
+    if context.iterations > 10:
+        return False  # Bloquear execução
+    return None
+
+@after_llm_call
+def sanitize_response(context):
+    """Remove dados sensíveis das respostas do LLM."""
+    if "API_KEY" in context.response:
+        return context.response.replace("API_KEY", "[CENSURADO]")
+    return None
+
+@before_tool_call
+def block_dangerous_tools(context):
+    """Bloqueia operações destrutivas."""
+    if context.tool_name == "delete_database":
+        return False  # Bloquear execução
+    return None
+
+@after_tool_call
+def log_tool_result(context):
+    """Registra execução de ferramenta."""
+    print(f"Ferramenta {context.tool_name} concluída")
+    return None
+```
+
+### 2. Hooks com Escopo de Crew
+
+Aplica hooks apenas a instâncias específicas de crew:
+
+```python
+from crewai import CrewBase
+from crewai.project import crew
+from crewai.hooks import before_llm_call_crew, after_tool_call_crew
+
+@CrewBase
+class MyProjCrew:
+    @before_llm_call_crew
+    def validate_inputs(self, context):
+        # Aplica-se apenas a esta crew
+        print(f"Chamada LLM em {self.__class__.__name__}")
+        return None
+
+    @after_tool_call_crew
+    def log_results(self, context):
+        # Logging específico da crew
+        print(f"Resultado da ferramenta: {context.tool_result[:50]}...")
+        return None
+
+    @crew
+    def crew(self) -> Crew:
+        return Crew(
+            agents=self.agents,
+            tasks=self.tasks,
+            process=Process.sequential
+        )
+```
+
+## Fluxo de Execução de Hooks
+
+### Fluxo de Chamada LLM
+
+```
+Agente precisa chamar LLM
+    ↓
+[Hooks Antes da Chamada LLM Executam]
+    ├→ Hook 1: Validar contagem de iterações
+    ├→ Hook 2: Adicionar contexto de segurança
+    └→ Hook 3: Registrar requisição
+    ↓
+Se algum hook retornar False:
+    ├→ Bloquear chamada LLM
+    └→ Lançar ValueError
+    ↓
+Se todos os hooks retornarem True/None:
+    ├→ Chamada LLM prossegue
+    └→ Resposta gerada
+    ↓
+[Hooks Depois da Chamada LLM Executam]
+    ├→ Hook 1: Sanitizar resposta
+    ├→ Hook 2: Registrar resposta
+    └→ Hook 3: Atualizar métricas
+    ↓
+Resposta final retornada
+```
+
+### Fluxo de Chamada de Ferramenta
+
+```
+Agente precisa executar ferramenta
+    ↓
+[Hooks Antes da Chamada de Ferramenta Executam]
+    ├→ Hook 1: Verificar se ferramenta é permitida
+    ├→ Hook 2: Validar entradas
+    └→ Hook 3: Solicitar aprovação se necessário
+    ↓
+Se algum hook retornar False:
+    ├→ Bloquear execução da ferramenta
+    └→ Retornar mensagem de erro
+    ↓
+Se todos os hooks retornarem True/None:
+    ├→ Execução da ferramenta prossegue
+    └→ Resultado gerado
+    ↓
+[Hooks Depois da Chamada de Ferramenta Executam]
+    ├→ Hook 1: Sanitizar resultado
+    ├→ Hook 2: Fazer cache do resultado
+    └→ Hook 3: Registrar métricas
+    ↓
+Resultado final retornado
+```
+
+## Objetos de Contexto de Hook
+
+### LLMCallHookContext
+
+Fornece acesso ao estado de execução do LLM:
+
+```python
+class LLMCallHookContext:
+    executor: CrewAgentExecutor  # Acesso completo ao executor
+    messages: list               # Lista de mensagens mutável
+    agent: Agent                 # Agente atual
+    task: Task                   # Tarefa atual
+    crew: Crew                   # Instância da crew
+    llm: BaseLLM                 # Instância do LLM
+    iterations: int              # Iteração atual
+    response: str | None         # Resposta do LLM (hooks posteriores)
+```
+
+### ToolCallHookContext
+
+Fornece acesso ao estado de execução da ferramenta:
+
+```python
+class ToolCallHookContext:
+    tool_name: str               # Ferramenta sendo chamada
+    tool_input: dict             # Parâmetros de entrada mutáveis
+    tool: CrewStructuredTool     # Instância da ferramenta
+    agent: Agent | None          # Agente executando
+    task: Task | None            # Tarefa atual
+    crew: Crew | None            # Instância da crew
+    tool_result: str | None      # Resultado da ferramenta (hooks posteriores)
+```
+
+## Padrões Comuns
+
+### Segurança e Validação
+
+```python
+@before_tool_call
+def safety_check(context):
+    """Bloqueia operações destrutivas."""
+    dangerous = ['delete_file', 'drop_table', 'system_shutdown']
+    if context.tool_name in dangerous:
+        print(f"🛑 Bloqueado: {context.tool_name}")
+        return False
+    return None
+
+@before_llm_call
+def iteration_limit(context):
+    """Previne loops infinitos."""
+    if context.iterations > 15:
+        print("⛔ Máximo de iterações excedido")
+        return False
+    return None
+```
+
+### Humano no Loop
+
+```python
+@before_tool_call
+def require_approval(context):
+    """Requer aprovação para operações sensíveis."""
+    sensitive = ['send_email', 'make_payment', 'post_message']
+
+    if context.tool_name in sensitive:
+        response = context.request_human_input(
+            prompt=f"Aprovar {context.tool_name}?",
+            default_message="Digite 'sim' para aprovar:"
+        )
+
+        if response.lower() != 'sim':
+            return False
+
+    return None
+```
+
+### Monitoramento e Análise
+
+```python
+from collections import defaultdict
+import time
+
+metrics = defaultdict(lambda: {'count': 0, 'total_time': 0})
+
+@before_tool_call
+def start_timer(context):
+    context.tool_input['_start'] = time.time()
+    return None
+
+@after_tool_call
+def track_metrics(context):
+    start = context.tool_input.get('_start', time.time())
+    duration = time.time() - start
+
+    metrics[context.tool_name]['count'] += 1
+    metrics[context.tool_name]['total_time'] += duration
+
+    return None
+```
+
+## Gerenciamento de Hooks
+
+### Limpar Todos os Hooks
+
+```python
+from crewai.hooks import clear_all_global_hooks
+
+# Limpa todos os hooks de uma vez
+result = clear_all_global_hooks()
+print(f"Limpou {result['total']} hooks")
+```
+
+### Limpar Tipos Específicos de Hooks
+
+```python
+from crewai.hooks import (
+    clear_before_llm_call_hooks,
+    clear_after_llm_call_hooks,
+    clear_before_tool_call_hooks,
+    clear_after_tool_call_hooks
+)
+
+# Limpar tipos específicos
+llm_before_count = clear_before_llm_call_hooks()
+tool_after_count = clear_after_tool_call_hooks()
+```
+
+## Melhores Práticas
+
+### 1. Mantenha os Hooks Focados
+Cada hook deve ter uma responsabilidade única e clara.
+
+### 2. Trate Erros Graciosamente
+```python
+@before_llm_call
+def safe_hook(context):
+    try:
+        if some_condition:
+            return False
+    except Exception as e:
+        print(f"Erro no hook: {e}")
+        return None  # Permitir execução apesar do erro
+```
+
+### 3. Modifique o Contexto In-Place
+```python
+# ✅ Correto - modificar in-place
+@before_llm_call
+def add_context(context):
+    context.messages.append({"role": "system", "content": "Seja conciso"})
+
+# ❌ Errado - substitui referência
+@before_llm_call
+def wrong_approach(context):
+    context.messages = [{"role": "system", "content": "Seja conciso"}]
+```
+
+### 4. Use Type Hints
+```python
+from crewai.hooks import LLMCallHookContext, ToolCallHookContext
+
+def my_llm_hook(context: LLMCallHookContext) -> bool | None:
+    return None
+
+def my_tool_hook(context: ToolCallHookContext) -> str | None:
+    return None
+```
+
+### 5. Limpe em Testes
+```python
+import pytest
+from crewai.hooks import clear_all_global_hooks
+
+@pytest.fixture(autouse=True)
+def clean_hooks():
+    """Reseta hooks antes de cada teste."""
+    yield
+    clear_all_global_hooks()
+```
+
+## Quando Usar Qual Hook
+
+### Use Hooks LLM Quando:
+- Implementar limites de iteração
+- Adicionar contexto ou diretrizes de segurança aos prompts
+- Rastrear uso de tokens e custos
+- Sanitizar ou transformar respostas
+- Implementar gates de aprovação para chamadas LLM
+- Fazer debug de interações de prompt/resposta
+
+### Use Hooks de Ferramenta Quando:
+- Bloquear operações perigosas ou destrutivas
+- Validar entradas de ferramenta antes da execução
+- Implementar gates de aprovação para ações sensíveis
+- Fazer cache de resultados de ferramenta
+- Rastrear uso e performance de ferramentas
+- Sanitizar saídas de ferramenta
+- Limitar taxa de chamadas de ferramenta
+
+### Use Ambos Quando:
+Construir sistemas abrangentes de observabilidade, segurança ou aprovação que precisam monitorar todas as operações do agente.
+
+## Documentação Relacionada
+
+- [Hooks de Chamada LLM →](/learn/llm-hooks) - Documentação detalhada de hooks LLM
+- [Hooks de Chamada de Ferramenta →](/learn/tool-hooks) - Documentação detalhada de hooks de ferramenta
+- [Hooks Antes e Depois do Kickoff →](/learn/before-and-after-kickoff-hooks) - Hooks do ciclo de vida da crew
+- [Humano no Loop →](/learn/human-in-the-loop) - Padrões de entrada humana
+
+## Conclusão
+
+Os Hooks de Execução fornecem controle poderoso sobre o comportamento em tempo de execução do agente. Use-os para implementar guardrails de segurança, fluxos de trabalho de aprovação, monitoramento abrangente e lógica de negócio personalizada. Combinados com tratamento adequado de erros, segurança de tipos e considerações de performance, os hooks permitem sistemas de agentes seguros, prontos para produção e observáveis.

docs/pt-BR/learn/hierarchical-process.mdx

@@ -96,7 +96,7 @@ project_crew = Crew(
 ```
 
 <Tip>
-    Para mais detalhes sobre a criação e personalização de um agente gerente, confira a [documentação do Custom Manager Agent](https://docs.crewai.com/how-to/custom-manager-agent#custom-manager-agent).
+    Para mais detalhes sobre a criação e personalização de um agente gerente, confira a [documentação do Custom Manager Agent](/pt-BR/learn/custom-manager-agent).
 </Tip>
 
 

docs/pt-BR/learn/llm-hooks.mdx

@@ -0,0 +1,388 @@
+---
+title: Hooks de Chamada LLM
+description: Aprenda a usar hooks de chamada LLM para interceptar, modificar e controlar interações com modelos de linguagem no CrewAI
+mode: "wide"
+---
+
+Os Hooks de Chamada LLM fornecem controle fino sobre interações com modelos de linguagem durante a execução do agente. Esses hooks permitem interceptar chamadas LLM, modificar prompts, transformar respostas, implementar gates de aprovação e adicionar logging ou monitoramento personalizado.
+
+## Visão Geral
+
+Os hooks LLM são executados em dois pontos críticos:
+- **Antes da Chamada LLM**: Modificar mensagens, validar entradas ou bloquear execução
+- **Depois da Chamada LLM**: Transformar respostas, sanitizar saídas ou modificar histórico de conversação
+
+## Tipos de Hook
+
+### Hooks Antes da Chamada LLM
+
+Executados antes de cada chamada LLM, esses hooks podem:
+- Inspecionar e modificar mensagens enviadas ao LLM
+- Bloquear execução LLM com base em condições
+- Implementar limitação de taxa ou gates de aprovação
+- Adicionar contexto ou mensagens do sistema
+- Registrar detalhes da requisição
+
+**Assinatura:**
+```python
+def before_hook(context: LLMCallHookContext) -> bool | None:
+    # Retorne False para bloquear execução
+    # Retorne True ou None para permitir execução
+    ...
+```
+
+### Hooks Depois da Chamada LLM
+
+Executados depois de cada chamada LLM, esses hooks podem:
+- Modificar ou sanitizar respostas do LLM
+- Adicionar metadados ou formatação
+- Registrar detalhes da resposta
+- Atualizar histórico de conversação
+- Implementar filtragem de conteúdo
+
+**Assinatura:**
+```python
+def after_hook(context: LLMCallHookContext) -> str | None:
+    # Retorne string de resposta modificada
+    # Retorne None para manter resposta original
+    ...
+```
+
+## Contexto do Hook LLM
+
+O objeto `LLMCallHookContext` fornece acesso abrangente ao estado de execução:
+
+```python
+class LLMCallHookContext:
+    executor: CrewAgentExecutor  # Referência completa ao executor
+    messages: list               # Lista de mensagens mutável
+    agent: Agent                 # Agente atual
+    task: Task                   # Tarefa atual
+    crew: Crew                   # Instância da crew
+    llm: BaseLLM                 # Instância do LLM
+    iterations: int              # Contagem de iteração atual
+    response: str | None         # Resposta do LLM (apenas hooks posteriores)
+```
+
+### Modificando Mensagens
+
+**Importante:** Sempre modifique mensagens in-place:
+
+```python
+# ✅ Correto - modificar in-place
+def add_context(context: LLMCallHookContext) -> None:
+    context.messages.append({"role": "system", "content": "Seja conciso"})
+
+# ❌ Errado - substitui referência da lista
+def wrong_approach(context: LLMCallHookContext) -> None:
+    context.messages = [{"role": "system", "content": "Seja conciso"}]
+```
+
+## Métodos de Registro
+
+### 1. Registro Baseado em Decoradores (Recomendado)
+
+Use decoradores para sintaxe mais limpa:
+
+```python
+from crewai.hooks import before_llm_call, after_llm_call
+
+@before_llm_call
+def validate_iteration_count(context):
+    """Valida a contagem de iterações."""
+    if context.iterations > 10:
+        print("⚠️ Máximo de iterações excedido")
+        return False  # Bloquear execução
+    return None
+
+@after_llm_call
+def sanitize_response(context):
+    """Remove dados sensíveis."""
+    if context.response and "API_KEY" in context.response:
+        return context.response.replace("API_KEY", "[CENSURADO]")
+    return None
+```
+
+### 2. Hooks com Escopo de Crew
+
+Registre hooks para uma instância específica de crew:
+
+```python
+from crewai import CrewBase
+from crewai.project import crew
+from crewai.hooks import before_llm_call_crew, after_llm_call_crew
+
+@CrewBase
+class MyProjCrew:
+    @before_llm_call_crew
+    def validate_inputs(self, context):
+        # Aplica-se apenas a esta crew
+        if context.iterations == 0:
+            print(f"Iniciando tarefa: {context.task.description}")
+        return None
+    
+    @after_llm_call_crew
+    def log_responses(self, context):
+        # Logging específico da crew
+        print(f"Comprimento da resposta: {len(context.response)}")
+        return None
+    
+    @crew
+    def crew(self) -> Crew:
+        return Crew(
+            agents=self.agents,
+            tasks=self.tasks,
+            process=Process.sequential,
+            verbose=True
+        )
+```
+
+## Casos de Uso Comuns
+
+### 1. Limitação de Iterações
+
+```python
+@before_llm_call
+def limit_iterations(context: LLMCallHookContext) -> bool | None:
+    """Previne loops infinitos limitando iterações."""
+    max_iterations = 15
+    if context.iterations > max_iterations:
+        print(f"⛔ Bloqueado: Excedeu {max_iterations} iterações")
+        return False  # Bloquear execução
+    return None
+```
+
+### 2. Gate de Aprovação Humana
+
+```python
+@before_llm_call
+def require_approval(context: LLMCallHookContext) -> bool | None:
+    """Requer aprovação após certas iterações."""
+    if context.iterations > 5:
+        response = context.request_human_input(
+            prompt=f"Iteração {context.iterations}: Aprovar chamada LLM?",
+            default_message="Pressione Enter para aprovar, ou digite 'não' para bloquear:"
+        )
+        if response.lower() == "não":
+            print("🚫 Chamada LLM bloqueada pelo usuário")
+            return False
+    return None
+```
+
+### 3. Adicionando Contexto do Sistema
+
+```python
+@before_llm_call
+def add_guardrails(context: LLMCallHookContext) -> None:
+    """Adiciona diretrizes de segurança a cada chamada LLM."""
+    context.messages.append({
+        "role": "system",
+        "content": "Garanta que as respostas sejam factuais e cite fontes quando possível."
+    })
+    return None
+```
+
+### 4. Sanitização de Resposta
+
+```python
+@after_llm_call
+def sanitize_sensitive_data(context: LLMCallHookContext) -> str | None:
+    """Remove padrões sensíveis."""
+    if not context.response:
+        return None
+    
+    import re
+    sanitized = context.response
+    sanitized = re.sub(r'\b\d{3}\.\d{3}\.\d{3}-\d{2}\b', '[CPF-CENSURADO]', sanitized)
+    sanitized = re.sub(r'\b\d{4}[- ]?\d{4}[- ]?\d{4}[- ]?\d{4}\b', '[CARTÃO-CENSURADO]', sanitized)
+    
+    return sanitized
+```
+
+### 5. Rastreamento de Custos
+
+```python
+import tiktoken
+
+@before_llm_call
+def track_token_usage(context: LLMCallHookContext) -> None:
+    """Rastreia tokens de entrada."""
+    encoding = tiktoken.get_encoding("cl100k_base")
+    total_tokens = sum(
+        len(encoding.encode(msg.get("content", ""))) 
+        for msg in context.messages
+    )
+    print(f"📊 Tokens de entrada: ~{total_tokens}")
+    return None
+
+@after_llm_call
+def track_response_tokens(context: LLMCallHookContext) -> None:
+    """Rastreia tokens de resposta."""
+    if context.response:
+        encoding = tiktoken.get_encoding("cl100k_base")
+        tokens = len(encoding.encode(context.response))
+        print(f"📊 Tokens de resposta: ~{tokens}")
+    return None
+```
+
+### 6. Logging de Debug
+
+```python
+@before_llm_call
+def debug_request(context: LLMCallHookContext) -> None:
+    """Debug de requisição LLM."""
+    print(f"""
+    🔍 Debug de Chamada LLM:
+    - Agente: {context.agent.role}
+    - Tarefa: {context.task.description[:50]}...
+    - Iteração: {context.iterations}
+    - Contagem de Mensagens: {len(context.messages)}
+    - Última Mensagem: {context.messages[-1] if context.messages else 'Nenhuma'}
+    """)
+    return None
+
+@after_llm_call
+def debug_response(context: LLMCallHookContext) -> None:
+    """Debug de resposta LLM."""
+    if context.response:
+        print(f"✅ Preview da Resposta: {context.response[:100]}...")
+    return None
+```
+
+## Gerenciamento de Hooks
+
+### Desregistrando Hooks
+
+```python
+from crewai.hooks import (
+    unregister_before_llm_call_hook,
+    unregister_after_llm_call_hook
+)
+
+# Desregistrar hook específico
+def my_hook(context):
+    ...
+
+register_before_llm_call_hook(my_hook)
+# Mais tarde...
+unregister_before_llm_call_hook(my_hook)  # Retorna True se encontrado
+```
+
+### Limpando Hooks
+
+```python
+from crewai.hooks import (
+    clear_before_llm_call_hooks,
+    clear_after_llm_call_hooks,
+    clear_all_llm_call_hooks
+)
+
+# Limpar tipo específico de hook
+count = clear_before_llm_call_hooks()
+print(f"Limpou {count} hooks antes")
+
+# Limpar todos os hooks LLM
+before_count, after_count = clear_all_llm_call_hooks()
+print(f"Limpou {before_count} hooks antes e {after_count} hooks depois")
+```
+
+## Padrões Avançados
+
+### Execução Condicional de Hook
+
+```python
+@before_llm_call
+def conditional_blocking(context: LLMCallHookContext) -> bool | None:
+    """Bloqueia apenas em condições específicas."""
+    # Bloquear apenas para agentes específicos
+    if context.agent.role == "researcher" and context.iterations > 10:
+        return False
+    
+    # Bloquear apenas para tarefas específicas
+    if "sensível" in context.task.description.lower() and context.iterations > 5:
+        return False
+    
+    return None
+```
+
+### Modificações com Consciência de Contexto
+
+```python
+@before_llm_call
+def adaptive_prompting(context: LLMCallHookContext) -> None:
+    """Adiciona contexto diferente baseado na iteração."""
+    if context.iterations == 0:
+        context.messages.append({
+            "role": "system",
+            "content": "Comece com uma visão geral de alto nível."
+        })
+    elif context.iterations > 3:
+        context.messages.append({
+            "role": "system",
+            "content": "Foque em detalhes específicos e forneça exemplos."
+        })
+    return None
+```
+
+## Melhores Práticas
+
+1. **Mantenha Hooks Focados**: Cada hook deve ter uma responsabilidade única
+2. **Evite Computação Pesada**: Hooks executam em cada chamada LLM
+3. **Trate Erros Graciosamente**: Use try-except para prevenir falhas de hooks
+4. **Use Type Hints**: Aproveite `LLMCallHookContext` para melhor suporte IDE
+5. **Documente Comportamento do Hook**: Especialmente para condições de bloqueio
+6. **Teste Hooks Independentemente**: Teste unitário de hooks antes de usar em produção
+7. **Limpe Hooks em Testes**: Use `clear_all_llm_call_hooks()` entre execuções de teste
+8. **Modifique In-Place**: Sempre modifique `context.messages` in-place, nunca substitua
+
+## Tratamento de Erros
+
+```python
+@before_llm_call
+def safe_hook(context: LLMCallHookContext) -> bool | None:
+    try:
+        # Sua lógica de hook
+        if some_condition:
+            return False
+    except Exception as e:
+        print(f"⚠️ Erro no hook: {e}")
+        # Decida: permitir ou bloquear em erro
+        return None  # Permitir execução apesar do erro
+```
+
+## Segurança de Tipos
+
+```python
+from crewai.hooks import LLMCallHookContext, BeforeLLMCallHookType, AfterLLMCallHookType
+
+# Anotações de tipo explícitas
+def my_before_hook(context: LLMCallHookContext) -> bool | None:
+    return None
+
+def my_after_hook(context: LLMCallHookContext) -> str | None:
+    return None
+
+# Registro type-safe
+register_before_llm_call_hook(my_before_hook)
+register_after_llm_call_hook(my_after_hook)
+```
+
+## Solução de Problemas
+
+### Hook Não Está Executando
+- Verifique se o hook está registrado antes da execução da crew
+- Verifique se hook anterior retornou `False` (bloqueia hooks subsequentes)
+- Garanta que assinatura do hook corresponda ao tipo esperado
+
+### Modificações de Mensagem Não Persistem
+- Use modificações in-place: `context.messages.append()`
+- Não substitua a lista: `context.messages = []`
+
+### Modificações de Resposta Não Funcionam
+- Retorne a string modificada dos hooks posteriores
+- Retornar `None` mantém a resposta original
+
+## Conclusão
+
+Os Hooks de Chamada LLM fornecem capacidades poderosas para controlar e monitorar interações com modelos de linguagem no CrewAI. Use-os para implementar guardrails de segurança, gates de aprovação, logging, rastreamento de custos e sanitização de respostas. Combinados com tratamento adequado de erros e segurança de tipos, os hooks permitem sistemas de agentes robustos e prontos para produção.
+

docs/pt-BR/learn/tool-hooks.mdx

@@ -0,0 +1,498 @@
+---
+title: Hooks de Chamada de Ferramenta
+description: Aprenda a usar hooks de chamada de ferramenta para interceptar, modificar e controlar execução de ferramentas no CrewAI
+mode: "wide"
+---
+
+Os Hooks de Chamada de Ferramenta fornecem controle fino sobre a execução de ferramentas durante operações do agente. Esses hooks permitem interceptar chamadas de ferramenta, modificar entradas, transformar saídas, implementar verificações de segurança e adicionar logging ou monitoramento abrangente.
+
+## Visão Geral
+
+Os hooks de ferramenta são executados em dois pontos críticos:
+- **Antes da Chamada de Ferramenta**: Modificar entradas, validar parâmetros ou bloquear execução
+- **Depois da Chamada de Ferramenta**: Transformar resultados, sanitizar saídas ou registrar detalhes de execução
+
+## Tipos de Hook
+
+### Hooks Antes da Chamada de Ferramenta
+
+Executados antes de cada execução de ferramenta, esses hooks podem:
+- Inspecionar e modificar entradas de ferramenta
+- Bloquear execução de ferramenta com base em condições
+- Implementar gates de aprovação para operações perigosas
+- Validar parâmetros
+- Registrar invocações de ferramenta
+
+**Assinatura:**
+```python
+def before_hook(context: ToolCallHookContext) -> bool | None:
+    # Retorne False para bloquear execução
+    # Retorne True ou None para permitir execução
+    ...
+```
+
+### Hooks Depois da Chamada de Ferramenta
+
+Executados depois de cada execução de ferramenta, esses hooks podem:
+- Modificar ou sanitizar resultados de ferramenta
+- Adicionar metadados ou formatação
+- Registrar resultados de execução
+- Implementar validação de resultado
+- Transformar formatos de saída
+
+**Assinatura:**
+```python
+def after_hook(context: ToolCallHookContext) -> str | None:
+    # Retorne string de resultado modificado
+    # Retorne None para manter resultado original
+    ...
+```
+
+## Contexto do Hook de Ferramenta
+
+O objeto `ToolCallHookContext` fornece acesso abrangente ao estado de execução da ferramenta:
+
+```python
+class ToolCallHookContext:
+    tool_name: str                    # Nome da ferramenta sendo chamada
+    tool_input: dict[str, Any]        # Parâmetros de entrada mutáveis da ferramenta
+    tool: CrewStructuredTool          # Referência da instância da ferramenta
+    agent: Agent | BaseAgent | None   # Agente executando a ferramenta
+    task: Task | None                 # Tarefa atual
+    crew: Crew | None                 # Instância da crew
+    tool_result: str | None           # Resultado da ferramenta (apenas hooks posteriores)
+```
+
+### Modificando Entradas de Ferramenta
+
+**Importante:** Sempre modifique entradas de ferramenta in-place:
+
+```python
+# ✅ Correto - modificar in-place
+def sanitize_input(context: ToolCallHookContext) -> None:
+    context.tool_input['query'] = context.tool_input['query'].lower()
+
+# ❌ Errado - substitui referência do dict
+def wrong_approach(context: ToolCallHookContext) -> None:
+    context.tool_input = {'query': 'nova consulta'}
+```
+
+## Métodos de Registro
+
+### 1. Registro Baseado em Decoradores (Recomendado)
+
+Use decoradores para sintaxe mais limpa:
+
+```python
+from crewai.hooks import before_tool_call, after_tool_call
+
+@before_tool_call
+def block_dangerous_tools(context):
+    """Bloqueia ferramentas perigosas."""
+    dangerous_tools = ['delete_database', 'drop_table', 'rm_rf']
+    if context.tool_name in dangerous_tools:
+        print(f"⛔ Ferramenta perigosa bloqueada: {context.tool_name}")
+        return False  # Bloquear execução
+    return None
+
+@after_tool_call
+def sanitize_results(context):
+    """Sanitiza resultados."""
+    if context.tool_result and "password" in context.tool_result.lower():
+        return context.tool_result.replace("password", "[CENSURADO]")
+    return None
+```
+
+### 2. Hooks com Escopo de Crew
+
+Registre hooks para uma instância específica de crew:
+
+```python
+from crewai import CrewBase
+from crewai.project import crew
+from crewai.hooks import before_tool_call_crew, after_tool_call_crew
+
+@CrewBase
+class MyProjCrew:
+    @before_tool_call_crew
+    def validate_tool_inputs(self, context):
+        # Aplica-se apenas a esta crew
+        if context.tool_name == "web_search":
+            if not context.tool_input.get('query'):
+                print("❌ Consulta de busca inválida")
+                return False
+        return None
+    
+    @after_tool_call_crew
+    def log_tool_results(self, context):
+        # Logging de ferramenta específico da crew
+        print(f"✅ {context.tool_name} concluída")
+        return None
+    
+    @crew
+    def crew(self) -> Crew:
+        return Crew(
+            agents=self.agents,
+            tasks=self.tasks,
+            process=Process.sequential,
+            verbose=True
+        )
+```
+
+## Casos de Uso Comuns
+
+### 1. Guardrails de Segurança
+
+```python
+@before_tool_call
+def safety_check(context: ToolCallHookContext) -> bool | None:
+    """Bloqueia ferramentas que podem causar danos."""
+    destructive_tools = [
+        'delete_file',
+        'drop_table',
+        'remove_user',
+        'system_shutdown'
+    ]
+    
+    if context.tool_name in destructive_tools:
+        print(f"🛑 Ferramenta destrutiva bloqueada: {context.tool_name}")
+        return False
+    
+    # Avisar em operações sensíveis
+    sensitive_tools = ['send_email', 'post_to_social_media', 'charge_payment']
+    if context.tool_name in sensitive_tools:
+        print(f"⚠️  Executando ferramenta sensível: {context.tool_name}")
+    
+    return None
+```
+
+### 2. Gate de Aprovação Humana
+
+```python
+@before_tool_call
+def require_approval_for_actions(context: ToolCallHookContext) -> bool | None:
+    """Requer aprovação para ações específicas."""
+    approval_required = [
+        'send_email',
+        'make_purchase',
+        'delete_file',
+        'post_message'
+    ]
+    
+    if context.tool_name in approval_required:
+        response = context.request_human_input(
+            prompt=f"Aprovar {context.tool_name}?",
+            default_message=f"Entrada: {context.tool_input}\nDigite 'sim' para aprovar:"
+        )
+        
+        if response.lower() != 'sim':
+            print(f"❌ Execução de ferramenta negada: {context.tool_name}")
+            return False
+    
+    return None
+```
+
+### 3. Validação e Sanitização de Entrada
+
+```python
+@before_tool_call
+def validate_and_sanitize_inputs(context: ToolCallHookContext) -> bool | None:
+    """Valida e sanitiza entradas."""
+    # Validar consultas de busca
+    if context.tool_name == 'web_search':
+        query = context.tool_input.get('query', '')
+        if len(query) < 3:
+            print("❌ Consulta de busca muito curta")
+            return False
+        
+        # Sanitizar consulta
+        context.tool_input['query'] = query.strip().lower()
+    
+    # Validar caminhos de arquivo
+    if context.tool_name == 'read_file':
+        path = context.tool_input.get('path', '')
+        if '..' in path or path.startswith('/'):
+            print("❌ Caminho de arquivo inválido")
+            return False
+    
+    return None
+```
+
+### 4. Sanitização de Resultado
+
+```python
+@after_tool_call
+def sanitize_sensitive_data(context: ToolCallHookContext) -> str | None:
+    """Sanitiza dados sensíveis."""
+    if not context.tool_result:
+        return None
+    
+    import re
+    result = context.tool_result
+    
+    # Remover chaves de API
+    result = re.sub(
+        r'(api[_-]?key|token)["\']?\s*[:=]\s*["\']?[\w-]+',
+        r'\1: [CENSURADO]',
+        result,
+        flags=re.IGNORECASE
+    )
+    
+    # Remover endereços de email
+    result = re.sub(
+        r'\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Z|a-z]{2,}\b',
+        '[EMAIL-CENSURADO]',
+        result
+    )
+    
+    # Remover números de cartão de crédito
+    result = re.sub(
+        r'\b\d{4}[- ]?\d{4}[- ]?\d{4}[- ]?\d{4}\b',
+        '[CARTÃO-CENSURADO]',
+        result
+    )
+    
+    return result
+```
+
+### 5. Análise de Uso de Ferramenta
+
+```python
+import time
+from collections import defaultdict
+
+tool_stats = defaultdict(lambda: {'count': 0, 'total_time': 0, 'failures': 0})
+
+@before_tool_call
+def start_timer(context: ToolCallHookContext) -> None:
+    context.tool_input['_start_time'] = time.time()
+    return None
+
+@after_tool_call
+def track_tool_usage(context: ToolCallHookContext) -> None:
+    start_time = context.tool_input.get('_start_time', time.time())
+    duration = time.time() - start_time
+    
+    tool_stats[context.tool_name]['count'] += 1
+    tool_stats[context.tool_name]['total_time'] += duration
+    
+    if not context.tool_result or 'error' in context.tool_result.lower():
+        tool_stats[context.tool_name]['failures'] += 1
+    
+    print(f"""
+    📊 Estatísticas da Ferramenta {context.tool_name}:
+    - Execuções: {tool_stats[context.tool_name]['count']}
+    - Tempo Médio: {tool_stats[context.tool_name]['total_time'] / tool_stats[context.tool_name]['count']:.2f}s
+    - Falhas: {tool_stats[context.tool_name]['failures']}
+    """)
+    
+    return None
+```
+
+### 6. Limitação de Taxa
+
+```python
+from collections import defaultdict
+from datetime import datetime, timedelta
+
+tool_call_history = defaultdict(list)
+
+@before_tool_call
+def rate_limit_tools(context: ToolCallHookContext) -> bool | None:
+    """Limita taxa de chamadas de ferramenta."""
+    tool_name = context.tool_name
+    now = datetime.now()
+    
+    # Limpar entradas antigas (mais antigas que 1 minuto)
+    tool_call_history[tool_name] = [
+        call_time for call_time in tool_call_history[tool_name]
+        if now - call_time < timedelta(minutes=1)
+    ]
+    
+    # Verificar limite de taxa (máximo 10 chamadas por minuto)
+    if len(tool_call_history[tool_name]) >= 10:
+        print(f"🚫 Limite de taxa excedido para {tool_name}")
+        return False
+    
+    # Registrar esta chamada
+    tool_call_history[tool_name].append(now)
+    return None
+```
+
+### 7. Logging de Debug
+
+```python
+@before_tool_call
+def debug_tool_call(context: ToolCallHookContext) -> None:
+    """Debug de chamada de ferramenta."""
+    print(f"""
+    🔍 Debug de Chamada de Ferramenta:
+    - Ferramenta: {context.tool_name}
+    - Agente: {context.agent.role if context.agent else 'Desconhecido'}
+    - Tarefa: {context.task.description[:50] if context.task else 'Desconhecida'}...
+    - Entrada: {context.tool_input}
+    """)
+    return None
+
+@after_tool_call
+def debug_tool_result(context: ToolCallHookContext) -> None:
+    """Debug de resultado de ferramenta."""
+    if context.tool_result:
+        result_preview = context.tool_result[:200]
+        print(f"✅ Preview do Resultado: {result_preview}...")
+    else:
+        print("⚠️  Nenhum resultado retornado")
+    return None
+```
+
+## Gerenciamento de Hooks
+
+### Desregistrando Hooks
+
+```python
+from crewai.hooks import (
+    unregister_before_tool_call_hook,
+    unregister_after_tool_call_hook
+)
+
+# Desregistrar hook específico
+def my_hook(context):
+    ...
+
+register_before_tool_call_hook(my_hook)
+# Mais tarde...
+success = unregister_before_tool_call_hook(my_hook)
+print(f"Desregistrado: {success}")
+```
+
+### Limpando Hooks
+
+```python
+from crewai.hooks import (
+    clear_before_tool_call_hooks,
+    clear_after_tool_call_hooks,
+    clear_all_tool_call_hooks
+)
+
+# Limpar tipo específico de hook
+count = clear_before_tool_call_hooks()
+print(f"Limpou {count} hooks antes")
+
+# Limpar todos os hooks de ferramenta
+before_count, after_count = clear_all_tool_call_hooks()
+print(f"Limpou {before_count} hooks antes e {after_count} hooks depois")
+```
+
+## Padrões Avançados
+
+### Execução Condicional de Hook
+
+```python
+@before_tool_call
+def conditional_blocking(context: ToolCallHookContext) -> bool | None:
+    """Bloqueia apenas em condições específicas."""
+    # Bloquear apenas para agentes específicos
+    if context.agent and context.agent.role == "junior_agent":
+        if context.tool_name in ['delete_file', 'send_email']:
+            print(f"❌ Agentes júnior não podem usar {context.tool_name}")
+            return False
+    
+    # Bloquear apenas durante tarefas específicas
+    if context.task and "sensível" in context.task.description.lower():
+        if context.tool_name == 'web_search':
+            print("❌ Busca na web bloqueada para tarefas sensíveis")
+            return False
+    
+    return None
+```
+
+### Modificação de Entrada com Consciência de Contexto
+
+```python
+@before_tool_call
+def enhance_tool_inputs(context: ToolCallHookContext) -> None:
+    """Adiciona contexto baseado no papel do agente."""
+    # Adicionar contexto baseado no papel do agente
+    if context.agent and context.agent.role == "researcher":
+        if context.tool_name == 'web_search':
+            # Adicionar restrições de domínio para pesquisadores
+            context.tool_input['domains'] = ['edu', 'gov', 'org']
+    
+    # Adicionar contexto baseado na tarefa
+    if context.task and "urgente" in context.task.description.lower():
+        if context.tool_name == 'send_email':
+            context.tool_input['priority'] = 'high'
+    
+    return None
+```
+
+## Melhores Práticas
+
+1. **Mantenha Hooks Focados**: Cada hook deve ter uma responsabilidade única
+2. **Evite Computação Pesada**: Hooks executam em cada chamada de ferramenta
+3. **Trate Erros Graciosamente**: Use try-except para prevenir falhas de hooks
+4. **Use Type Hints**: Aproveite `ToolCallHookContext` para melhor suporte IDE
+5. **Documente Condições de Bloqueio**: Deixe claro quando/por que ferramentas são bloqueadas
+6. **Teste Hooks Independentemente**: Teste unitário de hooks antes de usar em produção
+7. **Limpe Hooks em Testes**: Use `clear_all_tool_call_hooks()` entre execuções de teste
+8. **Modifique In-Place**: Sempre modifique `context.tool_input` in-place, nunca substitua
+9. **Registre Decisões Importantes**: Especialmente ao bloquear execução de ferramenta
+10. **Considere Performance**: Cache validações caras quando possível
+
+## Tratamento de Erros
+
+```python
+@before_tool_call
+def safe_validation(context: ToolCallHookContext) -> bool | None:
+    try:
+        # Sua lógica de validação
+        if not validate_input(context.tool_input):
+            return False
+    except Exception as e:
+        print(f"⚠️ Erro no hook: {e}")
+        # Decida: permitir ou bloquear em erro
+        return None  # Permitir execução apesar do erro
+```
+
+## Segurança de Tipos
+
+```python
+from crewai.hooks import ToolCallHookContext, BeforeToolCallHookType, AfterToolCallHookType
+
+# Anotações de tipo explícitas
+def my_before_hook(context: ToolCallHookContext) -> bool | None:
+    return None
+
+def my_after_hook(context: ToolCallHookContext) -> str | None:
+    return None
+
+# Registro type-safe
+register_before_tool_call_hook(my_before_hook)
+register_after_tool_call_hook(my_after_hook)
+```
+
+## Solução de Problemas
+
+### Hook Não Está Executando
+- Verifique se hook está registrado antes da execução da crew
+- Verifique se hook anterior retornou `False` (bloqueia execução e hooks subsequentes)
+- Garanta que assinatura do hook corresponda ao tipo esperado
+
+### Modificações de Entrada Não Funcionam
+- Use modificações in-place: `context.tool_input['key'] = value`
+- Não substitua o dict: `context.tool_input = {}`
+
+### Modificações de Resultado Não Funcionam
+- Retorne a string modificada dos hooks posteriores
+- Retornar `None` mantém o resultado original
+- Garanta que a ferramenta realmente retornou um resultado
+
+### Ferramenta Bloqueada Inesperadamente
+- Verifique todos os hooks antes por condições de bloqueio
+- Verifique ordem de execução do hook
+- Adicione logging de debug para identificar qual hook está bloqueando
+
+## Conclusão
+
+Os Hooks de Chamada de Ferramenta fornecem capacidades poderosas para controlar e monitorar execução de ferramentas no CrewAI. Use-os para implementar guardrails de segurança, gates de aprovação, validação de entrada, sanitização de resultado, logging e análise. Combinados com tratamento adequado de erros e segurança de tipos, os hooks permitem sistemas de agentes seguros e prontos para produção com observabilidade abrangente.
+

docs/pt-BR/observability/portkey.mdx

@@ -733,9 +733,7 @@ Aqui está um exemplo básico para rotear requisições ao OpenAI, usando especi
     - Coletam metadados relevantes para filtragem de logs
     - Impõem permissões de acesso
 
-    Crie chaves de API através de:
-    - [Portkey App](https://app.portkey.ai/)
-    - [API Key Management API](/pt-BR/api-reference/admin-api/control-plane/api-keys/create-api-key)
+    Crie chaves de API através do [Portkey App](https://app.portkey.ai/)
 
     Exemplo usando Python SDK:
     ```python
@@ -758,7 +756,7 @@ Aqui está um exemplo básico para rotear requisições ao OpenAI, usando especi
     )
     ```
 
-    Para instruções detalhadas de gerenciamento de chaves, veja nossa [documentação de API Keys](/pt-BR/api-reference/admin-api/control-plane/api-keys/create-api-key).
+    Para instruções detalhadas de gerenciamento de chaves, veja a [documentação Portkey](https://portkey.ai/docs).
   </Accordion>
 
   <Accordion title="Etapa 4: Implante & Monitore">

docs/pt-BR/tools/cloud-storage/overview.mdx

@@ -18,7 +18,7 @@ Essas ferramentas permitem que seus agentes interajam com serviços em nuvem, ac
     Escreva e faça upload de arquivos para o armazenamento Amazon S3.
   </Card>
 
-  <Card title="Bedrock Invoke Agent" icon="aws" href="/pt-BR/tools/cloud-storage/bedrockinvokeagenttool">
+  <Card title="Bedrock Invoke Agent" icon="aws" href="/pt-BR/tools/integration/bedrockinvokeagenttool">
     Acione agentes Amazon Bedrock para tarefas orientadas por IA.
   </Card>
 

docs/pt-BR/tools/tool-integrations/overview.mdx

@@ -11,7 +11,7 @@ mode: "wide"
   <Card
     title="Bedrock Invoke Agent Tool"
     icon="cloud"
-    href="/en/tools/tool-integrations/bedrockinvokeagenttool"
+    href="/pt-BR/tools/integration/bedrockinvokeagenttool"
     color="#0891B2"
   >
     Invoke Amazon Bedrock Agents from CrewAI to orchestrate actions across AWS services.
@@ -20,7 +20,7 @@ mode: "wide"
   <Card
     title="CrewAI Automation Tool"
     icon="bolt"
-    href="/en/tools/tool-integrations/crewaiautomationtool"
+    href="/pt-BR/tools/integration/crewaiautomationtool"
     color="#7C3AED"
   >
     Automate deployment and operations by integrating CrewAI with external platforms and workflows.

lib/crewai-tools/src/crewai_tools/tools/qdrant_vector_search_tool/qdrant_search_tool.py

@@ -12,12 +12,16 @@ from pydantic.types import ImportString
 
 
 class QdrantToolSchema(BaseModel):
-    query: str = Field(..., description="Query to search in Qdrant DB")
+    query: str = Field(
+        ..., description="Query to search in Qdrant DB - always required."
+    )
     filter_by: str | None = Field(
-        default=None, description="Parameter to filter the search by."
+        default=None,
+        description="Parameter to filter the search by. When filtering, needs to be used in conjunction with filter_value.",
     )
     filter_value: Any | None = Field(
-        default=None, description="Value to filter the search by."
+        default=None,
+        description="Value to filter the search by. When filtering, needs to be used in conjunction with filter_by.",
     )
 
 

lib/crewai/src/crewai/__init__.py

@@ -8,8 +8,8 @@ from crewai.crew import Crew
 from crewai.crews.crew_output import CrewOutput
 from crewai.flow.flow import Flow
 from crewai.knowledge.knowledge import Knowledge
-from crewai.llm import LLM
-from crewai.llms.base_llm import BaseLLM
+from crewai.llm.base_llm import BaseLLM
+from crewai.llm.core import LLM
 from crewai.process import Process
 from crewai.task import Task
 from crewai.tasks.llm_guardrail import LLMGuardrail

lib/crewai/src/crewai/a2a/config.py

@@ -38,6 +38,7 @@ class A2AConfig(BaseModel):
         max_turns: Maximum conversation turns with A2A agent (default: 10).
         response_model: Optional Pydantic model for structured A2A agent responses.
         fail_fast: If True, raise error when agent unreachable; if False, skip and continue (default: True).
+        trust_remote_completion_status: If True, return A2A agent's result directly when status is "completed"; if False, always ask server agent to respond (default: False).
     """
 
     endpoint: Url = Field(description="A2A agent endpoint URL")
@@ -57,3 +58,7 @@ class A2AConfig(BaseModel):
         default=True,
         description="If True, raise an error immediately when the A2A agent is unreachable. If False, skip the A2A agent and continue execution.",
     )
+    trust_remote_completion_status: bool = Field(
+        default=False,
+        description='If True, return the A2A agent\'s result directly when status is "completed" without asking the server agent to respond. If False, always ask the server agent to respond, allowing it to potentially delegate again.',
+    )

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

@@ -52,7 +52,7 @@ def wrap_agent_with_a2a_instance(agent: Agent) -> None:
     Args:
         agent: The agent instance to wrap
     """
-    original_execute_task = agent.execute_task.__func__
+    original_execute_task = agent.execute_task.__func__  # type: ignore[attr-defined]
 
     @wraps(original_execute_task)
     def execute_task_with_a2a(
@@ -73,7 +73,7 @@ def wrap_agent_with_a2a_instance(agent: Agent) -> None:
             Task execution result
         """
         if not self.a2a:
-            return original_execute_task(self, task, context, tools)
+            return original_execute_task(self, task, context, tools)  # type: ignore[no-any-return]
 
         a2a_agents, agent_response_model = get_a2a_agents_and_response_model(self.a2a)
 
@@ -498,6 +498,23 @@ def _delegate_to_a2a(
             conversation_history = a2a_result.get("history", [])
 
             if a2a_result["status"] in ["completed", "input_required"]:
+                if (
+                    a2a_result["status"] == "completed"
+                    and agent_config.trust_remote_completion_status
+                ):
+                    result_text = a2a_result.get("result", "")
+                    final_turn_number = turn_num + 1
+                    crewai_event_bus.emit(
+                        None,
+                        A2AConversationCompletedEvent(
+                            status="completed",
+                            final_result=result_text,
+                            error=None,
+                            total_turns=final_turn_number,
+                        ),
+                    )
+                    return result_text  # type: ignore[no-any-return]
+
                 final_result, next_request = _handle_agent_response_and_continue(
                     self=self,
                     a2a_result=a2a_result,

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

@@ -39,7 +39,7 @@ 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.llm.base_llm import BaseLLM
 from crewai.mcp import (
     MCPClient,
     MCPServerConfig,
@@ -119,6 +119,7 @@ class Agent(BaseAgent):
 
     _times_executed: int = PrivateAttr(default=0)
     _mcp_clients: list[Any] = PrivateAttr(default_factory=list)
+    _last_messages: list[LLMMessage] = PrivateAttr(default_factory=list)
     max_execution_time: int | None = Field(
         default=None,
         description="Maximum execution time for an agent to execute a task",
@@ -538,6 +539,12 @@ class Agent(BaseAgent):
             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 []
+        )
+
         self._cleanup_mcp_clients()
 
         return result
@@ -626,7 +633,7 @@ 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,
@@ -803,6 +810,7 @@ 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,
@@ -896,10 +904,12 @@ class Agent(BaseAgent):
                                 server_name=server_name,
                                 run_context=None,
                             )
-                            if mcp_config.tool_filter(context, tool):
+                            # Try new signature first
+                            if mcp_config.tool_filter(context, tool):  # type: ignore[arg-type,call-arg]
                                 filtered_tools.append(tool)
                         except (TypeError, AttributeError):
-                            if mcp_config.tool_filter(tool):
+                            # Fallback to old signature
+                            if mcp_config.tool_filter(tool):  # type: ignore[arg-type,call-arg]
                                 filtered_tools.append(tool)
                     else:
                         # Not callable - include tool
@@ -974,7 +984,9 @@ 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"]
 
@@ -988,7 +1000,7 @@ class Agent(BaseAgent):
                 self._logger.log(
                     "debug", f"Using cached MCP tool schemas for {server_url}"
                 )
-                return cached_data
+                return cast(dict[str, dict[str, Any]], cached_data)
 
         try:
             schemas = asyncio.run(self._get_mcp_tool_schemas_async(server_params))
@@ -1006,7 +1018,7 @@ 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(
@@ -1014,7 +1026,7 @@ 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
@@ -1045,7 +1057,7 @@ 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:
@@ -1149,13 +1161,13 @@ class Agent(BaseAgent):
                     Field(..., description=field_description),
                 )
             else:
-                field_definitions[field_name] = (
+                field_definitions[field_name] = (  # type: ignore[assignment]
                     field_type | None,
                     Field(default=None, description=field_description),
                 )
 
         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,call-overload]
 
     def _json_type_to_python(self, field_schema: dict[str, Any]) -> type:
         """Convert JSON Schema type to Python type.
@@ -1175,16 +1187,16 @@ class Agent(BaseAgent):
                 if "const" in option:
                     types.append(str)
                 else:
-                    types.append(self._json_type_to_python(option))
+                    types.append(self._json_type_to_python(option))  # type: ignore[arg-type]
             unique_types = list(set(types))
             if len(unique_types) > 1:
                 result = unique_types[0]
                 for t in unique_types[1:]:
-                    result = result | t
+                    result = result | t  # type: ignore[assignment]
                 return result
             return unique_types[0]
 
-        type_mapping = {
+        type_mapping: dict[str, type] = {
             "string": str,
             "number": float,
             "integer": int,
@@ -1193,10 +1205,10 @@ class Agent(BaseAgent):
             "object": dict,
         }
 
-        return type_mapping.get(json_type, Any)
+        return type_mapping.get(json_type or "", 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 AMP API."""
         # TODO: Implement AMP API call to "integrations/mcps" endpoint
         # Should return list of server configs with URLs
@@ -1341,6 +1353,15 @@ class Agent(BaseAgent):
     def set_fingerprint(self, fingerprint: Fingerprint) -> None:
         self.security_config.fingerprint = fingerprint
 
+    @property
+    def last_messages(self) -> list[LLMMessage]:
+        """Get messages from the last task execution.
+
+        Returns:
+            List of LLM messages from the most recent task execution.
+        """
+        return self._last_messages
+
     def _get_knowledge_search_query(self, task_prompt: str, task: Task) -> str | None:
         """Generate a search query for the knowledge base based on the task description."""
         crewai_event_bus.emit(

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

@@ -137,7 +137,7 @@ class BaseAgent(BaseModel, ABC, metaclass=AgentMeta):
         default=False,
         description="Enable agent to delegate and ask questions among each other.",
     )
-    tools: list[BaseTool] | None = Field(
+    tools: list[BaseTool] = Field(
         default_factory=list, description="Tools at agents' disposal"
     )
     max_iter: int = Field(
@@ -161,7 +161,7 @@ class BaseAgent(BaseModel, ABC, metaclass=AgentMeta):
         description="An instance of the ToolsHandler class.",
     )
     tools_results: list[dict[str, Any]] = Field(
-        default=[], description="Results of the tools used by the agent."
+        default_factory=list, description="Results of the tools used by the agent."
     )
     max_tokens: int | None = Field(
         default=None, description="Maximum number of tokens for the agent's execution."
@@ -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:")):

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

@@ -23,6 +23,10 @@ from crewai.events.types.logging_events import (
     AgentLogsExecutionEvent,
     AgentLogsStartedEvent,
 )
+from crewai.hooks.llm_hooks import (
+    get_after_llm_call_hooks,
+    get_before_llm_call_hooks,
+)
 from crewai.utilities.agent_utils import (
     enforce_rpm_limit,
     format_message_for_llm,
@@ -47,7 +51,7 @@ if TYPE_CHECKING:
     from crewai.agent import Agent
     from crewai.agents.tools_handler import ToolsHandler
     from crewai.crew import Crew
-    from crewai.llms.base_llm import BaseLLM
+    from crewai.llm.base_llm import BaseLLM
     from crewai.task import Task
     from crewai.tools.base_tool import BaseTool
     from crewai.tools.structured_tool import CrewStructuredTool
@@ -130,6 +134,10 @@ 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.extend(get_before_llm_call_hooks())
+        self.after_llm_call_hooks.extend(get_after_llm_call_hooks())
         if self.llm:
             # This may be mutating the shared llm object and needs further evaluation
             existing_stop = getattr(self.llm, "stop", [])
@@ -226,6 +234,7 @@ class CrewAgentExecutor(CrewAgentExecutorMixin):
                     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]
 
@@ -254,6 +263,7 @@ class CrewAgentExecutor(CrewAgentExecutorMixin):
                         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

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

@@ -1,5 +1,5 @@
 import time
-from typing import Any
+from typing import TYPE_CHECKING, Any, TypeVar, cast
 import webbrowser
 
 from pydantic import BaseModel, Field
@@ -13,6 +13,8 @@ from crewai.cli.shared.token_manager import TokenManager
 
 console = Console()
 
+TOauth2Settings = TypeVar("TOauth2Settings", bound="Oauth2Settings")
+
 
 class Oauth2Settings(BaseModel):
     provider: str = Field(
@@ -28,9 +30,15 @@ class Oauth2Settings(BaseModel):
         description="OAuth2 audience value, typically used to identify the target API or resource.",
         default=None,
     )
+    extra: dict[str, Any] = Field(
+        description="Extra configuration for the OAuth2 provider.",
+        default={},
+    )
 
     @classmethod
-    def from_settings(cls):
+    def from_settings(cls: type[TOauth2Settings]) -> TOauth2Settings:
+        """Create an Oauth2Settings instance from the CLI settings."""
+
         settings = Settings()
 
         return cls(
@@ -38,12 +46,20 @@ class Oauth2Settings(BaseModel):
             domain=settings.oauth2_domain,
             client_id=settings.oauth2_client_id,
             audience=settings.oauth2_audience,
+            extra=settings.oauth2_extra,
         )
 
 
+if TYPE_CHECKING:
+    from crewai.cli.authentication.providers.base_provider import BaseProvider
+
+
 class ProviderFactory:
     @classmethod
-    def from_settings(cls, settings: Oauth2Settings | None = None):
+    def from_settings(
+        cls: type["ProviderFactory"],  # noqa: UP037
+        settings: Oauth2Settings | None = None,
+    ) -> "BaseProvider":  # noqa: UP037
         settings = settings or Oauth2Settings.from_settings()
 
         import importlib
@@ -53,11 +69,11 @@ class ProviderFactory:
         )
         provider = getattr(module, f"{settings.provider.capitalize()}Provider")
 
-        return provider(settings)
+        return cast("BaseProvider", provider(settings))
 
 
 class AuthenticationCommand:
-    def __init__(self):
+    def __init__(self) -> None:
         self.token_manager = TokenManager()
         self.oauth2_provider = ProviderFactory.from_settings()
 
@@ -84,7 +100,7 @@ class AuthenticationCommand:
             timeout=20,
         )
         response.raise_for_status()
-        return response.json()
+        return cast(dict[str, Any], response.json())
 
     def _display_auth_instructions(self, device_code_data: dict[str, str]) -> None:
         """Display the authentication instructions to the user."""

lib/crewai/src/crewai/cli/authentication/providers/base_provider.py

@@ -24,3 +24,7 @@ class BaseProvider(ABC):
 
     @abstractmethod
     def get_client_id(self) -> str: ...
+
+    def get_required_fields(self) -> list[str]:
+        """Returns which provider-specific fields inside the "extra" dict will be required"""
+        return []

lib/crewai/src/crewai/cli/authentication/providers/okta.py

@@ -3,16 +3,16 @@ from crewai.cli.authentication.providers.base_provider import BaseProvider
 
 class OktaProvider(BaseProvider):
     def get_authorize_url(self) -> str:
-        return f"https://{self.settings.domain}/oauth2/default/v1/device/authorize"
+        return f"{self._oauth2_base_url()}/v1/device/authorize"
 
     def get_token_url(self) -> str:
-        return f"https://{self.settings.domain}/oauth2/default/v1/token"
+        return f"{self._oauth2_base_url()}/v1/token"
 
     def get_jwks_url(self) -> str:
-        return f"https://{self.settings.domain}/oauth2/default/v1/keys"
+        return f"{self._oauth2_base_url()}/v1/keys"
 
     def get_issuer(self) -> str:
-        return f"https://{self.settings.domain}/oauth2/default"
+        return self._oauth2_base_url().removesuffix("/oauth2")
 
     def get_audience(self) -> str:
         if self.settings.audience is None:
@@ -27,3 +27,16 @@ class OktaProvider(BaseProvider):
                 "Client ID is required. Please set it in the configuration."
             )
         return self.settings.client_id
+
+    def get_required_fields(self) -> list[str]:
+        return ["authorization_server_name", "using_org_auth_server"]
+
+    def _oauth2_base_url(self) -> str:
+        using_org_auth_server = self.settings.extra.get("using_org_auth_server", False)
+
+        if using_org_auth_server:
+            base_url = f"https://{self.settings.domain}/oauth2"
+        else:
+            base_url = f"https://{self.settings.domain}/oauth2/{self.settings.extra.get('authorization_server_name', 'default')}"
+
+        return f"{base_url}"

lib/crewai/src/crewai/cli/command.py

@@ -11,18 +11,18 @@ console = Console()
 
 
 class BaseCommand:
-    def __init__(self):
+    def __init__(self) -> None:
         self._telemetry = Telemetry()
         self._telemetry.set_tracer()
 
 
 class PlusAPIMixin:
-    def __init__(self, telemetry):
+    def __init__(self, telemetry: Telemetry) -> None:
         try:
             telemetry.set_tracer()
             self.plus_api_client = PlusAPI(api_key=get_auth_token())
         except Exception:
-            self._deploy_signup_error_span = telemetry.deploy_signup_error_span()
+            telemetry.deploy_signup_error_span()
             console.print(
                 "Please sign up/login to CrewAI+ before using the CLI.",
                 style="bold red",

lib/crewai/src/crewai/cli/config.py

@@ -2,6 +2,7 @@ import json
 from logging import getLogger
 from pathlib import Path
 import tempfile
+from typing import Any
 
 from pydantic import BaseModel, Field
 
@@ -136,7 +137,12 @@ class Settings(BaseModel):
         default=DEFAULT_CLI_SETTINGS["oauth2_domain"],
     )
 
-    def __init__(self, config_path: Path | None = None, **data):
+    oauth2_extra: dict[str, Any] = Field(
+        description="Extra configuration for the OAuth2 provider.",
+        default={},
+    )
+
+    def __init__(self, config_path: Path | None = None, **data: dict[str, Any]) -> None:
         """Load Settings from config path with fallback support"""
         if config_path is None:
             config_path = get_writable_config_path()

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.llm.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

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

@@ -1,9 +1,10 @@
-from typing import Any
+from typing import Any, cast
 
 import requests
 from requests.exceptions import JSONDecodeError, RequestException
 from rich.console import Console
 
+from crewai.cli.authentication.main import Oauth2Settings, ProviderFactory
 from crewai.cli.command import BaseCommand
 from crewai.cli.settings.main import SettingsCommand
 from crewai.cli.version import get_crewai_version
@@ -13,7 +14,7 @@ console = Console()
 
 
 class EnterpriseConfigureCommand(BaseCommand):
-    def __init__(self):
+    def __init__(self) -> None:
         super().__init__()
         self.settings_command = SettingsCommand()
 
@@ -54,25 +55,12 @@ class EnterpriseConfigureCommand(BaseCommand):
             except JSONDecodeError as e:
                 raise ValueError(f"Invalid JSON response from {oauth_endpoint}") from e
 
-            required_fields = [
-                "audience",
-                "domain",
-                "device_authorization_client_id",
-                "provider",
-            ]
-            missing_fields = [
-                field for field in required_fields if field not in oauth_config
-            ]
-
-            if missing_fields:
-                raise ValueError(
-                    f"Missing required fields in OAuth2 configuration: {', '.join(missing_fields)}"
-                )
+            self._validate_oauth_config(oauth_config)
 
             console.print(
                 "✅ Successfully retrieved OAuth2 configuration", style="green"
             )
-            return oauth_config
+            return cast(dict[str, Any], oauth_config)
 
         except RequestException as e:
             raise ValueError(f"Failed to connect to enterprise URL: {e!s}") from e
@@ -89,6 +77,7 @@ class EnterpriseConfigureCommand(BaseCommand):
                 "oauth2_audience": oauth_config["audience"],
                 "oauth2_client_id": oauth_config["device_authorization_client_id"],
                 "oauth2_domain": oauth_config["domain"],
+                "oauth2_extra": oauth_config["extra"],
             }
 
             console.print("🔄 Updating local OAuth2 configuration...")
@@ -99,3 +88,38 @@ class EnterpriseConfigureCommand(BaseCommand):
 
         except Exception as e:
             raise ValueError(f"Failed to update OAuth2 settings: {e!s}") from e
+
+    def _validate_oauth_config(self, oauth_config: dict[str, Any]) -> None:
+        required_fields = [
+            "audience",
+            "domain",
+            "device_authorization_client_id",
+            "provider",
+            "extra",
+        ]
+
+        missing_basic_fields = [
+            field for field in required_fields if field not in oauth_config
+        ]
+        missing_provider_specific_fields = [
+            field
+            for field in self._get_provider_specific_fields(oauth_config["provider"])
+            if field not in oauth_config.get("extra", {})
+        ]
+
+        if missing_basic_fields:
+            raise ValueError(
+                f"Missing required fields in OAuth2 configuration: [{', '.join(missing_basic_fields)}]"
+            )
+
+        if missing_provider_specific_fields:
+            raise ValueError(
+                f"Missing authentication provider required fields in OAuth2 configuration: [{', '.join(missing_provider_specific_fields)}] (Configured provider: '{oauth_config['provider']}')"
+            )
+
+    def _get_provider_specific_fields(self, provider_name: str) -> list[str]:
+        provider = ProviderFactory.from_settings(
+            Oauth2Settings(provider=provider_name, client_id="dummy", domain="dummy")
+        )
+
+        return provider.get_required_fields()

lib/crewai/src/crewai/cli/git.py

@@ -3,7 +3,7 @@ import subprocess
 
 
 class Repository:
-    def __init__(self, path="."):
+    def __init__(self, path: str = ".") -> None:
         self.path = path
 
         if not self.is_git_installed():

lib/crewai/src/crewai/cli/plus_api.py

@@ -1,3 +1,4 @@
+from typing import Any
 from urllib.parse import urljoin
 
 import requests
@@ -36,19 +37,21 @@ class PlusAPI:
             str(settings.enterprise_base_url) or DEFAULT_CREWAI_ENTERPRISE_URL
         )
 
-    def _make_request(self, method: str, endpoint: str, **kwargs) -> requests.Response:
+    def _make_request(
+        self, method: str, endpoint: str, **kwargs: Any
+    ) -> requests.Response:
         url = urljoin(self.base_url, endpoint)
         session = requests.Session()
         session.trust_env = False
         return session.request(method, url, headers=self.headers, **kwargs)
 
-    def login_to_tool_repository(self):
+    def login_to_tool_repository(self) -> requests.Response:
         return self._make_request("POST", f"{self.TOOLS_RESOURCE}/login")
 
-    def get_tool(self, handle: str):
+    def get_tool(self, handle: str) -> requests.Response:
         return self._make_request("GET", f"{self.TOOLS_RESOURCE}/{handle}")
 
-    def get_agent(self, handle: str):
+    def get_agent(self, handle: str) -> requests.Response:
         return self._make_request("GET", f"{self.AGENTS_RESOURCE}/{handle}")
 
     def publish_tool(
@@ -58,8 +61,8 @@ class PlusAPI:
         version: str,
         description: str | None,
         encoded_file: str,
-        available_exports: list[str] | None = None,
-    ):
+        available_exports: list[dict[str, Any]] | None = None,
+    ) -> requests.Response:
         params = {
             "handle": handle,
             "public": is_public,
@@ -111,13 +114,13 @@ class PlusAPI:
     def list_crews(self) -> requests.Response:
         return self._make_request("GET", self.CREWS_RESOURCE)
 
-    def create_crew(self, payload) -> requests.Response:
+    def create_crew(self, payload: dict[str, Any]) -> requests.Response:
         return self._make_request("POST", self.CREWS_RESOURCE, json=payload)
 
     def get_organizations(self) -> requests.Response:
         return self._make_request("GET", self.ORGANIZATIONS_RESOURCE)
 
-    def initialize_trace_batch(self, payload) -> requests.Response:
+    def initialize_trace_batch(self, payload: dict[str, Any]) -> requests.Response:
         return self._make_request(
             "POST",
             f"{self.TRACING_RESOURCE}/batches",
@@ -125,14 +128,18 @@ class PlusAPI:
             timeout=30,
         )
 
-    def initialize_ephemeral_trace_batch(self, payload) -> requests.Response:
+    def initialize_ephemeral_trace_batch(
+        self, payload: dict[str, Any]
+    ) -> requests.Response:
         return self._make_request(
             "POST",
             f"{self.EPHEMERAL_TRACING_RESOURCE}/batches",
             json=payload,
         )
 
-    def send_trace_events(self, trace_batch_id: str, payload) -> requests.Response:
+    def send_trace_events(
+        self, trace_batch_id: str, payload: dict[str, Any]
+    ) -> requests.Response:
         return self._make_request(
             "POST",
             f"{self.TRACING_RESOURCE}/batches/{trace_batch_id}/events",
@@ -141,7 +148,7 @@ class PlusAPI:
         )
 
     def send_ephemeral_trace_events(
-        self, trace_batch_id: str, payload
+        self, trace_batch_id: str, payload: dict[str, Any]
     ) -> requests.Response:
         return self._make_request(
             "POST",
@@ -150,7 +157,9 @@ class PlusAPI:
             timeout=30,
         )
 
-    def finalize_trace_batch(self, trace_batch_id: str, payload) -> requests.Response:
+    def finalize_trace_batch(
+        self, trace_batch_id: str, payload: dict[str, Any]
+    ) -> requests.Response:
         return self._make_request(
             "PATCH",
             f"{self.TRACING_RESOURCE}/batches/{trace_batch_id}/finalize",
@@ -159,7 +168,7 @@ class PlusAPI:
         )
 
     def finalize_ephemeral_trace_batch(
-        self, trace_batch_id: str, payload
+        self, trace_batch_id: str, payload: dict[str, Any]
     ) -> requests.Response:
         return self._make_request(
             "PATCH",

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

@@ -34,7 +34,7 @@ class SettingsCommand(BaseCommand):
             current_value = getattr(self.settings, field_name)
             description = field_info.description or "No description available"
             display_value = (
-                str(current_value) if current_value is not None else "Not set"
+                str(current_value) if current_value not in [None, {}] else "Not set"
             )
 
             table.add_row(field_name, display_value, description)

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

@@ -30,11 +30,11 @@ class ToolCommand(BaseCommand, PlusAPIMixin):
     A class to handle tool repository related operations for CrewAI projects.
     """
 
-    def __init__(self):
+    def __init__(self) -> None:
         BaseCommand.__init__(self)
         PlusAPIMixin.__init__(self, telemetry=self._telemetry)
 
-    def create(self, handle: str):
+    def create(self, handle: str) -> None:
         self._ensure_not_in_project()
 
         folder_name = handle.replace(" ", "_").replace("-", "_").lower()
@@ -64,7 +64,7 @@ class ToolCommand(BaseCommand, PlusAPIMixin):
         finally:
             os.chdir(old_directory)
 
-    def publish(self, is_public: bool, force: bool = False):
+    def publish(self, is_public: bool, force: bool = False) -> None:
         if not git.Repository().is_synced() and not force:
             console.print(
                 "[bold red]Failed to publish tool.[/bold red]\n"
@@ -137,7 +137,7 @@ class ToolCommand(BaseCommand, PlusAPIMixin):
             style="bold green",
         )
 
-    def install(self, handle: str):
+    def install(self, handle: str) -> None:
         self._print_current_organization()
         get_response = self.plus_api_client.get_tool(handle)
 
@@ -180,7 +180,7 @@ class ToolCommand(BaseCommand, PlusAPIMixin):
         settings.org_name = login_response_json["current_organization"]["name"]
         settings.dump()
 
-    def _add_package(self, tool_details: dict[str, Any]):
+    def _add_package(self, tool_details: dict[str, Any]) -> None:
         is_from_pypi = tool_details.get("source", None) == "pypi"
         tool_handle = tool_details["handle"]
         repository_handle = tool_details["repository"]["handle"]
@@ -209,7 +209,7 @@ class ToolCommand(BaseCommand, PlusAPIMixin):
             click.echo(add_package_result.stderr, err=True)
             raise SystemExit
 
-    def _ensure_not_in_project(self):
+    def _ensure_not_in_project(self) -> None:
         if os.path.isfile("./pyproject.toml"):
             console.print(
                 "[bold red]Oops! It looks like you're inside a project.[/bold red]"

lib/crewai/src/crewai/cli/utils.py

@@ -5,7 +5,7 @@ import os
 from pathlib import Path
 import shutil
 import sys
-from typing import Any, get_type_hints
+from typing import Any, cast, get_type_hints
 
 import click
 from rich.console import Console
@@ -23,7 +23,9 @@ if sys.version_info >= (3, 11):
 console = Console()
 
 
-def copy_template(src, dst, name, class_name, folder_name):
+def copy_template(
+    src: Path, dst: Path, name: str, class_name: str, folder_name: str
+) -> None:
     """Copy a file from src to dst."""
     with open(src, "r") as file:
         content = file.read()
@@ -40,13 +42,13 @@ def copy_template(src, dst, name, class_name, folder_name):
     click.secho(f"  - Created {dst}", fg="green")
 
 
-def read_toml(file_path: str = "pyproject.toml"):
+def read_toml(file_path: str = "pyproject.toml") -> dict[str, Any]:
     """Read the content of a TOML file and return it as a dictionary."""
     with open(file_path, "rb") as f:
         return tomli.load(f)
 
 
-def parse_toml(content):
+def parse_toml(content: str) -> dict[str, Any]:
     if sys.version_info >= (3, 11):
         return tomllib.loads(content)
     return tomli.loads(content)
@@ -103,7 +105,7 @@ def _get_project_attribute(
         )
     except Exception as e:
         # Handle TOML decode errors for Python 3.11+
-        if sys.version_info >= (3, 11) and isinstance(e, tomllib.TOMLDecodeError):  # type: ignore
+        if sys.version_info >= (3, 11) and isinstance(e, tomllib.TOMLDecodeError):
             console.print(
                 f"Error: {pyproject_path} is not a valid TOML file.", style="bold red"
             )
@@ -126,7 +128,7 @@ def _get_nested_value(data: dict[str, Any], keys: list[str]) -> Any:
     return reduce(dict.__getitem__, keys, data)
 
 
-def fetch_and_json_env_file(env_file_path: str = ".env") -> dict:
+def fetch_and_json_env_file(env_file_path: str = ".env") -> dict[str, Any]:
     """Fetch the environment variables from a .env file and return them as a dictionary."""
     try:
         # Read the .env file
@@ -150,7 +152,7 @@ def fetch_and_json_env_file(env_file_path: str = ".env") -> dict:
     return {}
 
 
-def tree_copy(source, destination):
+def tree_copy(source: Path, destination: Path) -> None:
     """Copies the entire directory structure from the source to the destination."""
     for item in os.listdir(source):
         source_item = os.path.join(source, item)
@@ -161,7 +163,7 @@ def tree_copy(source, destination):
             shutil.copy2(source_item, destination_item)
 
 
-def tree_find_and_replace(directory, find, replace):
+def tree_find_and_replace(directory: Path, find: str, replace: str) -> None:
     """Recursively searches through a directory, replacing a target string in
     both file contents and filenames with a specified replacement string.
     """
@@ -187,7 +189,7 @@ def tree_find_and_replace(directory, find, replace):
                 os.rename(old_dirpath, new_dirpath)
 
 
-def load_env_vars(folder_path):
+def load_env_vars(folder_path: Path) -> dict[str, Any]:
     """
     Loads environment variables from a .env file in the specified folder path.
 
@@ -208,7 +210,9 @@ def load_env_vars(folder_path):
     return env_vars
 
 
-def update_env_vars(env_vars, provider, model):
+def update_env_vars(
+    env_vars: dict[str, Any], provider: str, model: str
+) -> dict[str, Any] | None:
     """
     Updates environment variables with the API key for the selected provider and model.
 
@@ -220,15 +224,20 @@ def update_env_vars(env_vars, provider, model):
     Returns:
     - None
     """
-    api_key_var = ENV_VARS.get(
-        provider,
-        [
-            click.prompt(
-                f"Enter the environment variable name for your {provider.capitalize()} API key",
-                type=str,
-            )
-        ],
-    )[0]
+    provider_config = cast(
+        list[str],
+        ENV_VARS.get(
+            provider,
+            [
+                click.prompt(
+                    f"Enter the environment variable name for your {provider.capitalize()} API key",
+                    type=str,
+                )
+            ],
+        ),
+    )
+
+    api_key_var = provider_config[0]
 
     if api_key_var not in env_vars:
         try:
@@ -246,7 +255,7 @@ def update_env_vars(env_vars, provider, model):
     return env_vars
 
 
-def write_env_file(folder_path, env_vars):
+def write_env_file(folder_path: Path, env_vars: dict[str, Any]) -> None:
     """
     Writes environment variables to a .env file in the specified folder.
 
@@ -342,18 +351,18 @@ def get_crews(crew_path: str = "crew.py", require: bool = False) -> list[Crew]:
     return crew_instances
 
 
-def get_crew_instance(module_attr) -> Crew | None:
+def get_crew_instance(module_attr: Any) -> Crew | None:
     if (
         callable(module_attr)
         and hasattr(module_attr, "is_crew_class")
         and module_attr.is_crew_class
     ):
-        return module_attr().crew()
+        return cast(Crew, module_attr().crew())
     try:
         if (ismethod(module_attr) or isfunction(module_attr)) and get_type_hints(
             module_attr
         ).get("return") is Crew:
-            return module_attr()
+            return cast(Crew, module_attr())
     except Exception:
         return None
 
@@ -362,7 +371,7 @@ def get_crew_instance(module_attr) -> Crew | None:
     return None
 
 
-def fetch_crews(module_attr) -> list[Crew]:
+def fetch_crews(module_attr: Any) -> list[Crew]:
     crew_instances: list[Crew] = []
 
     if crew_instance := get_crew_instance(module_attr):
@@ -377,7 +386,7 @@ def fetch_crews(module_attr) -> list[Crew]:
     return crew_instances
 
 
-def is_valid_tool(obj):
+def is_valid_tool(obj: Any) -> bool:
     from crewai.tools.base_tool import Tool
 
     if isclass(obj):
@@ -389,7 +398,7 @@ def is_valid_tool(obj):
     return isinstance(obj, Tool)
 
 
-def extract_available_exports(dir_path: str = "src"):
+def extract_available_exports(dir_path: str = "src") -> list[dict[str, Any]]:
     """
     Extract available tool classes from the project's __init__.py files.
     Only includes classes that inherit from BaseTool or functions decorated with @tool.
@@ -419,7 +428,9 @@ def extract_available_exports(dir_path: str = "src"):
         raise SystemExit(1) from e
 
 
-def build_env_with_tool_repository_credentials(repository_handle: str):
+def build_env_with_tool_repository_credentials(
+    repository_handle: str,
+) -> dict[str, Any]:
     repository_handle = repository_handle.upper().replace("-", "_")
     settings = Settings()
 
@@ -472,7 +483,7 @@ def _load_tools_from_init(init_file: Path) -> list[dict[str, Any]]:
         sys.modules.pop("temp_module", None)
 
 
-def _print_no_tools_warning():
+def _print_no_tools_warning() -> None:
     """
     Display warning and usage instructions if no tools were found.
     """

lib/crewai/src/crewai/crew.py

@@ -56,8 +56,8 @@ from crewai.events.types.crew_events import (
 from crewai.flow.flow_trackable import FlowTrackable
 from crewai.knowledge.knowledge import Knowledge
 from crewai.knowledge.source.base_knowledge_source import BaseKnowledgeSource
-from crewai.llm import LLM
-from crewai.llms.base_llm import BaseLLM
+from crewai.llm.base_llm import BaseLLM
+from crewai.llm.core import LLM
 from crewai.memory.entity.entity_memory import EntityMemory
 from crewai.memory.external.external_memory import ExternalMemory
 from crewai.memory.long_term.long_term_memory import LongTermMemory
@@ -809,6 +809,7 @@ class Crew(FlowTrackable, BaseModel):
                 "json_dict": output.json_dict,
                 "output_format": output.output_format,
                 "agent": output.agent,
+                "messages": output.messages,
             },
             "task_index": task_index,
             "inputs": inputs,
@@ -1236,6 +1237,7 @@ class Crew(FlowTrackable, BaseModel):
                 pydantic=stored_output["pydantic"],
                 json_dict=stored_output["json_dict"],
                 output_format=stored_output["output_format"],
+                messages=stored_output.get("messages", []),
             )
             self.tasks[i].output = task_output
 

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

@@ -89,7 +89,7 @@ from crewai.events.types.tool_usage_events import (
     ToolUsageStartedEvent,
 )
 from crewai.events.utils.console_formatter import ConsoleFormatter
-from crewai.llm import LLM
+from crewai.llm.core import LLM
 from crewai.task import Task
 from crewai.telemetry.telemetry import Telemetry
 from crewai.utilities import Logger

lib/crewai/src/crewai/experimental/evaluation/base_evaluator.py

@@ -7,7 +7,7 @@ from pydantic import BaseModel, Field
 
 from crewai.agent import Agent
 from crewai.agents.agent_builder.base_agent import BaseAgent
-from crewai.llm import BaseLLM
+from crewai.llm.base_llm import BaseLLM
 from crewai.task import Task
 from crewai.utilities.llm_utils import create_llm
 

lib/crewai/src/crewai/hooks/__init__.py

@@ -0,0 +1,108 @@
+from __future__ import annotations
+
+from crewai.hooks.decorators import (
+    after_llm_call,
+    after_tool_call,
+    before_llm_call,
+    before_tool_call,
+)
+from crewai.hooks.llm_hooks import (
+    LLMCallHookContext,
+    clear_after_llm_call_hooks,
+    clear_all_llm_call_hooks,
+    clear_before_llm_call_hooks,
+    get_after_llm_call_hooks,
+    get_before_llm_call_hooks,
+    register_after_llm_call_hook,
+    register_before_llm_call_hook,
+    unregister_after_llm_call_hook,
+    unregister_before_llm_call_hook,
+)
+from crewai.hooks.tool_hooks import (
+    ToolCallHookContext,
+    clear_after_tool_call_hooks,
+    clear_all_tool_call_hooks,
+    clear_before_tool_call_hooks,
+    get_after_tool_call_hooks,
+    get_before_tool_call_hooks,
+    register_after_tool_call_hook,
+    register_before_tool_call_hook,
+    unregister_after_tool_call_hook,
+    unregister_before_tool_call_hook,
+)
+
+
+def clear_all_global_hooks() -> dict[str, tuple[int, int]]:
+    """Clear all global hooks across all hook types (LLM and Tool).
+
+    This is a convenience function that clears all registered hooks in one call.
+    Useful for testing, resetting state, or cleaning up between different
+    execution contexts.
+
+    Returns:
+        Dictionary with counts of cleared hooks:
+        {
+            "llm_hooks": (before_count, after_count),
+            "tool_hooks": (before_count, after_count),
+            "total": (total_before_count, total_after_count)
+        }
+
+    Example:
+        >>> # Register various hooks
+        >>> register_before_llm_call_hook(llm_hook1)
+        >>> register_after_llm_call_hook(llm_hook2)
+        >>> register_before_tool_call_hook(tool_hook1)
+        >>> register_after_tool_call_hook(tool_hook2)
+        >>>
+        >>> # Clear all hooks at once
+        >>> result = clear_all_global_hooks()
+        >>> print(result)
+        {
+            'llm_hooks': (1, 1),
+            'tool_hooks': (1, 1),
+            'total': (2, 2)
+        }
+    """
+    llm_counts = clear_all_llm_call_hooks()
+    tool_counts = clear_all_tool_call_hooks()
+
+    return {
+        "llm_hooks": llm_counts,
+        "tool_hooks": tool_counts,
+        "total": (llm_counts[0] + tool_counts[0], llm_counts[1] + tool_counts[1]),
+    }
+
+
+__all__ = [
+    # Context classes
+    "LLMCallHookContext",
+    "ToolCallHookContext",
+    # Decorators
+    "after_llm_call",
+    "after_tool_call",
+    "before_llm_call",
+    "before_tool_call",
+    "clear_after_llm_call_hooks",
+    "clear_after_tool_call_hooks",
+    "clear_all_global_hooks",
+    "clear_all_llm_call_hooks",
+    "clear_all_tool_call_hooks",
+    # Clear hooks
+    "clear_before_llm_call_hooks",
+    "clear_before_tool_call_hooks",
+    "get_after_llm_call_hooks",
+    "get_after_tool_call_hooks",
+    # Get hooks
+    "get_before_llm_call_hooks",
+    "get_before_tool_call_hooks",
+    "register_after_llm_call_hook",
+    "register_after_tool_call_hook",
+    # LLM Hook registration
+    "register_before_llm_call_hook",
+    # Tool Hook registration
+    "register_before_tool_call_hook",
+    "unregister_after_llm_call_hook",
+    "unregister_after_tool_call_hook",
+    "unregister_before_llm_call_hook",
+    "unregister_before_tool_call_hook",
+]

lib/crewai/src/crewai/hooks/decorators.py

@@ -0,0 +1,300 @@
+from __future__ import annotations
+
+from collections.abc import Callable
+from functools import wraps
+import inspect
+from typing import TYPE_CHECKING, Any, TypeVar, overload
+
+
+if TYPE_CHECKING:
+    from crewai.hooks.llm_hooks import LLMCallHookContext
+    from crewai.hooks.tool_hooks import ToolCallHookContext
+
+F = TypeVar("F", bound=Callable[..., Any])
+
+
+def _create_hook_decorator(
+    hook_type: str,
+    register_function: Callable[..., Any],
+    marker_attribute: str,
+) -> Callable[..., Any]:
+    """Create a hook decorator with filtering support.
+
+    This factory function eliminates code duplication across the four hook decorators.
+
+    Args:
+        hook_type: Type of hook ("llm" or "tool")
+        register_function: Function to call for registration (e.g., register_before_llm_call_hook)
+        marker_attribute: Attribute name to mark functions (e.g., "is_before_llm_call_hook")
+
+    Returns:
+        A decorator function that supports filters and auto-registration
+    """
+
+    def decorator_factory(
+        func: Callable[..., Any] | None = None,
+        *,
+        tools: list[str] | None = None,
+        agents: list[str] | None = None,
+    ) -> Callable[..., Any]:
+        def decorator(f: Callable[..., Any]) -> Callable[..., Any]:
+            setattr(f, marker_attribute, True)
+
+            sig = inspect.signature(f)
+            params = list(sig.parameters.keys())
+            is_method = len(params) >= 2 and params[0] == "self"
+
+            if tools:
+                f._filter_tools = tools  # type: ignore[attr-defined]
+            if agents:
+                f._filter_agents = agents  # type: ignore[attr-defined]
+
+            if tools or agents:
+
+                @wraps(f)
+                def filtered_hook(context: Any) -> Any:
+                    if tools and hasattr(context, "tool_name"):
+                        if context.tool_name not in tools:
+                            return None
+
+                    if agents and hasattr(context, "agent"):
+                        if context.agent and context.agent.role not in agents:
+                            return None
+
+                    return f(context)
+
+                if not is_method:
+                    register_function(filtered_hook)
+
+                return f
+
+            if not is_method:
+                register_function(f)
+
+            return f
+
+        if func is None:
+            return decorator
+        return decorator(func)
+
+    return decorator_factory
+
+
+@overload
+def before_llm_call(
+    func: Callable[[LLMCallHookContext], None],
+) -> Callable[[LLMCallHookContext], None]: ...
+
+
+@overload
+def before_llm_call(
+    *,
+    agents: list[str] | None = None,
+) -> Callable[
+    [Callable[[LLMCallHookContext], None]], Callable[[LLMCallHookContext], None]
+]: ...
+
+
+def before_llm_call(
+    func: Callable[[LLMCallHookContext], None] | None = None,
+    *,
+    agents: list[str] | None = None,
+) -> (
+    Callable[[LLMCallHookContext], None]
+    | Callable[
+        [Callable[[LLMCallHookContext], None]], Callable[[LLMCallHookContext], None]
+    ]
+):
+    """Decorator to register a function as a before_llm_call hook.
+
+    Example:
+        Simple usage::
+
+            @before_llm_call
+            def log_calls(context):
+                print(f"LLM call by {context.agent.role}")
+
+        With agent filter::
+
+            @before_llm_call(agents=["Researcher", "Analyst"])
+            def log_specific_agents(context):
+                print(f"Filtered LLM call: {context.agent.role}")
+    """
+    from crewai.hooks.llm_hooks import register_before_llm_call_hook
+
+    return _create_hook_decorator(  # type: ignore[return-value]
+        hook_type="llm",
+        register_function=register_before_llm_call_hook,
+        marker_attribute="is_before_llm_call_hook",
+    )(func=func, agents=agents)
+
+
+@overload
+def after_llm_call(
+    func: Callable[[LLMCallHookContext], str | None],
+) -> Callable[[LLMCallHookContext], str | None]: ...
+
+
+@overload
+def after_llm_call(
+    *,
+    agents: list[str] | None = None,
+) -> Callable[
+    [Callable[[LLMCallHookContext], str | None]],
+    Callable[[LLMCallHookContext], str | None],
+]: ...
+
+
+def after_llm_call(
+    func: Callable[[LLMCallHookContext], str | None] | None = None,
+    *,
+    agents: list[str] | None = None,
+) -> (
+    Callable[[LLMCallHookContext], str | None]
+    | Callable[
+        [Callable[[LLMCallHookContext], str | None]],
+        Callable[[LLMCallHookContext], str | None],
+    ]
+):
+    """Decorator to register a function as an after_llm_call hook.
+
+    Example:
+        Simple usage::
+
+            @after_llm_call
+            def sanitize(context):
+                if "SECRET" in context.response:
+                    return context.response.replace("SECRET", "[REDACTED]")
+                return None
+
+        With agent filter::
+
+            @after_llm_call(agents=["Researcher"])
+            def log_researcher_responses(context):
+                print(f"Response length: {len(context.response)}")
+                return None
+    """
+    from crewai.hooks.llm_hooks import register_after_llm_call_hook
+
+    return _create_hook_decorator(  # type: ignore[return-value]
+        hook_type="llm",
+        register_function=register_after_llm_call_hook,
+        marker_attribute="is_after_llm_call_hook",
+    )(func=func, agents=agents)
+
+
+@overload
+def before_tool_call(
+    func: Callable[[ToolCallHookContext], bool | None],
+) -> Callable[[ToolCallHookContext], bool | None]: ...
+
+
+@overload
+def before_tool_call(
+    *,
+    tools: list[str] | None = None,
+    agents: list[str] | None = None,
+) -> Callable[
+    [Callable[[ToolCallHookContext], bool | None]],
+    Callable[[ToolCallHookContext], bool | None],
+]: ...
+
+
+def before_tool_call(
+    func: Callable[[ToolCallHookContext], bool | None] | None = None,
+    *,
+    tools: list[str] | None = None,
+    agents: list[str] | None = None,
+) -> (
+    Callable[[ToolCallHookContext], bool | None]
+    | Callable[
+        [Callable[[ToolCallHookContext], bool | None]],
+        Callable[[ToolCallHookContext], bool | None],
+    ]
+):
+    """Decorator to register a function as a before_tool_call hook.
+
+    Example:
+        Simple usage::
+
+            @before_tool_call
+            def log_all_tools(context):
+                print(f"Tool: {context.tool_name}")
+                return None
+
+        With tool filter::
+
+            @before_tool_call(tools=["delete_file", "execute_code"])
+            def approve_dangerous(context):
+                response = context.request_human_input(prompt="Approve?")
+                return None if response == "yes" else False
+
+        With combined filters::
+
+            @before_tool_call(tools=["write_file"], agents=["Developer"])
+            def approve_dev_writes(context):
+                return None  # Only for Developer writing files
+    """
+    from crewai.hooks.tool_hooks import register_before_tool_call_hook
+
+    return _create_hook_decorator(  # type: ignore[return-value]
+        hook_type="tool",
+        register_function=register_before_tool_call_hook,
+        marker_attribute="is_before_tool_call_hook",
+    )(func=func, tools=tools, agents=agents)
+
+
+@overload
+def after_tool_call(
+    func: Callable[[ToolCallHookContext], str | None],
+) -> Callable[[ToolCallHookContext], str | None]: ...
+
+
+@overload
+def after_tool_call(
+    *,
+    tools: list[str] | None = None,
+    agents: list[str] | None = None,
+) -> Callable[
+    [Callable[[ToolCallHookContext], str | None]],
+    Callable[[ToolCallHookContext], str | None],
+]: ...
+
+
+def after_tool_call(
+    func: Callable[[ToolCallHookContext], str | None] | None = None,
+    *,
+    tools: list[str] | None = None,
+    agents: list[str] | None = None,
+) -> (
+    Callable[[ToolCallHookContext], str | None]
+    | Callable[
+        [Callable[[ToolCallHookContext], str | None]],
+        Callable[[ToolCallHookContext], str | None],
+    ]
+):
+    """Decorator to register a function as an after_tool_call hook.
+
+    Example:
+        Simple usage::
+
+            @after_tool_call
+            def log_results(context):
+                print(f"Result: {len(context.tool_result)} chars")
+                return None
+
+        With tool filter::
+
+            @after_tool_call(tools=["web_search", "ExaSearchTool"])
+            def sanitize_search_results(context):
+                if "SECRET" in context.tool_result:
+                    return context.tool_result.replace("SECRET", "[REDACTED]")
+                return None
+    """
+    from crewai.hooks.tool_hooks import register_after_tool_call_hook
+
+    return _create_hook_decorator(  # type: ignore[return-value]
+        hook_type="tool",
+        register_function=register_after_tool_call_hook,
+        marker_attribute="is_after_tool_call_hook",
+    )(func=func, tools=tools, agents=agents)

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

@@ -0,0 +1,290 @@
+from __future__ import annotations
+
+from typing import TYPE_CHECKING
+
+from crewai.events.event_listener import event_listener
+from crewai.hooks.types import AfterLLMCallHookType, BeforeLLMCallHookType
+from crewai.utilities.printer import Printer
+
+
+if TYPE_CHECKING:
+    from crewai.agents.crew_agent_executor import CrewAgentExecutor
+
+
+class LLMCallHookContext:
+    """Context object passed to LLM call hooks with full executor access.
+
+    Provides hooks with complete access to the executor state, allowing
+    modification of messages, responses, and executor attributes.
+
+    Attributes:
+        executor: Full reference to the CrewAgentExecutor instance
+        messages: Direct reference to executor.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.
+            IMPORTANT: Modify messages in-place (e.g., append, extend, remove items).
+            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
+        llm: Reference to the LLM instance
+        iterations: Current iteration count
+        response: LLM response string (only set for after_llm_call hooks).
+            Can be modified by returning a new string from after_llm_call hook.
+    """
+
+    def __init__(
+        self,
+        executor: CrewAgentExecutor,
+        response: str | None = None,
+    ) -> None:
+        """Initialize hook context with executor reference.
+
+        Args:
+            executor: The CrewAgentExecutor instance
+            response: Optional response string (for after_llm_call hooks)
+        """
+        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
+        self.response = response
+
+    def request_human_input(
+        self,
+        prompt: str,
+        default_message: str = "Press Enter to continue, or provide feedback:",
+    ) -> str:
+        """Request human input during LLM hook execution.
+
+        This method pauses live console updates, displays a prompt to the user,
+        waits for their input, and then resumes live updates. This is useful for
+        approval gates, debugging, or getting human feedback during execution.
+
+        Args:
+            prompt: Custom message to display to the user
+            default_message: Message shown after the prompt
+
+        Returns:
+            User's input as a string (empty string if just Enter pressed)
+
+        Example:
+            >>> def approval_hook(context: LLMCallHookContext) -> None:
+            ...     if context.iterations > 5:
+            ...         response = context.request_human_input(
+            ...             prompt="Allow this LLM call?",
+            ...             default_message="Type 'no' to skip, or press Enter:",
+            ...         )
+            ...         if response.lower() == "no":
+            ...             print("LLM call skipped by user")
+        """
+
+        printer = Printer()
+        event_listener.formatter.pause_live_updates()
+
+        try:
+            printer.print(content=f"\n{prompt}", color="bold_yellow")
+            printer.print(content=default_message, color="cyan")
+            response = input().strip()
+
+            if response:
+                printer.print(content="\nProcessing your input...", color="cyan")
+
+            return response
+        finally:
+            event_listener.formatter.resume_live_updates()
+
+
+_before_llm_call_hooks: list[BeforeLLMCallHookType] = []
+_after_llm_call_hooks: list[AfterLLMCallHookType] = []
+
+
+def register_before_llm_call_hook(
+    hook: BeforeLLMCallHookType,
+) -> None:
+    """Register a global before_llm_call hook.
+
+    Global hooks are added to all executors automatically.
+    This is a convenience function for registering hooks that should
+    apply to all LLM calls across all executors.
+
+    Args:
+        hook: Function that receives LLMCallHookContext and can:
+            - Modify context.messages directly (in-place)
+            - Return False to block LLM execution
+            - Return True or None to allow execution
+            IMPORTANT: Modify messages in-place (append, extend, remove items).
+            Do NOT replace the list (context.messages = []), as this will break execution.
+
+    Example:
+        >>> def log_llm_calls(context: LLMCallHookContext) -> None:
+        ...     print(f"LLM call by {context.agent.role}")
+        ...     print(f"Messages: {len(context.messages)}")
+        ...     return None  # Allow execution
+        >>>
+        >>> register_before_llm_call_hook(log_llm_calls)
+        >>>
+        >>> def block_excessive_iterations(context: LLMCallHookContext) -> bool | None:
+        ...     if context.iterations > 10:
+        ...         print("Blocked: Too many iterations")
+        ...         return False  # Block execution
+        ...     return None  # Allow execution
+        >>>
+        >>> register_before_llm_call_hook(block_excessive_iterations)
+    """
+    _before_llm_call_hooks.append(hook)
+
+
+def register_after_llm_call_hook(
+    hook: AfterLLMCallHookType,
+) -> None:
+    """Register a global after_llm_call hook.
+
+    Global hooks are added to all executors automatically.
+    This is a convenience function for registering hooks that should
+    apply to all LLM calls across all executors.
+
+    Args:
+        hook: Function that receives LLMCallHookContext and can modify:
+            - The response: Return modified response string or None to keep original
+            - The messages: Modify context.messages directly (mutable reference)
+            Both modifications are supported and can be used together.
+            IMPORTANT: Modify messages in-place (append, extend, remove items).
+            Do NOT replace the list (context.messages = []), as this will break execution.
+
+    Example:
+        >>> def sanitize_response(context: LLMCallHookContext) -> str | None:
+        ...     if context.response and "SECRET" in context.response:
+        ...         return context.response.replace("SECRET", "[REDACTED]")
+        ...     return None
+        >>>
+        >>> register_after_llm_call_hook(sanitize_response)
+    """
+    _after_llm_call_hooks.append(hook)
+
+
+def get_before_llm_call_hooks() -> list[BeforeLLMCallHookType]:
+    """Get all registered global before_llm_call hooks.
+
+    Returns:
+        List of registered before hooks
+    """
+    return _before_llm_call_hooks.copy()
+
+
+def get_after_llm_call_hooks() -> list[AfterLLMCallHookType]:
+    """Get all registered global after_llm_call hooks.
+
+    Returns:
+        List of registered after hooks
+    """
+    return _after_llm_call_hooks.copy()
+
+
+def unregister_before_llm_call_hook(
+    hook: BeforeLLMCallHookType,
+) -> bool:
+    """Unregister a specific global before_llm_call hook.
+
+    Args:
+        hook: The hook function to remove
+
+    Returns:
+        True if the hook was found and removed, False otherwise
+
+    Example:
+        >>> def my_hook(context: LLMCallHookContext) -> None:
+        ...     print("Before LLM call")
+        >>>
+        >>> register_before_llm_call_hook(my_hook)
+        >>> unregister_before_llm_call_hook(my_hook)
+        True
+    """
+    try:
+        _before_llm_call_hooks.remove(hook)
+        return True
+    except ValueError:
+        return False
+
+
+def unregister_after_llm_call_hook(
+    hook: AfterLLMCallHookType,
+) -> bool:
+    """Unregister a specific global after_llm_call hook.
+
+    Args:
+        hook: The hook function to remove
+
+    Returns:
+        True if the hook was found and removed, False otherwise
+
+    Example:
+        >>> def my_hook(context: LLMCallHookContext) -> str | None:
+        ...     return None
+        >>>
+        >>> register_after_llm_call_hook(my_hook)
+        >>> unregister_after_llm_call_hook(my_hook)
+        True
+    """
+    try:
+        _after_llm_call_hooks.remove(hook)
+        return True
+    except ValueError:
+        return False
+
+
+def clear_before_llm_call_hooks() -> int:
+    """Clear all registered global before_llm_call hooks.
+
+    Returns:
+        Number of hooks that were cleared
+
+    Example:
+        >>> register_before_llm_call_hook(hook1)
+        >>> register_before_llm_call_hook(hook2)
+        >>> clear_before_llm_call_hooks()
+        2
+    """
+    count = len(_before_llm_call_hooks)
+    _before_llm_call_hooks.clear()
+    return count
+
+
+def clear_after_llm_call_hooks() -> int:
+    """Clear all registered global after_llm_call hooks.
+
+    Returns:
+        Number of hooks that were cleared
+
+    Example:
+        >>> register_after_llm_call_hook(hook1)
+        >>> register_after_llm_call_hook(hook2)
+        >>> clear_after_llm_call_hooks()
+        2
+    """
+    count = len(_after_llm_call_hooks)
+    _after_llm_call_hooks.clear()
+    return count
+
+
+def clear_all_llm_call_hooks() -> tuple[int, int]:
+    """Clear all registered global LLM call hooks (both before and after).
+
+    Returns:
+        Tuple of (before_hooks_cleared, after_hooks_cleared)
+
+    Example:
+        >>> register_before_llm_call_hook(before_hook)
+        >>> register_after_llm_call_hook(after_hook)
+        >>> clear_all_llm_call_hooks()
+        (1, 1)
+    """
+    before_count = clear_before_llm_call_hooks()
+    after_count = clear_after_llm_call_hooks()
+    return (before_count, after_count)

lib/crewai/src/crewai/hooks/tool_hooks.py

@@ -0,0 +1,305 @@
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, Any
+
+from crewai.events.event_listener import event_listener
+from crewai.hooks.types import AfterToolCallHookType, BeforeToolCallHookType
+from crewai.utilities.printer import Printer
+
+
+if TYPE_CHECKING:
+    from crewai.agent import Agent
+    from crewai.agents.agent_builder.base_agent import BaseAgent
+    from crewai.crew import Crew
+    from crewai.task import Task
+    from crewai.tools.structured_tool import CrewStructuredTool
+
+
+class ToolCallHookContext:
+    """Context object passed to tool call hooks.
+
+    Provides hooks with access to the tool being called, its input,
+    the agent/task/crew context, and the result (for after hooks).
+
+    Attributes:
+        tool_name: Name of the tool being called
+        tool_input: Tool input parameters (mutable dict).
+            Can be modified in-place by before_tool_call hooks.
+            IMPORTANT: Modify in-place (e.g., context.tool_input['key'] = value).
+            Do NOT replace the dict (e.g., context.tool_input = {}), as this
+            will not affect the actual tool execution.
+        tool: Reference to the CrewStructuredTool instance
+        agent: Agent executing the tool (may be None)
+        task: Current task being executed (may be None)
+        crew: Crew instance (may be None)
+        tool_result: Tool execution result (only set for after_tool_call hooks).
+            Can be modified by returning a new string from after_tool_call hook.
+    """
+
+    def __init__(
+        self,
+        tool_name: str,
+        tool_input: dict[str, Any],
+        tool: CrewStructuredTool,
+        agent: Agent | BaseAgent | None = None,
+        task: Task | None = None,
+        crew: Crew | None = None,
+        tool_result: str | None = None,
+    ) -> None:
+        """Initialize tool call hook context.
+
+        Args:
+            tool_name: Name of the tool being called
+            tool_input: Tool input parameters (mutable)
+            tool: Tool instance reference
+            agent: Optional agent executing the tool
+            task: Optional current task
+            crew: Optional crew instance
+            tool_result: Optional tool result (for after hooks)
+        """
+        self.tool_name = tool_name
+        self.tool_input = tool_input
+        self.tool = tool
+        self.agent = agent
+        self.task = task
+        self.crew = crew
+        self.tool_result = tool_result
+
+    def request_human_input(
+        self,
+        prompt: str,
+        default_message: str = "Press Enter to continue, or provide feedback:",
+    ) -> str:
+        """Request human input during tool hook execution.
+
+        This method pauses live console updates, displays a prompt to the user,
+        waits for their input, and then resumes live updates. This is useful for
+        approval gates, reviewing tool results, or getting human feedback during execution.
+
+        Args:
+            prompt: Custom message to display to the user
+            default_message: Message shown after the prompt
+
+        Returns:
+            User's input as a string (empty string if just Enter pressed)
+
+        Example:
+            >>> def approval_hook(context: ToolCallHookContext) -> bool | None:
+            ...     if context.tool_name == "delete_file":
+            ...         response = context.request_human_input(
+            ...             prompt="Allow file deletion?",
+            ...             default_message="Type 'approve' to continue:",
+            ...         )
+            ...         if response.lower() != "approve":
+            ...             return False  # Block execution
+            ...     return None  # Allow execution
+        """
+
+        printer = Printer()
+        event_listener.formatter.pause_live_updates()
+
+        try:
+            printer.print(content=f"\n{prompt}", color="bold_yellow")
+            printer.print(content=default_message, color="cyan")
+            response = input().strip()
+
+            if response:
+                printer.print(content="\nProcessing your input...", color="cyan")
+
+            return response
+        finally:
+            event_listener.formatter.resume_live_updates()
+
+
+# Global hook registries
+_before_tool_call_hooks: list[BeforeToolCallHookType] = []
+_after_tool_call_hooks: list[AfterToolCallHookType] = []
+
+
+def register_before_tool_call_hook(
+    hook: BeforeToolCallHookType,
+) -> None:
+    """Register a global before_tool_call hook.
+
+    Global hooks are added to all tool executions automatically.
+    This is a convenience function for registering hooks that should
+    apply to all tool calls across all agents and crews.
+
+    Args:
+        hook: Function that receives ToolCallHookContext and can:
+            - Modify tool_input in-place
+            - Return False to block tool execution
+            - Return True or None to allow execution
+            IMPORTANT: Modify tool_input in-place (e.g., context.tool_input['key'] = value).
+            Do NOT replace the dict (context.tool_input = {}), as this will not affect
+            the actual tool execution.
+
+    Example:
+        >>> def log_tool_usage(context: ToolCallHookContext) -> None:
+        ...     print(f"Executing tool: {context.tool_name}")
+        ...     print(f"Input: {context.tool_input}")
+        ...     return None  # Allow execution
+        >>>
+        >>> register_before_tool_call_hook(log_tool_usage)
+
+        >>> def block_dangerous_tools(context: ToolCallHookContext) -> bool | None:
+        ...     if context.tool_name == "delete_database":
+        ...         print("Blocked dangerous tool execution!")
+        ...         return False  # Block execution
+        ...     return None  # Allow execution
+        >>>
+        >>> register_before_tool_call_hook(block_dangerous_tools)
+    """
+    _before_tool_call_hooks.append(hook)
+
+
+def register_after_tool_call_hook(
+    hook: AfterToolCallHookType,
+) -> None:
+    """Register a global after_tool_call hook.
+
+    Global hooks are added to all tool executions automatically.
+    This is a convenience function for registering hooks that should
+    apply to all tool calls across all agents and crews.
+
+    Args:
+        hook: Function that receives ToolCallHookContext and can modify
+            the tool result. Return modified result string or None to keep
+            the original result. The tool_result is available in context.tool_result.
+
+    Example:
+        >>> def sanitize_output(context: ToolCallHookContext) -> str | None:
+        ...     if context.tool_result and "SECRET_KEY" in context.tool_result:
+        ...         return context.tool_result.replace("SECRET_KEY=...", "[REDACTED]")
+        ...     return None  # Keep original result
+        >>>
+        >>> register_after_tool_call_hook(sanitize_output)
+
+        >>> def log_tool_results(context: ToolCallHookContext) -> None:
+        ...     print(f"Tool {context.tool_name} returned: {context.tool_result[:100]}")
+        ...     return None  # Keep original result
+        >>>
+        >>> register_after_tool_call_hook(log_tool_results)
+    """
+    _after_tool_call_hooks.append(hook)
+
+
+def get_before_tool_call_hooks() -> list[BeforeToolCallHookType]:
+    """Get all registered global before_tool_call hooks.
+
+    Returns:
+        List of registered before hooks
+    """
+    return _before_tool_call_hooks.copy()
+
+
+def get_after_tool_call_hooks() -> list[AfterToolCallHookType]:
+    """Get all registered global after_tool_call hooks.
+
+    Returns:
+        List of registered after hooks
+    """
+    return _after_tool_call_hooks.copy()
+
+
+def unregister_before_tool_call_hook(
+    hook: BeforeToolCallHookType,
+) -> bool:
+    """Unregister a specific global before_tool_call hook.
+
+    Args:
+        hook: The hook function to remove
+
+    Returns:
+        True if the hook was found and removed, False otherwise
+
+    Example:
+        >>> def my_hook(context: ToolCallHookContext) -> None:
+        ...     print("Before tool call")
+        >>>
+        >>> register_before_tool_call_hook(my_hook)
+        >>> unregister_before_tool_call_hook(my_hook)
+        True
+    """
+    try:
+        _before_tool_call_hooks.remove(hook)
+        return True
+    except ValueError:
+        return False
+
+
+def unregister_after_tool_call_hook(
+    hook: AfterToolCallHookType,
+) -> bool:
+    """Unregister a specific global after_tool_call hook.
+
+    Args:
+        hook: The hook function to remove
+
+    Returns:
+        True if the hook was found and removed, False otherwise
+
+    Example:
+        >>> def my_hook(context: ToolCallHookContext) -> str | None:
+        ...     return None
+        >>>
+        >>> register_after_tool_call_hook(my_hook)
+        >>> unregister_after_tool_call_hook(my_hook)
+        True
+    """
+    try:
+        _after_tool_call_hooks.remove(hook)
+        return True
+    except ValueError:
+        return False
+
+
+def clear_before_tool_call_hooks() -> int:
+    """Clear all registered global before_tool_call hooks.
+
+    Returns:
+        Number of hooks that were cleared
+
+    Example:
+        >>> register_before_tool_call_hook(hook1)
+        >>> register_before_tool_call_hook(hook2)
+        >>> clear_before_tool_call_hooks()
+        2
+    """
+    count = len(_before_tool_call_hooks)
+    _before_tool_call_hooks.clear()
+    return count
+
+
+def clear_after_tool_call_hooks() -> int:
+    """Clear all registered global after_tool_call hooks.
+
+    Returns:
+        Number of hooks that were cleared
+
+    Example:
+        >>> register_after_tool_call_hook(hook1)
+        >>> register_after_tool_call_hook(hook2)
+        >>> clear_after_tool_call_hooks()
+        2
+    """
+    count = len(_after_tool_call_hooks)
+    _after_tool_call_hooks.clear()
+    return count
+
+
+def clear_all_tool_call_hooks() -> tuple[int, int]:
+    """Clear all registered global tool call hooks (both before and after).
+
+    Returns:
+        Tuple of (before_hooks_cleared, after_hooks_cleared)
+
+    Example:
+        >>> register_before_tool_call_hook(before_hook)
+        >>> register_after_tool_call_hook(after_hook)
+        >>> clear_all_tool_call_hooks()
+        (1, 1)
+    """
+    before_count = clear_before_tool_call_hooks()
+    after_count = clear_after_tool_call_hooks()
+    return (before_count, after_count)

lib/crewai/src/crewai/hooks/types.py

@@ -0,0 +1,137 @@
+from __future__ import annotations
+
+from collections.abc import Callable
+from typing import TYPE_CHECKING, Generic, Protocol, TypeVar, runtime_checkable
+
+
+if TYPE_CHECKING:
+    from crewai.hooks.llm_hooks import LLMCallHookContext
+    from crewai.hooks.tool_hooks import ToolCallHookContext
+
+
+ContextT = TypeVar("ContextT", contravariant=True)
+ReturnT = TypeVar("ReturnT", covariant=True)
+
+
+@runtime_checkable
+class Hook(Protocol, Generic[ContextT, ReturnT]):
+    """Generic protocol for hook functions.
+
+    This protocol defines the common interface for all hook types in CrewAI.
+    Hooks receive a context object and optionally return a modified result.
+
+    Type Parameters:
+        ContextT: The context type (LLMCallHookContext or ToolCallHookContext)
+        ReturnT: The return type (None, str | None, or bool | None)
+
+    Example:
+        >>> # Before LLM call hook: receives LLMCallHookContext, returns None
+        >>> hook: Hook[LLMCallHookContext, None] = lambda ctx: print(ctx.iterations)
+        >>>
+        >>> # After LLM call hook: receives LLMCallHookContext, returns str | None
+        >>> hook: Hook[LLMCallHookContext, str | None] = lambda ctx: ctx.response
+    """
+
+    def __call__(self, context: ContextT) -> ReturnT:
+        """Execute the hook with the given context.
+
+        Args:
+            context: Context object with relevant execution state
+
+        Returns:
+            Hook-specific return value (None, str | None, or bool | None)
+        """
+        ...
+
+
+class BeforeLLMCallHook(Hook["LLMCallHookContext", bool | None], Protocol):
+    """Protocol for before_llm_call hooks.
+
+    These hooks are called before an LLM is invoked and can modify the messages
+    that will be sent to the LLM or block the execution entirely.
+    """
+
+    def __call__(self, context: LLMCallHookContext) -> bool | None:
+        """Execute the before LLM call hook.
+
+        Args:
+            context: Context object with executor, messages, agent, task, etc.
+                Messages can be modified in-place.
+
+        Returns:
+            False to block LLM execution, True or None to allow execution
+        """
+        ...
+
+
+class AfterLLMCallHook(Hook["LLMCallHookContext", str | None], Protocol):
+    """Protocol for after_llm_call hooks.
+
+    These hooks are called after an LLM returns a response and can modify
+    the response or the message history.
+    """
+
+    def __call__(self, context: LLMCallHookContext) -> str | None:
+        """Execute the after LLM call hook.
+
+        Args:
+            context: Context object with executor, messages, agent, task, response, etc.
+                Messages can be modified in-place. Response is available in context.response.
+
+        Returns:
+            Modified response string, or None to keep the original response
+        """
+        ...
+
+
+class BeforeToolCallHook(Hook["ToolCallHookContext", bool | None], Protocol):
+    """Protocol for before_tool_call hooks.
+
+    These hooks are called before a tool is executed and can modify the tool
+    input or block the execution entirely.
+    """
+
+    def __call__(self, context: ToolCallHookContext) -> bool | None:
+        """Execute the before tool call hook.
+
+        Args:
+            context: Context object with tool_name, tool_input, tool, agent, task, etc.
+                Tool input can be modified in-place.
+
+        Returns:
+            False to block tool execution, True or None to allow execution
+        """
+        ...
+
+
+class AfterToolCallHook(Hook["ToolCallHookContext", str | None], Protocol):
+    """Protocol for after_tool_call hooks.
+
+    These hooks are called after a tool executes and can modify the result.
+    """
+
+    def __call__(self, context: ToolCallHookContext) -> str | None:
+        """Execute the after tool call hook.
+
+        Args:
+            context: Context object with tool_name, tool_input, tool_result, etc.
+                Tool result is available in context.tool_result.
+
+        Returns:
+            Modified tool result string, or None to keep the original result
+        """
+        ...
+
+
+# - All before hooks: bool | None (False = block execution, True/None = allow)
+# - All after hooks: str | None (str = modified result, None = keep original)
+BeforeLLMCallHookType = Hook["LLMCallHookContext", bool | None]
+AfterLLMCallHookType = Hook["LLMCallHookContext", str | None]
+BeforeToolCallHookType = Hook["ToolCallHookContext", bool | None]
+AfterToolCallHookType = Hook["ToolCallHookContext", str | None]
+
+# Alternative Callable-based type aliases for compatibility
+BeforeLLMCallHookCallable = Callable[["LLMCallHookContext"], bool | None]
+AfterLLMCallHookCallable = Callable[["LLMCallHookContext"], str | None]
+BeforeToolCallHookCallable = Callable[["ToolCallHookContext"], bool | None]
+AfterToolCallHookCallable = Callable[["ToolCallHookContext"], str | None]

lib/crewai/src/crewai/hooks/wrappers.py

@@ -0,0 +1,157 @@
+from __future__ import annotations
+
+from collections.abc import Callable
+from typing import TYPE_CHECKING, Any, TypeVar
+
+
+if TYPE_CHECKING:
+    from crewai.hooks.llm_hooks import LLMCallHookContext
+    from crewai.hooks.tool_hooks import ToolCallHookContext
+
+P = TypeVar("P")
+R = TypeVar("R")
+
+
+def _copy_method_metadata(wrapper: Any, original: Callable[..., Any]) -> None:
+    """Copy metadata from original function to wrapper.
+
+    Args:
+        wrapper: The wrapper object to copy metadata to
+        original: The original function to copy from
+    """
+    wrapper.__name__ = original.__name__
+    wrapper.__doc__ = original.__doc__
+    wrapper.__module__ = original.__module__
+    wrapper.__qualname__ = original.__qualname__
+    wrapper.__annotations__ = original.__annotations__
+
+
+class BeforeLLMCallHookMethod:
+    """Wrapper for methods marked as before_llm_call hooks within @CrewBase classes.
+
+    This wrapper marks a method so it can be detected and registered as a
+    crew-scoped hook during crew initialization.
+    """
+
+    is_before_llm_call_hook: bool = True
+
+    def __init__(
+        self,
+        meth: Callable[[Any, LLMCallHookContext], None],
+        agents: list[str] | None = None,
+    ) -> None:
+        """Initialize the hook method wrapper.
+
+        Args:
+            meth: The method to wrap
+            agents: Optional list of agent roles to filter
+        """
+        self._meth = meth
+        self.agents = agents
+        _copy_method_metadata(self, meth)
+
+    def __call__(self, *args: Any, **kwargs: Any) -> None:
+        """Call the wrapped method.
+
+        Args:
+            *args: Positional arguments
+            **kwargs: Keyword arguments
+        """
+        return self._meth(*args, **kwargs)
+
+    def __get__(self, obj: Any, objtype: type[Any] | None = None) -> Any:
+        """Support instance methods by implementing descriptor protocol.
+
+        Args:
+            obj: The instance that the method is accessed through
+            objtype: The type of the instance
+
+        Returns:
+            Self when accessed through class, bound method when accessed through instance
+        """
+        if obj is None:
+            return self
+        # Return bound method
+        return lambda context: self._meth(obj, context)
+
+
+class AfterLLMCallHookMethod:
+    """Wrapper for methods marked as after_llm_call hooks within @CrewBase classes."""
+
+    is_after_llm_call_hook: bool = True
+
+    def __init__(
+        self,
+        meth: Callable[[Any, LLMCallHookContext], str | None],
+        agents: list[str] | None = None,
+    ) -> None:
+        """Initialize the hook method wrapper."""
+        self._meth = meth
+        self.agents = agents
+        _copy_method_metadata(self, meth)
+
+    def __call__(self, *args: Any, **kwargs: Any) -> str | None:
+        """Call the wrapped method."""
+        return self._meth(*args, **kwargs)
+
+    def __get__(self, obj: Any, objtype: type[Any] | None = None) -> Any:
+        """Support instance methods."""
+        if obj is None:
+            return self
+        return lambda context: self._meth(obj, context)
+
+
+class BeforeToolCallHookMethod:
+    """Wrapper for methods marked as before_tool_call hooks within @CrewBase classes."""
+
+    is_before_tool_call_hook: bool = True
+
+    def __init__(
+        self,
+        meth: Callable[[Any, ToolCallHookContext], bool | None],
+        tools: list[str] | None = None,
+        agents: list[str] | None = None,
+    ) -> None:
+        """Initialize the hook method wrapper."""
+        self._meth = meth
+        self.tools = tools
+        self.agents = agents
+        _copy_method_metadata(self, meth)
+
+    def __call__(self, *args: Any, **kwargs: Any) -> bool | None:
+        """Call the wrapped method."""
+        return self._meth(*args, **kwargs)
+
+    def __get__(self, obj: Any, objtype: type[Any] | None = None) -> Any:
+        """Support instance methods."""
+        if obj is None:
+            return self
+        return lambda context: self._meth(obj, context)
+
+
+class AfterToolCallHookMethod:
+    """Wrapper for methods marked as after_tool_call hooks within @CrewBase classes."""
+
+    is_after_tool_call_hook: bool = True
+
+    def __init__(
+        self,
+        meth: Callable[[Any, ToolCallHookContext], str | None],
+        tools: list[str] | None = None,
+        agents: list[str] | None = None,
+    ) -> None:
+        """Initialize the hook method wrapper."""
+        self._meth = meth
+        self.tools = tools
+        self.agents = agents
+        _copy_method_metadata(self, meth)
+
+    def __call__(self, *args: Any, **kwargs: Any) -> str | None:
+        """Call the wrapped method."""
+        return self._meth(*args, **kwargs)
+
+    def __get__(self, obj: Any, objtype: type[Any] | None = None) -> Any:
+        """Support instance methods."""
+        if obj is None:
+            return self
+        return lambda context: self._meth(obj, context)

lib/crewai/src/crewai/lite_agent.py

@@ -39,8 +39,8 @@ from crewai.events.types.agent_events import (
 from crewai.events.types.logging_events import AgentLogsExecutionEvent
 from crewai.flow.flow_trackable import FlowTrackable
 from crewai.lite_agent_output import LiteAgentOutput
-from crewai.llm import LLM
-from crewai.llms.base_llm import BaseLLM
+from crewai.llm.base_llm import BaseLLM
+from crewai.llm.core import LLM
 from crewai.tools.base_tool import BaseTool
 from crewai.tools.structured_tool import CrewStructuredTool
 from crewai.utilities.agent_utils import (
@@ -358,6 +358,7 @@ class LiteAgent(FlowTrackable, BaseModel):
             pydantic=formatted_result,
             agent_role=self.role,
             usage_metrics=usage_metrics.model_dump() if usage_metrics else None,
+            messages=self._messages,
         )
 
         # Process guardrail if set
@@ -503,7 +504,7 @@ 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):
@@ -541,6 +542,7 @@ class LiteAgent(FlowTrackable, BaseModel):
                             agent_key=self.key,
                             agent_role=self.role,
                             agent=self.original_agent,
+                            crew=None,
                         )
                     except Exception as e:
                         raise e
@@ -551,7 +553,8 @@ class LiteAgent(FlowTrackable, BaseModel):
                         show_logs=self._show_logs,
                     )
 
-                self._append_message(formatted_answer.text, role="assistant")
+                if formatted_answer is not None:
+                    self._append_message(formatted_answer.text, role="assistant")
             except OutputParserError as e:  # noqa: PERF203
                 self._printer.print(
                     content="Failed to parse LLM output. Retrying...",

lib/crewai/src/crewai/lite_agent_output.py

@@ -6,6 +6,8 @@ from typing import Any
 
 from pydantic import BaseModel, Field
 
+from crewai.utilities.types import LLMMessage
+
 
 class LiteAgentOutput(BaseModel):
     """Class that represents the result of a LiteAgent execution."""
@@ -20,6 +22,7 @@ class LiteAgentOutput(BaseModel):
     usage_metrics: dict[str, Any] | None = Field(
         description="Token usage metrics for this execution", default=None
     )
+    messages: list[LLMMessage] = Field(description="Messages of the agent", default=[])
 
     def to_dict(self) -> dict[str, Any]:
         """Convert pydantic_output to a dictionary."""

lib/crewai/src/crewai/llm/__init__.py

@@ -0,0 +1,4 @@
+from crewai.llm.core import LLM
+
+
+__all__ = ["LLM"]

lib/crewai/src/crewai/llm/base_llm.py

@@ -0,0 +1,588 @@
+"""Base LLM abstract class for CrewAI.
+
+This module provides the abstract base class for all LLM implementations
+in CrewAI, including common functionality for native SDK implementations.
+"""
+
+from __future__ import annotations
+
+from abc import ABC, abstractmethod
+from datetime import datetime
+import json
+import logging
+import os
+import re
+from typing import TYPE_CHECKING, Any, Final
+
+from dotenv import load_dotenv
+import httpx
+from pydantic import BaseModel, Field, field_validator
+
+from crewai.events.event_bus import crewai_event_bus
+from crewai.events.types.llm_events import (
+    LLMCallCompletedEvent,
+    LLMCallFailedEvent,
+    LLMCallStartedEvent,
+    LLMCallType,
+    LLMStreamChunkEvent,
+)
+from crewai.events.types.tool_usage_events import (
+    ToolUsageErrorEvent,
+    ToolUsageFinishedEvent,
+    ToolUsageStartedEvent,
+)
+from crewai.llm.hooks.base import BaseInterceptor
+from crewai.llm.internal.meta import LLMMeta
+from crewai.types.usage_metrics import UsageMetrics
+
+
+if TYPE_CHECKING:
+    from crewai.agent.core import Agent
+    from crewai.task import Task
+    from crewai.tools.base_tool import BaseTool
+    from crewai.utilities.types import LLMMessage
+
+
+load_dotenv()
+
+DEFAULT_CONTEXT_WINDOW_SIZE: Final[int] = 4096
+DEFAULT_SUPPORTS_STOP_WORDS: Final[bool] = True
+_JSON_EXTRACTION_PATTERN: Final[re.Pattern[str]] = re.compile(r"\{.*}", re.DOTALL)
+
+
+class BaseLLM(BaseModel, ABC, metaclass=LLMMeta):
+    """Abstract base class for LLM implementations.
+
+    This class defines the interface that all LLM implementations must follow.
+    Users can extend this class to create custom LLM implementations that don't
+    rely on litellm's authentication mechanism.
+
+    Custom LLM implementations should handle error cases gracefully, including
+    timeouts, authentication failures, and malformed responses. They should also
+    implement proper validation for input parameters and provide clear error
+    messages when things go wrong.
+
+    Attributes:
+        model: The model identifier/name.
+        temperature: Optional temperature setting for response generation.
+        stop: A list of stop sequences that the LLM should use to stop generation.
+    """
+
+    # Core fields
+    model: str = Field(..., description="The model identifier/name")
+    temperature: float | None = Field(
+        None, description="Temperature setting for response generation"
+    )
+    api_key: str | None = Field(None, description="API key for authentication")
+    base_url: str | None = Field(None, description="Base URL for API requests")
+    provider: str = Field(
+        default="openai", description="Provider name (openai, anthropic, etc.)"
+    )
+    stop: list[str] = Field(
+        default_factory=list,
+        description="Stop sequences for generation",
+        alias="stop_sequences",
+    )
+
+    # Internal fields
+    is_litellm: bool = Field(
+        default=False, description="Whether this instance uses LiteLLM"
+    )
+    interceptor: BaseInterceptor[httpx.Request, httpx.Response] | None = Field(
+        default=None, description="HTTP request/response interceptor"
+    )
+    _token_usage: dict[str, int] = {
+        "total_tokens": 0,
+        "prompt_tokens": 0,
+        "completion_tokens": 0,
+        "successful_requests": 0,
+        "cached_prompt_tokens": 0,
+    }
+
+    @field_validator("api_key", mode="after")
+    @classmethod
+    def _validate_api_key(cls, value: str | None) -> str | None:
+        """Validate API key for authentication.
+
+        Args:
+            value: API key value or None
+
+        Returns:
+            API key from environment if not provided, or the original value
+        """
+        if value is None:
+            cls_name = cls.__name__
+            provider_prefix = cls_name.replace("Completion", "").upper()
+            env_var = f"{provider_prefix}_API_KEY"
+            value = os.getenv(env_var)
+        return value
+
+    @field_validator("stop", mode="before")
+    @classmethod
+    def _normalize_stop(cls, value: Any) -> list[str]:
+        """Normalize stop sequences to a list.
+
+        Args:
+            value: Stop sequences as string, list, or None
+
+        Returns:
+            Normalized list of stop sequences
+        """
+        if value is None:
+            return []
+        if isinstance(value, str):
+            return [value]
+        if isinstance(value, list):
+            return value
+        return []
+
+    @property
+    def additional_params(self) -> dict[str, Any]:
+        """Get additional parameters stored as extra fields.
+
+        Returns:
+            Dictionary of additional parameters
+        """
+        return self.__pydantic_extra__ or {}
+
+    @additional_params.setter
+    def additional_params(self, value: dict[str, Any]) -> None:
+        """Set additional parameters as extra fields.
+
+        Args:
+            value: Dictionary of additional parameters to set
+        """
+        if not isinstance(value, dict):
+            raise ValueError("additional_params must be a dictionary")
+        if self.__pydantic_extra__ is None:
+            self.__pydantic_extra__ = {}
+        self.__pydantic_extra__.update(value)
+
+    @abstractmethod
+    def call(
+        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.
+        """
+
+    def _convert_tools_for_interference(
+        self, tools: list[dict[str, BaseTool]]
+    ) -> list[dict[str, BaseTool]]:
+        """Convert tools to a format that can be used for interference.
+
+        Args:
+            tools: List of tools to convert.
+
+        Returns:
+            List of converted tools (default implementation returns as-is)
+        """
+        return tools
+
+    def supports_stop_words(self) -> bool:
+        """Check if the LLM supports stop words.
+
+        Returns:
+            True if the LLM supports stop words, False otherwise.
+        """
+        return DEFAULT_SUPPORTS_STOP_WORDS
+
+    def _supports_stop_words_implementation(self) -> bool:
+        """Check if stop words are configured for this LLM instance.
+
+        Native providers can override supports_stop_words() to return this value
+        to ensure consistent behavior based on whether stop words are actually configured.
+
+        Returns:
+            True if stop words are configured and can be applied
+        """
+        return bool(self.stop)
+
+    def _apply_stop_words(self, content: str) -> str:
+        """Apply stop words to truncate response content.
+
+        This method provides consistent stop word behavior across all native SDK providers.
+        Native providers should call this method to post-process their responses.
+
+        Args:
+            content: The raw response content from the LLM
+
+        Returns:
+            Content truncated at the first occurrence of any stop word
+
+        Example:
+            >>> llm = MyNativeLLM(stop=["Observation:", "Final Answer:"])
+            >>> response = (
+            ...     "I need to search.\\n\\nAction: search\\nObservation: Found results"
+            ... )
+            >>> llm._apply_stop_words(response)
+            "I need to search.\\n\\nAction: search"
+        """
+        if not self.stop or not content:
+            return content
+
+        # Find the earliest occurrence of any stop word
+        earliest_stop_pos = len(content)
+        found_stop_word = None
+
+        for stop_word in self.stop:
+            stop_pos = content.find(stop_word)
+            if stop_pos != -1 and stop_pos < earliest_stop_pos:
+                earliest_stop_pos = stop_pos
+                found_stop_word = stop_word
+
+        # Truncate at the stop word if found
+        if found_stop_word is not None:
+            truncated = content[:earliest_stop_pos].strip()
+            logging.debug(
+                f"Applied stop word '{found_stop_word}' at position {earliest_stop_pos}"
+            )
+            return truncated
+
+        return content
+
+    def get_context_window_size(self) -> int:
+        """Get the context window size for the LLM.
+
+        Returns:
+            The number of tokens/characters the model can handle.
+        """
+        # Default implementation - subclasses should override with model-specific values
+        return DEFAULT_CONTEXT_WINDOW_SIZE
+
+    # Common helper methods for native SDK implementations
+
+    def _emit_call_started_event(
+        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,
+    ) -> None:
+        """Emit LLM call started event."""
+        if not hasattr(crewai_event_bus, "emit"):
+            raise ValueError("crewai_event_bus does not have an emit method") from None
+
+        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,
+            ),
+        )
+
+    def _emit_call_completed_event(
+        self,
+        response: Any,
+        call_type: LLMCallType,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
+        messages: str | list[dict[str, Any]] | None = None,
+    ) -> None:
+        """Emit LLM call completed event."""
+        crewai_event_bus.emit(
+            self,
+            event=LLMCallCompletedEvent(
+                messages=messages,
+                response=response,
+                call_type=call_type,
+                from_task=from_task,
+                from_agent=from_agent,
+                model=self.model,
+            ),
+        )
+
+    def _emit_call_failed_event(
+        self,
+        error: str,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
+    ) -> None:
+        """Emit LLM call failed event."""
+        if not hasattr(crewai_event_bus, "emit"):
+            raise ValueError("crewai_event_bus does not have an emit method") from None
+
+        crewai_event_bus.emit(
+            self,
+            event=LLMCallFailedEvent(
+                error=error,
+                from_task=from_task,
+                from_agent=from_agent,
+            ),
+        )
+
+    def _emit_stream_chunk_event(
+        self,
+        chunk: str,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
+        tool_call: dict[str, Any] | None = None,
+    ) -> None:
+        """Emit stream chunk event."""
+        if not hasattr(crewai_event_bus, "emit"):
+            raise ValueError("crewai_event_bus does not have an emit method") from None
+
+        crewai_event_bus.emit(
+            self,
+            event=LLMStreamChunkEvent(
+                chunk=chunk,
+                tool_call=tool_call,
+                from_task=from_task,
+                from_agent=from_agent,
+            ),
+        )
+
+    def _handle_tool_execution(
+        self,
+        function_name: str,
+        function_args: dict[str, Any],
+        available_functions: dict[str, Any],
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
+    ) -> str | None:
+        """Handle tool execution with proper event emission.
+
+        Args:
+            function_name: Name of the function to execute
+            function_args: Arguments to pass to the function
+            available_functions: Dict of available functions
+            from_task: Optional task object
+            from_agent: Optional agent object
+
+        Returns:
+            Result of function execution or None if function not found
+        """
+        if function_name not in available_functions:
+            logging.warning(
+                f"Function '{function_name}' not found in available functions"
+            )
+            return None
+
+        try:
+            # Emit tool usage started event
+            started_at = datetime.now()
+
+            crewai_event_bus.emit(
+                self,
+                event=ToolUsageStartedEvent(
+                    tool_name=function_name,
+                    tool_args=function_args,
+                    from_agent=from_agent,
+                    from_task=from_task,
+                ),
+            )
+
+            # Execute the function
+            fn = available_functions[function_name]
+            result = fn(**function_args)
+
+            # Emit tool usage finished event
+            crewai_event_bus.emit(
+                self,
+                event=ToolUsageFinishedEvent(
+                    output=result,
+                    tool_name=function_name,
+                    tool_args=function_args,
+                    started_at=started_at,
+                    finished_at=datetime.now(),
+                    from_task=from_task,
+                    from_agent=from_agent,
+                ),
+            )
+
+            # Emit LLM call completed event for tool call
+            self._emit_call_completed_event(
+                response=result,
+                call_type=LLMCallType.TOOL_CALL,
+                from_task=from_task,
+                from_agent=from_agent,
+            )
+
+            return str(result)
+
+        except Exception as e:
+            error_msg = f"Error executing function '{function_name}': {e!s}"
+            logging.error(error_msg)
+
+            # Emit tool usage error event
+            if not hasattr(crewai_event_bus, "emit"):
+                raise ValueError(
+                    "crewai_event_bus does not have an emit method"
+                ) from None
+
+            crewai_event_bus.emit(
+                self,
+                event=ToolUsageErrorEvent(
+                    tool_name=function_name,
+                    tool_args=function_args,
+                    error=error_msg,
+                    from_task=from_task,
+                    from_agent=from_agent,
+                ),
+            )
+
+            # Emit LLM call failed event
+            self._emit_call_failed_event(
+                error=error_msg,
+                from_task=from_task,
+                from_agent=from_agent,
+            )
+
+            return None
+
+    def _format_messages(self, messages: str | list[LLMMessage]) -> list[LLMMessage]:
+        """Convert messages to standard format.
+
+        Args:
+            messages: Input messages (string or list of message dicts)
+
+        Returns:
+            List of message dictionaries with 'role' and 'content' keys
+
+        Raises:
+            ValueError: If message format is invalid
+        """
+        if isinstance(messages, str):
+            return [{"role": "user", "content": messages}]
+
+        # Validate message format
+        for i, msg in enumerate(messages):
+            if not isinstance(msg, dict):
+                raise ValueError(f"Message at index {i} must be a dictionary")
+            if "role" not in msg or "content" not in msg:
+                raise ValueError(
+                    f"Message at index {i} must have 'role' and 'content' keys"
+                )
+
+        return messages
+
+    @staticmethod
+    def _validate_structured_output(
+        response: str,
+        response_format: type[BaseModel] | None,
+    ) -> str | BaseModel:
+        """Validate and parse structured output.
+
+        Args:
+            response: Raw response string
+            response_format: Optional Pydantic model for structured output
+
+        Returns:
+            Parsed response (BaseModel instance if response_format provided, otherwise string)
+
+        Raises:
+            ValueError: If structured output validation fails
+        """
+        if response_format is None:
+            return response
+
+        try:
+            # Try to parse as JSON first
+            if response.strip().startswith("{") or response.strip().startswith("["):
+                data = json.loads(response)
+                return response_format.model_validate(data)
+
+            json_match = _JSON_EXTRACTION_PATTERN.search(response)
+            if json_match:
+                data = json.loads(json_match.group())
+                return response_format.model_validate(data)
+
+            raise ValueError("No JSON found in response")
+
+        except (json.JSONDecodeError, ValueError) as e:
+            logging.warning(f"Failed to parse structured output: {e}")
+            raise ValueError(
+                f"Failed to parse response into {response_format.__name__}: {e}"
+            ) from e
+
+    @staticmethod
+    def _extract_provider(model: str) -> str:
+        """Extract provider from model string.
+
+        Args:
+            model: Model string (e.g., 'openai/gpt-4' or 'gpt-4')
+
+        Returns:
+            Provider name (e.g., 'openai')
+        """
+        if "/" in model:
+            return model.partition("/")[0]
+        return "openai"  # Default provider
+
+    def _track_token_usage_internal(self, usage_data: dict[str, Any]) -> None:
+        """Track token usage internally in the LLM instance.
+
+        Args:
+            usage_data: Token usage data from the API response
+        """
+        # Extract tokens in a provider-agnostic way
+        prompt_tokens = (
+            usage_data.get("prompt_tokens")
+            or usage_data.get("prompt_token_count")
+            or usage_data.get("input_tokens")
+            or 0
+        )
+
+        completion_tokens = (
+            usage_data.get("completion_tokens")
+            or usage_data.get("candidates_token_count")
+            or usage_data.get("output_tokens")
+            or 0
+        )
+
+        cached_tokens = (
+            usage_data.get("cached_tokens")
+            or usage_data.get("cached_prompt_tokens")
+            or 0
+        )
+
+        self._token_usage["prompt_tokens"] += prompt_tokens
+        self._token_usage["completion_tokens"] += completion_tokens
+        self._token_usage["total_tokens"] += prompt_tokens + completion_tokens
+        self._token_usage["successful_requests"] += 1
+        self._token_usage["cached_prompt_tokens"] += cached_tokens
+
+    def get_token_usage_summary(self) -> UsageMetrics:
+        """Get summary of token usage for this LLM instance.
+
+        Returns:
+            Dictionary with token usage totals
+        """
+        return UsageMetrics(**self._token_usage)

lib/crewai/src/crewai/llm/constants.py

@@ -0,0 +1,587 @@
+from typing import Literal, TypeAlias
+
+
+SupportedNativeProviders: TypeAlias = Literal[
+    "openai",
+    "anthropic",
+    "claude",
+    "azure",
+    "azure_openai",
+    "google",
+    "gemini",
+    "bedrock",
+    "aws",
+]
+
+SUPPORTED_NATIVE_PROVIDERS: list[SupportedNativeProviders] = [
+    "openai",
+    "anthropic",
+    "claude",
+    "azure",
+    "azure_openai",
+    "google",
+    "gemini",
+    "bedrock",
+    "aws",
+]
+
+
+OpenAIModels: TypeAlias = Literal[
+    "gpt-3.5-turbo",
+    "gpt-3.5-turbo-0125",
+    "gpt-3.5-turbo-0301",
+    "gpt-3.5-turbo-0613",
+    "gpt-3.5-turbo-1106",
+    "gpt-3.5-turbo-16k",
+    "gpt-3.5-turbo-16k-0613",
+    "gpt-3.5-turbo-instruct",
+    "gpt-3.5-turbo-instruct-0914",
+    "gpt-4",
+    "gpt-4-0125-preview",
+    "gpt-4-0314",
+    "gpt-4-0613",
+    "gpt-4-1106-preview",
+    "gpt-4-32k",
+    "gpt-4-32k-0314",
+    "gpt-4-32k-0613",
+    "gpt-4-turbo",
+    "gpt-4-turbo-2024-04-09",
+    "gpt-4-turbo-preview",
+    "gpt-4-vision-preview",
+    "gpt-4.1",
+    "gpt-4.1-2025-04-14",
+    "gpt-4.1-mini",
+    "gpt-4.1-mini-2025-04-14",
+    "gpt-4.1-nano",
+    "gpt-4.1-nano-2025-04-14",
+    "gpt-4o",
+    "gpt-4o-2024-05-13",
+    "gpt-4o-2024-08-06",
+    "gpt-4o-2024-11-20",
+    "gpt-4o-audio-preview",
+    "gpt-4o-audio-preview-2024-10-01",
+    "gpt-4o-audio-preview-2024-12-17",
+    "gpt-4o-audio-preview-2025-06-03",
+    "gpt-4o-mini",
+    "gpt-4o-mini-2024-07-18",
+    "gpt-4o-mini-audio-preview",
+    "gpt-4o-mini-audio-preview-2024-12-17",
+    "gpt-4o-mini-realtime-preview",
+    "gpt-4o-mini-realtime-preview-2024-12-17",
+    "gpt-4o-mini-search-preview",
+    "gpt-4o-mini-search-preview-2025-03-11",
+    "gpt-4o-mini-transcribe",
+    "gpt-4o-mini-tts",
+    "gpt-4o-realtime-preview",
+    "gpt-4o-realtime-preview-2024-10-01",
+    "gpt-4o-realtime-preview-2024-12-17",
+    "gpt-4o-realtime-preview-2025-06-03",
+    "gpt-4o-search-preview",
+    "gpt-4o-search-preview-2025-03-11",
+    "gpt-4o-transcribe",
+    "gpt-4o-transcribe-diarize",
+    "gpt-5",
+    "gpt-5-2025-08-07",
+    "gpt-5-chat",
+    "gpt-5-chat-latest",
+    "gpt-5-codex",
+    "gpt-5-mini",
+    "gpt-5-mini-2025-08-07",
+    "gpt-5-nano",
+    "gpt-5-nano-2025-08-07",
+    "gpt-5-pro",
+    "gpt-5-pro-2025-10-06",
+    "gpt-5-search-api",
+    "gpt-5-search-api-2025-10-14",
+    "gpt-audio",
+    "gpt-audio-2025-08-28",
+    "gpt-audio-mini",
+    "gpt-audio-mini-2025-10-06",
+    "gpt-image-1",
+    "gpt-image-1-mini",
+    "gpt-realtime",
+    "gpt-realtime-2025-08-28",
+    "gpt-realtime-mini",
+    "gpt-realtime-mini-2025-10-06",
+    "o1",
+    "o1-preview",
+    "o1-2024-12-17",
+    "o1-mini",
+    "o1-mini-2024-09-12",
+    "o1-pro",
+    "o1-pro-2025-03-19",
+    "o3-mini",
+    "o3",
+    "o4-mini",
+    "whisper-1",
+]
+OPENAI_MODELS: list[OpenAIModels] = [
+    "gpt-3.5-turbo",
+    "gpt-3.5-turbo-0125",
+    "gpt-3.5-turbo-0301",
+    "gpt-3.5-turbo-0613",
+    "gpt-3.5-turbo-1106",
+    "gpt-3.5-turbo-16k",
+    "gpt-3.5-turbo-16k-0613",
+    "gpt-3.5-turbo-instruct",
+    "gpt-3.5-turbo-instruct-0914",
+    "gpt-4",
+    "gpt-4-0125-preview",
+    "gpt-4-0314",
+    "gpt-4-0613",
+    "gpt-4-1106-preview",
+    "gpt-4-32k",
+    "gpt-4-32k-0314",
+    "gpt-4-32k-0613",
+    "gpt-4-turbo",
+    "gpt-4-turbo-2024-04-09",
+    "gpt-4-turbo-preview",
+    "gpt-4-vision-preview",
+    "gpt-4.1",
+    "gpt-4.1-2025-04-14",
+    "gpt-4.1-mini",
+    "gpt-4.1-mini-2025-04-14",
+    "gpt-4.1-nano",
+    "gpt-4.1-nano-2025-04-14",
+    "gpt-4o",
+    "gpt-4o-2024-05-13",
+    "gpt-4o-2024-08-06",
+    "gpt-4o-2024-11-20",
+    "gpt-4o-audio-preview",
+    "gpt-4o-audio-preview-2024-10-01",
+    "gpt-4o-audio-preview-2024-12-17",
+    "gpt-4o-audio-preview-2025-06-03",
+    "gpt-4o-mini",
+    "gpt-4o-mini-2024-07-18",
+    "gpt-4o-mini-audio-preview",
+    "gpt-4o-mini-audio-preview-2024-12-17",
+    "gpt-4o-mini-realtime-preview",
+    "gpt-4o-mini-realtime-preview-2024-12-17",
+    "gpt-4o-mini-search-preview",
+    "gpt-4o-mini-search-preview-2025-03-11",
+    "gpt-4o-mini-transcribe",
+    "gpt-4o-mini-tts",
+    "gpt-4o-realtime-preview",
+    "gpt-4o-realtime-preview-2024-10-01",
+    "gpt-4o-realtime-preview-2024-12-17",
+    "gpt-4o-realtime-preview-2025-06-03",
+    "gpt-4o-search-preview",
+    "gpt-4o-search-preview-2025-03-11",
+    "gpt-4o-transcribe",
+    "gpt-4o-transcribe-diarize",
+    "gpt-5",
+    "gpt-5-2025-08-07",
+    "gpt-5-chat",
+    "gpt-5-chat-latest",
+    "gpt-5-codex",
+    "gpt-5-mini",
+    "gpt-5-mini-2025-08-07",
+    "gpt-5-nano",
+    "gpt-5-nano-2025-08-07",
+    "gpt-5-pro",
+    "gpt-5-pro-2025-10-06",
+    "gpt-5-search-api",
+    "gpt-5-search-api-2025-10-14",
+    "gpt-audio",
+    "gpt-audio-2025-08-28",
+    "gpt-audio-mini",
+    "gpt-audio-mini-2025-10-06",
+    "gpt-image-1",
+    "gpt-image-1-mini",
+    "gpt-realtime",
+    "gpt-realtime-2025-08-28",
+    "gpt-realtime-mini",
+    "gpt-realtime-mini-2025-10-06",
+    "o1",
+    "o1-preview",
+    "o1-2024-12-17",
+    "o1-mini",
+    "o1-mini-2024-09-12",
+    "o1-pro",
+    "o1-pro-2025-03-19",
+    "o3-mini",
+    "o3",
+    "o4-mini",
+    "whisper-1",
+]
+
+
+AnthropicModels: TypeAlias = Literal[
+    "claude-3-7-sonnet-latest",
+    "claude-3-7-sonnet-20250219",
+    "claude-3-5-haiku-latest",
+    "claude-3-5-haiku-20241022",
+    "claude-haiku-4-5",
+    "claude-haiku-4-5-20251001",
+    "claude-sonnet-4-20250514",
+    "claude-sonnet-4-0",
+    "claude-4-sonnet-20250514",
+    "claude-sonnet-4-5",
+    "claude-sonnet-4-5-20250929",
+    "claude-3-5-sonnet-latest",
+    "claude-3-5-sonnet-20241022",
+    "claude-3-5-sonnet-20240620",
+    "claude-opus-4-0",
+    "claude-opus-4-20250514",
+    "claude-4-opus-20250514",
+    "claude-opus-4-1",
+    "claude-opus-4-1-20250805",
+    "claude-3-opus-latest",
+    "claude-3-opus-20240229",
+    "claude-3-sonnet-20240229",
+    "claude-3-haiku-latest",
+    "claude-3-haiku-20240307",
+]
+ANTHROPIC_MODELS: list[AnthropicModels] = [
+    "claude-3-7-sonnet-latest",
+    "claude-3-7-sonnet-20250219",
+    "claude-3-5-haiku-latest",
+    "claude-3-5-haiku-20241022",
+    "claude-haiku-4-5",
+    "claude-haiku-4-5-20251001",
+    "claude-sonnet-4-20250514",
+    "claude-sonnet-4-0",
+    "claude-4-sonnet-20250514",
+    "claude-sonnet-4-5",
+    "claude-sonnet-4-5-20250929",
+    "claude-3-5-sonnet-latest",
+    "claude-3-5-sonnet-20241022",
+    "claude-3-5-sonnet-20240620",
+    "claude-opus-4-0",
+    "claude-opus-4-20250514",
+    "claude-4-opus-20250514",
+    "claude-opus-4-1",
+    "claude-opus-4-1-20250805",
+    "claude-3-opus-latest",
+    "claude-3-opus-20240229",
+    "claude-3-sonnet-20240229",
+    "claude-3-haiku-latest",
+    "claude-3-haiku-20240307",
+]
+
+GeminiModels: TypeAlias = Literal[
+    "gemini-2.5-pro",
+    "gemini-2.5-pro-preview-03-25",
+    "gemini-2.5-pro-preview-05-06",
+    "gemini-2.5-pro-preview-06-05",
+    "gemini-2.5-flash",
+    "gemini-2.5-flash-preview-05-20",
+    "gemini-2.5-flash-preview-04-17",
+    "gemini-2.5-flash-image",
+    "gemini-2.5-flash-image-preview",
+    "gemini-2.5-flash-lite",
+    "gemini-2.5-flash-lite-preview-06-17",
+    "gemini-2.5-flash-preview-09-2025",
+    "gemini-2.5-flash-lite-preview-09-2025",
+    "gemini-2.5-flash-preview-tts",
+    "gemini-2.5-pro-preview-tts",
+    "gemini-2.5-computer-use-preview-10-2025",
+    "gemini-2.0-flash",
+    "gemini-2.0-flash-001",
+    "gemini-2.0-flash-exp",
+    "gemini-2.0-flash-exp-image-generation",
+    "gemini-2.0-flash-lite",
+    "gemini-2.0-flash-lite-001",
+    "gemini-2.0-flash-lite-preview",
+    "gemini-2.0-flash-lite-preview-02-05",
+    "gemini-2.0-flash-preview-image-generation",
+    "gemini-2.0-flash-thinking-exp",
+    "gemini-2.0-flash-thinking-exp-01-21",
+    "gemini-2.0-flash-thinking-exp-1219",
+    "gemini-2.0-pro-exp",
+    "gemini-2.0-pro-exp-02-05",
+    "gemini-exp-1206",
+    "gemini-1.5-pro",
+    "gemini-1.5-flash",
+    "gemini-1.5-flash-8b",
+    "gemini-flash-latest",
+    "gemini-flash-lite-latest",
+    "gemini-pro-latest",
+    "gemini-2.0-flash-live-001",
+    "gemini-live-2.5-flash-preview",
+    "gemini-2.5-flash-live-preview",
+    "gemini-robotics-er-1.5-preview",
+    "gemini-gemma-2-27b-it",
+    "gemini-gemma-2-9b-it",
+    "gemma-3-1b-it",
+    "gemma-3-4b-it",
+    "gemma-3-12b-it",
+    "gemma-3-27b-it",
+    "gemma-3n-e2b-it",
+    "gemma-3n-e4b-it",
+    "learnlm-2.0-flash-experimental",
+]
+GEMINI_MODELS: list[GeminiModels] = [
+    "gemini-2.5-pro",
+    "gemini-2.5-pro-preview-03-25",
+    "gemini-2.5-pro-preview-05-06",
+    "gemini-2.5-pro-preview-06-05",
+    "gemini-2.5-flash",
+    "gemini-2.5-flash-preview-05-20",
+    "gemini-2.5-flash-preview-04-17",
+    "gemini-2.5-flash-image",
+    "gemini-2.5-flash-image-preview",
+    "gemini-2.5-flash-lite",
+    "gemini-2.5-flash-lite-preview-06-17",
+    "gemini-2.5-flash-preview-09-2025",
+    "gemini-2.5-flash-lite-preview-09-2025",
+    "gemini-2.5-flash-preview-tts",
+    "gemini-2.5-pro-preview-tts",
+    "gemini-2.5-computer-use-preview-10-2025",
+    "gemini-2.0-flash",
+    "gemini-2.0-flash-001",
+    "gemini-2.0-flash-exp",
+    "gemini-2.0-flash-exp-image-generation",
+    "gemini-2.0-flash-lite",
+    "gemini-2.0-flash-lite-001",
+    "gemini-2.0-flash-lite-preview",
+    "gemini-2.0-flash-lite-preview-02-05",
+    "gemini-2.0-flash-preview-image-generation",
+    "gemini-2.0-flash-thinking-exp",
+    "gemini-2.0-flash-thinking-exp-01-21",
+    "gemini-2.0-flash-thinking-exp-1219",
+    "gemini-2.0-pro-exp",
+    "gemini-2.0-pro-exp-02-05",
+    "gemini-exp-1206",
+    "gemini-1.5-pro",
+    "gemini-1.5-flash",
+    "gemini-1.5-flash-8b",
+    "gemini-flash-latest",
+    "gemini-flash-lite-latest",
+    "gemini-pro-latest",
+    "gemini-2.0-flash-live-001",
+    "gemini-live-2.5-flash-preview",
+    "gemini-2.5-flash-live-preview",
+    "gemini-robotics-er-1.5-preview",
+    "gemini-gemma-2-27b-it",
+    "gemini-gemma-2-9b-it",
+    "gemma-3-1b-it",
+    "gemma-3-4b-it",
+    "gemma-3-12b-it",
+    "gemma-3-27b-it",
+    "gemma-3n-e2b-it",
+    "gemma-3n-e4b-it",
+    "learnlm-2.0-flash-experimental",
+]
+
+
+AzureModels: TypeAlias = Literal[
+    "gpt-3.5-turbo",
+    "gpt-3.5-turbo-0301",
+    "gpt-3.5-turbo-0613",
+    "gpt-3.5-turbo-16k",
+    "gpt-3.5-turbo-16k-0613",
+    "gpt-35-turbo",
+    "gpt-35-turbo-0125",
+    "gpt-35-turbo-1106",
+    "gpt-35-turbo-16k-0613",
+    "gpt-35-turbo-instruct-0914",
+    "gpt-4",
+    "gpt-4-0314",
+    "gpt-4-0613",
+    "gpt-4-1106-preview",
+    "gpt-4-0125-preview",
+    "gpt-4-32k",
+    "gpt-4-32k-0314",
+    "gpt-4-32k-0613",
+    "gpt-4-turbo",
+    "gpt-4-turbo-2024-04-09",
+    "gpt-4-vision",
+    "gpt-4o",
+    "gpt-4o-2024-05-13",
+    "gpt-4o-2024-08-06",
+    "gpt-4o-2024-11-20",
+    "gpt-4o-mini",
+    "gpt-5",
+    "o1",
+    "o1-mini",
+    "o1-preview",
+    "o3-mini",
+    "o3",
+    "o4-mini",
+]
+AZURE_MODELS: list[AzureModels] = [
+    "gpt-3.5-turbo",
+    "gpt-3.5-turbo-0301",
+    "gpt-3.5-turbo-0613",
+    "gpt-3.5-turbo-16k",
+    "gpt-3.5-turbo-16k-0613",
+    "gpt-35-turbo",
+    "gpt-35-turbo-0125",
+    "gpt-35-turbo-1106",
+    "gpt-35-turbo-16k-0613",
+    "gpt-35-turbo-instruct-0914",
+    "gpt-4",
+    "gpt-4-0314",
+    "gpt-4-0613",
+    "gpt-4-1106-preview",
+    "gpt-4-0125-preview",
+    "gpt-4-32k",
+    "gpt-4-32k-0314",
+    "gpt-4-32k-0613",
+    "gpt-4-turbo",
+    "gpt-4-turbo-2024-04-09",
+    "gpt-4-vision",
+    "gpt-4o",
+    "gpt-4o-2024-05-13",
+    "gpt-4o-2024-08-06",
+    "gpt-4o-2024-11-20",
+    "gpt-4o-mini",
+    "gpt-5",
+    "o1",
+    "o1-mini",
+    "o1-preview",
+    "o3-mini",
+    "o3",
+    "o4-mini",
+]
+
+
+BedrockModels: TypeAlias = Literal[
+    "ai21.jamba-1-5-large-v1:0",
+    "ai21.jamba-1-5-mini-v1:0",
+    "amazon.nova-lite-v1:0",
+    "amazon.nova-lite-v1:0:24k",
+    "amazon.nova-lite-v1:0:300k",
+    "amazon.nova-micro-v1:0",
+    "amazon.nova-micro-v1:0:128k",
+    "amazon.nova-micro-v1:0:24k",
+    "amazon.nova-premier-v1:0",
+    "amazon.nova-premier-v1:0:1000k",
+    "amazon.nova-premier-v1:0:20k",
+    "amazon.nova-premier-v1:0:8k",
+    "amazon.nova-premier-v1:0:mm",
+    "amazon.nova-pro-v1:0",
+    "amazon.nova-pro-v1:0:24k",
+    "amazon.nova-pro-v1:0:300k",
+    "amazon.titan-text-express-v1",
+    "amazon.titan-text-express-v1:0:8k",
+    "amazon.titan-text-lite-v1",
+    "amazon.titan-text-lite-v1:0:4k",
+    "amazon.titan-tg1-large",
+    "anthropic.claude-3-5-haiku-20241022-v1:0",
+    "anthropic.claude-3-5-sonnet-20240620-v1:0",
+    "anthropic.claude-3-5-sonnet-20241022-v2:0",
+    "anthropic.claude-3-7-sonnet-20250219-v1:0",
+    "anthropic.claude-3-haiku-20240307-v1:0",
+    "anthropic.claude-3-haiku-20240307-v1:0:200k",
+    "anthropic.claude-3-haiku-20240307-v1:0:48k",
+    "anthropic.claude-3-opus-20240229-v1:0",
+    "anthropic.claude-3-opus-20240229-v1:0:12k",
+    "anthropic.claude-3-opus-20240229-v1:0:200k",
+    "anthropic.claude-3-opus-20240229-v1:0:28k",
+    "anthropic.claude-3-sonnet-20240229-v1:0",
+    "anthropic.claude-3-sonnet-20240229-v1:0:200k",
+    "anthropic.claude-3-sonnet-20240229-v1:0:28k",
+    "anthropic.claude-haiku-4-5-20251001-v1:0",
+    "anthropic.claude-instant-v1:2:100k",
+    "anthropic.claude-opus-4-1-20250805-v1:0",
+    "anthropic.claude-opus-4-20250514-v1:0",
+    "anthropic.claude-sonnet-4-20250514-v1:0",
+    "anthropic.claude-sonnet-4-5-20250929-v1:0",
+    "anthropic.claude-v2:0:100k",
+    "anthropic.claude-v2:0:18k",
+    "anthropic.claude-v2:1:18k",
+    "anthropic.claude-v2:1:200k",
+    "cohere.command-r-plus-v1:0",
+    "cohere.command-r-v1:0",
+    "cohere.rerank-v3-5:0",
+    "deepseek.r1-v1:0",
+    "meta.llama3-1-70b-instruct-v1:0",
+    "meta.llama3-1-8b-instruct-v1:0",
+    "meta.llama3-2-11b-instruct-v1:0",
+    "meta.llama3-2-1b-instruct-v1:0",
+    "meta.llama3-2-3b-instruct-v1:0",
+    "meta.llama3-2-90b-instruct-v1:0",
+    "meta.llama3-3-70b-instruct-v1:0",
+    "meta.llama3-70b-instruct-v1:0",
+    "meta.llama3-8b-instruct-v1:0",
+    "meta.llama4-maverick-17b-instruct-v1:0",
+    "meta.llama4-scout-17b-instruct-v1:0",
+    "mistral.mistral-7b-instruct-v0:2",
+    "mistral.mistral-large-2402-v1:0",
+    "mistral.mistral-small-2402-v1:0",
+    "mistral.mixtral-8x7b-instruct-v0:1",
+    "mistral.pixtral-large-2502-v1:0",
+    "openai.gpt-oss-120b-1:0",
+    "openai.gpt-oss-20b-1:0",
+    "qwen.qwen3-32b-v1:0",
+    "qwen.qwen3-coder-30b-a3b-v1:0",
+    "twelvelabs.pegasus-1-2-v1:0",
+]
+BEDROCK_MODELS: list[BedrockModels] = [
+    "ai21.jamba-1-5-large-v1:0",
+    "ai21.jamba-1-5-mini-v1:0",
+    "amazon.nova-lite-v1:0",
+    "amazon.nova-lite-v1:0:24k",
+    "amazon.nova-lite-v1:0:300k",
+    "amazon.nova-micro-v1:0",
+    "amazon.nova-micro-v1:0:128k",
+    "amazon.nova-micro-v1:0:24k",
+    "amazon.nova-premier-v1:0",
+    "amazon.nova-premier-v1:0:1000k",
+    "amazon.nova-premier-v1:0:20k",
+    "amazon.nova-premier-v1:0:8k",
+    "amazon.nova-premier-v1:0:mm",
+    "amazon.nova-pro-v1:0",
+    "amazon.nova-pro-v1:0:24k",
+    "amazon.nova-pro-v1:0:300k",
+    "amazon.titan-text-express-v1",
+    "amazon.titan-text-express-v1:0:8k",
+    "amazon.titan-text-lite-v1",
+    "amazon.titan-text-lite-v1:0:4k",
+    "amazon.titan-tg1-large",
+    "anthropic.claude-3-5-haiku-20241022-v1:0",
+    "anthropic.claude-3-5-sonnet-20240620-v1:0",
+    "anthropic.claude-3-5-sonnet-20241022-v2:0",
+    "anthropic.claude-3-7-sonnet-20250219-v1:0",
+    "anthropic.claude-3-haiku-20240307-v1:0",
+    "anthropic.claude-3-haiku-20240307-v1:0:200k",
+    "anthropic.claude-3-haiku-20240307-v1:0:48k",
+    "anthropic.claude-3-opus-20240229-v1:0",
+    "anthropic.claude-3-opus-20240229-v1:0:12k",
+    "anthropic.claude-3-opus-20240229-v1:0:200k",
+    "anthropic.claude-3-opus-20240229-v1:0:28k",
+    "anthropic.claude-3-sonnet-20240229-v1:0",
+    "anthropic.claude-3-sonnet-20240229-v1:0:200k",
+    "anthropic.claude-3-sonnet-20240229-v1:0:28k",
+    "anthropic.claude-haiku-4-5-20251001-v1:0",
+    "anthropic.claude-instant-v1:2:100k",
+    "anthropic.claude-opus-4-1-20250805-v1:0",
+    "anthropic.claude-opus-4-20250514-v1:0",
+    "anthropic.claude-sonnet-4-20250514-v1:0",
+    "anthropic.claude-sonnet-4-5-20250929-v1:0",
+    "anthropic.claude-v2:0:100k",
+    "anthropic.claude-v2:0:18k",
+    "anthropic.claude-v2:1:18k",
+    "anthropic.claude-v2:1:200k",
+    "cohere.command-r-plus-v1:0",
+    "cohere.command-r-v1:0",
+    "cohere.rerank-v3-5:0",
+    "deepseek.r1-v1:0",
+    "meta.llama3-1-70b-instruct-v1:0",
+    "meta.llama3-1-8b-instruct-v1:0",
+    "meta.llama3-2-11b-instruct-v1:0",
+    "meta.llama3-2-1b-instruct-v1:0",
+    "meta.llama3-2-3b-instruct-v1:0",
+    "meta.llama3-2-90b-instruct-v1:0",
+    "meta.llama3-3-70b-instruct-v1:0",
+    "meta.llama3-70b-instruct-v1:0",
+    "meta.llama3-8b-instruct-v1:0",
+    "meta.llama4-maverick-17b-instruct-v1:0",
+    "meta.llama4-scout-17b-instruct-v1:0",
+    "mistral.mistral-7b-instruct-v0:2",
+    "mistral.mistral-large-2402-v1:0",
+    "mistral.mistral-small-2402-v1:0",
+    "mistral.mixtral-8x7b-instruct-v0:1",
+    "mistral.pixtral-large-2502-v1:0",
+    "openai.gpt-oss-120b-1:0",
+    "openai.gpt-oss-20b-1:0",
+    "qwen.qwen3-32b-v1:0",
+    "qwen.qwen3-coder-30b-a3b-v1:0",
+    "twelvelabs.pegasus-1-2-v1:0",
+]
+
+SupportedModels: TypeAlias = (
+    OpenAIModels | AnthropicModels | GeminiModels | AzureModels | BedrockModels
+)

lib/crewai/src/crewai/llm/core.py

@@ -20,9 +20,7 @@ from typing import (
 )
 
 from dotenv import load_dotenv
-import httpx
 from pydantic import BaseModel, Field
-from typing_extensions import Self
 
 from crewai.events.event_bus import crewai_event_bus
 from crewai.events.types.llm_events import (
@@ -37,14 +35,7 @@ from crewai.events.types.tool_usage_events import (
     ToolUsageFinishedEvent,
     ToolUsageStartedEvent,
 )
-from crewai.llms.base_llm import BaseLLM
-from crewai.llms.constants import (
-    ANTHROPIC_MODELS,
-    AZURE_MODELS,
-    BEDROCK_MODELS,
-    GEMINI_MODELS,
-    OPENAI_MODELS,
-)
+from crewai.llm.base_llm import BaseLLM
 from crewai.utilities import InternalInstructor
 from crewai.utilities.exceptions.context_window_exceeding_exception import (
     LLMContextLengthExceededError,
@@ -61,7 +52,6 @@ if TYPE_CHECKING:
     from litellm.utils import supports_response_schema
 
     from crewai.agent.core import Agent
-    from crewai.llms.hooks.base import BaseInterceptor
     from crewai.task import Task
     from crewai.tools.base_tool import BaseTool
     from crewai.utilities.types import LLMMessage
@@ -327,249 +317,57 @@ class AccumulatedToolArgs(BaseModel):
 
 
 class LLM(BaseLLM):
-    completion_cost: float | None = None
+    """LiteLLM-based LLM implementation for CrewAI.
 
-    def __new__(cls, model: str, is_litellm: bool = False, **kwargs: Any) -> LLM:
-        """Factory method that routes to native SDK or falls back to LiteLLM.
+    This class provides LiteLLM integration for models not covered by native providers.
+    The metaclass (LLMMeta) automatically routes to native providers when appropriate.
+    """
 
-        Routing priority:
-            1. If 'provider' kwarg is present, use that provider with constants
-            2. If only 'model' kwarg, use constants to infer provider
-            3. If "/" in model name:
-               - Check if prefix is a native provider (openai/anthropic/azure/bedrock/gemini)
-               - If yes, validate model against constants
-               - If valid, route to native SDK; otherwise route to LiteLLM
-        """
-        if not model or not isinstance(model, str):
-            raise ValueError("Model must be a non-empty string")
+    # LiteLLM-specific fields
+    completion_cost: float | None = Field(None, description="Cost of completion")
+    timeout: float | int | None = Field(None, description="Request timeout")
+    top_p: float | None = Field(None, description="Top-p sampling parameter")
+    n: int | None = Field(None, description="Number of completions to generate")
+    max_completion_tokens: int | None = Field(
+        None, description="Maximum completion tokens"
+    )
+    max_tokens: int | float | None = Field(None, description="Maximum total tokens")
+    presence_penalty: float | None = Field(None, description="Presence penalty")
+    frequency_penalty: float | None = Field(None, description="Frequency penalty")
+    logit_bias: dict[int, float] | None = Field(None, description="Logit bias")
+    response_format: type[BaseModel] | None = Field(
+        None, description="Response format model"
+    )
+    seed: int | None = Field(None, description="Random seed for reproducibility")
+    logprobs: int | None = Field(None, description="Log probabilities to return")
+    top_logprobs: int | None = Field(None, description="Top log probabilities")
+    api_base: str | None = Field(None, description="API base URL (alias for base_url)")
+    api_version: str | None = Field(None, description="API version")
+    callbacks: list[Any] | None = Field(None, description="Callback functions")
+    context_window_size: int = Field(0, description="Context window size in tokens")
+    reasoning_effort: Literal["none", "low", "medium", "high"] | None = Field(
+        None, description="Reasoning effort level"
+    )
+    is_anthropic: bool = Field(False, description="Whether model is from Anthropic")
+    stream: bool = Field(False, description="Whether to stream responses")
 
-        explicit_provider = kwargs.get("provider")
-
-        if explicit_provider:
-            provider = explicit_provider
-            use_native = True
-            model_string = model
-        elif "/" in model:
-            prefix, _, model_part = model.partition("/")
-
-            provider_mapping = {
-                "openai": "openai",
-                "anthropic": "anthropic",
-                "claude": "anthropic",
-                "azure": "azure",
-                "azure_openai": "azure",
-                "google": "gemini",
-                "gemini": "gemini",
-                "bedrock": "bedrock",
-                "aws": "bedrock",
-            }
-
-            canonical_provider = provider_mapping.get(prefix.lower())
-
-            if canonical_provider and cls._validate_model_in_constants(
-                model_part, canonical_provider
-            ):
-                provider = canonical_provider
-                use_native = True
-                model_string = model_part
-            else:
-                provider = prefix
-                use_native = False
-                model_string = model_part
-        else:
-            provider = cls._infer_provider_from_model(model)
-            use_native = True
-            model_string = model
-
-        native_class = cls._get_native_provider(provider) if use_native else None
-        if native_class and not is_litellm and provider in SUPPORTED_NATIVE_PROVIDERS:
-            try:
-                # Remove 'provider' from kwargs if it exists to avoid duplicate keyword argument
-                kwargs_copy = {k: v for k, v in kwargs.items() if k != 'provider'}
-                return cast(
-                    Self, native_class(model=model_string, provider=provider, **kwargs_copy)
-                )
-            except NotImplementedError:
-                raise
-            except Exception as e:
-                raise ImportError(f"Error importing native provider: {e}") from e
-
-        # FALLBACK to LiteLLM
-        if not LITELLM_AVAILABLE:
-            logger.error("LiteLLM is not available, falling back to LiteLLM")
-            raise ImportError("Fallback to LiteLLM is not available") from None
-
-        instance = object.__new__(cls)
-        super(LLM, instance).__init__(model=model, is_litellm=True, **kwargs)
-        instance.is_litellm = True
-        return instance
-
-    @classmethod
-    def _validate_model_in_constants(cls, model: str, provider: str) -> bool:
-        """Validate if a model name exists in the provider's constants.
+    def model_post_init(self, __context: Any) -> None:
+        """Initialize LiteLLM-specific settings after model initialization.
 
         Args:
-            model: The model name to validate
-            provider: The provider to check against (canonical name)
-
-        Returns:
-            True if the model exists in the provider's constants, False otherwise
+            __context: Pydantic context
         """
-        if provider == "openai":
-            return model in OPENAI_MODELS
+        super().model_post_init(__context)
 
-        if provider == "anthropic" or provider == "claude":
-            return model in ANTHROPIC_MODELS
+        # Configure LiteLLM
+        if LITELLM_AVAILABLE:
+            litellm.drop_params = True
 
-        if provider == "gemini":
-            return model in GEMINI_MODELS
+        # Determine if this is an Anthropic model
+        self.is_anthropic = self._is_anthropic_model(self.model)
 
-        if provider == "bedrock":
-            return model in BEDROCK_MODELS
-
-        if provider == "azure":
-            # azure does not provide a list of available models, determine a better way to handle this
-            return True
-
-        return False
-
-    @classmethod
-    def _infer_provider_from_model(cls, model: str) -> str:
-        """Infer the provider from the model name.
-
-        Args:
-            model: The model name without provider prefix
-
-        Returns:
-            The inferred provider name, defaults to "openai"
-        """
-
-        if model in OPENAI_MODELS:
-            return "openai"
-
-        if model in ANTHROPIC_MODELS:
-            return "anthropic"
-
-        if model in GEMINI_MODELS:
-            return "gemini"
-
-        if model in BEDROCK_MODELS:
-            return "bedrock"
-
-        if model in AZURE_MODELS:
-            return "azure"
-
-        return "openai"
-
-    @classmethod
-    def _get_native_provider(cls, provider: str) -> type | None:
-        """Get native provider class if available."""
-        if provider == "openai":
-            from crewai.llms.providers.openai.completion import OpenAICompletion
-
-            return OpenAICompletion
-
-        if provider == "anthropic" or provider == "claude":
-            from crewai.llms.providers.anthropic.completion import (
-                AnthropicCompletion,
-            )
-
-            return AnthropicCompletion
-
-        if provider == "azure" or provider == "azure_openai":
-            from crewai.llms.providers.azure.completion import AzureCompletion
-
-            return AzureCompletion
-
-        if provider == "google" or provider == "gemini":
-            from crewai.llms.providers.gemini.completion import GeminiCompletion
-
-            return GeminiCompletion
-
-        if provider == "bedrock":
-            from crewai.llms.providers.bedrock.completion import BedrockCompletion
-
-            return BedrockCompletion
-
-        return None
-
-    def __init__(
-        self,
-        model: str,
-        timeout: float | int | None = None,
-        temperature: float | None = None,
-        top_p: float | None = None,
-        n: int | None = None,
-        stop: str | list[str] | None = None,
-        max_completion_tokens: int | None = None,
-        max_tokens: int | float | None = None,
-        presence_penalty: float | None = None,
-        frequency_penalty: float | None = None,
-        logit_bias: dict[int, float] | None = None,
-        response_format: type[BaseModel] | None = None,
-        seed: int | None = None,
-        logprobs: int | None = None,
-        top_logprobs: int | None = None,
-        base_url: str | None = None,
-        api_base: str | None = None,
-        api_version: str | None = None,
-        api_key: str | None = None,
-        callbacks: list[Any] | None = None,
-        reasoning_effort: Literal["none", "low", "medium", "high"] | None = None,
-        stream: bool = False,
-        interceptor: BaseInterceptor[httpx.Request, httpx.Response] | None = None,
-        **kwargs: Any,
-    ) -> None:
-        """Initialize LLM instance.
-
-        Note: This __init__ method is only called for fallback instances.
-        Native provider instances handle their own initialization in their respective classes.
-        """
-        super().__init__(
-            model=model,
-            temperature=temperature,
-            api_key=api_key,
-            base_url=base_url,
-            timeout=timeout,
-            **kwargs,
-        )
-        self.model = model
-        self.timeout = timeout
-        self.temperature = temperature
-        self.top_p = top_p
-        self.n = n
-        self.max_completion_tokens = max_completion_tokens
-        self.max_tokens = max_tokens
-        self.presence_penalty = presence_penalty
-        self.frequency_penalty = frequency_penalty
-        self.logit_bias = logit_bias
-        self.response_format = response_format
-        self.seed = seed
-        self.logprobs = logprobs
-        self.top_logprobs = top_logprobs
-        self.base_url = base_url
-        self.api_base = api_base
-        self.api_version = api_version
-        self.api_key = api_key
-        self.callbacks = callbacks
-        self.context_window_size = 0
-        self.reasoning_effort = reasoning_effort
-        self.additional_params = kwargs
-        self.is_anthropic = self._is_anthropic_model(model)
-        self.stream = stream
-        self.interceptor = interceptor
-
-        litellm.drop_params = True
-
-        # Normalize self.stop to always be a list[str]
-        if stop is None:
-            self.stop: list[str] = []
-        elif isinstance(stop, str):
-            self.stop = [stop]
-        else:
-            self.stop = stop
-
-        self.set_callbacks(callbacks or [])
+        # Set up callbacks
+        self.set_callbacks(self.callbacks or [])
         self.set_env_callbacks()
 
     @staticmethod
@@ -1649,7 +1447,7 @@ class LLM(BaseLLM):
             **filtered_params,
         )
 
-    def __deepcopy__(self, memo: dict[int, Any] | None) -> LLM:
+    def __deepcopy__(self, memo: dict[int, Any] | None) -> LLM:  # type: ignore[override]
         """Create a deep copy of the LLM instance."""
         import copy
 

lib/crewai/src/crewai/llm/hooks/__init__.py

@@ -0,0 +1,6 @@
+"""Interceptor contracts for crewai"""
+
+from crewai.llm.hooks.base import BaseInterceptor
+
+
+__all__ = ["BaseInterceptor"]

lib/crewai/src/crewai/llm/hooks/base.py

@@ -0,0 +1,133 @@
+"""Base classes for LLM transport interceptors.
+
+This module provides abstract base classes for intercepting and modifying
+outbound and inbound messages at the transport level.
+"""
+
+from __future__ import annotations
+
+from abc import ABC, abstractmethod
+from typing import TYPE_CHECKING, Any, Generic, TypeVar
+
+from pydantic_core import core_schema
+
+
+if TYPE_CHECKING:
+    from pydantic import GetCoreSchemaHandler
+    from pydantic_core import CoreSchema
+
+
+T = TypeVar("T")
+U = TypeVar("U")
+
+
+class BaseInterceptor(ABC, Generic[T, U]):
+    """Abstract base class for intercepting transport-level messages.
+
+    Provides hooks to intercept and modify outbound and inbound messages
+    at the transport layer.
+
+    Type parameters:
+        T: Outbound message type (e.g., httpx.Request)
+        U: Inbound message type (e.g., httpx.Response)
+
+    Example:
+        >>> import httpx
+        >>> class CustomInterceptor(BaseInterceptor[httpx.Request, httpx.Response]):
+        ...     def on_outbound(self, message: httpx.Request) -> httpx.Request:
+        ...         message.headers["X-Custom-Header"] = "value"
+        ...         return message
+        ...
+        ...     def on_inbound(self, message: httpx.Response) -> httpx.Response:
+        ...         print(f"Status: {message.status_code}")
+        ...         return message
+    """
+
+    @abstractmethod
+    def on_outbound(self, message: T) -> T:
+        """Intercept outbound message before sending.
+
+        Args:
+            message: Outbound message object.
+
+        Returns:
+            Modified message object.
+        """
+        ...
+
+    @abstractmethod
+    def on_inbound(self, message: U) -> U:
+        """Intercept inbound message after receiving.
+
+        Args:
+            message: Inbound message object.
+
+        Returns:
+            Modified message object.
+        """
+        ...
+
+    async def aon_outbound(self, message: T) -> T:
+        """Async version of on_outbound.
+
+        Args:
+            message: Outbound message object.
+
+        Returns:
+            Modified message object.
+        """
+        raise NotImplementedError
+
+    async def aon_inbound(self, message: U) -> U:
+        """Async version of on_inbound.
+
+        Args:
+            message: Inbound message object.
+
+        Returns:
+            Modified message object.
+        """
+        raise NotImplementedError
+
+    @classmethod
+    def __get_pydantic_core_schema__(
+        cls, _source_type: Any, _handler: GetCoreSchemaHandler
+    ) -> CoreSchema:
+        """Generate Pydantic core schema for BaseInterceptor.
+
+        This allows the generic BaseInterceptor to be used in Pydantic models
+        without requiring arbitrary_types_allowed=True. The schema validates
+        that the value is an instance of BaseInterceptor.
+
+        Args:
+            _source_type: The source type being validated (unused).
+            _handler: Handler for generating schemas (unused).
+
+        Returns:
+            A Pydantic core schema that validates BaseInterceptor instances.
+        """
+        return core_schema.no_info_plain_validator_function(
+            _validate_interceptor,
+            serialization=core_schema.plain_serializer_function_ser_schema(
+                lambda x: x, return_schema=core_schema.any_schema()
+            ),
+        )
+
+
+def _validate_interceptor(value: Any) -> BaseInterceptor[T, U]:
+    """Validate that the value is a BaseInterceptor instance.
+
+    Args:
+        value: The value to validate.
+
+    Returns:
+        The validated BaseInterceptor instance.
+
+    Raises:
+        ValueError: If the value is not a BaseInterceptor instance.
+    """
+    if not isinstance(value, BaseInterceptor):
+        raise ValueError(
+            f"Expected BaseInterceptor instance, got {type(value).__name__}"
+        )
+    return value

lib/crewai/src/crewai/llm/hooks/transport.py

@@ -0,0 +1,123 @@
+"""HTTP transport implementations for LLM request/response interception.
+
+This module provides internal transport classes that integrate with BaseInterceptor
+to enable request/response modification at the transport level.
+"""
+
+from __future__ import annotations
+
+from collections.abc import Iterable
+from typing import TYPE_CHECKING, TypedDict
+
+from httpx import (
+    AsyncHTTPTransport as _AsyncHTTPTransport,
+    HTTPTransport as _HTTPTransport,
+)
+from typing_extensions import NotRequired, Unpack
+
+
+if TYPE_CHECKING:
+    from ssl import SSLContext
+
+    from httpx import Limits, Request, Response
+    from httpx._types import CertTypes, ProxyTypes
+
+    from crewai.llm.hooks.base import BaseInterceptor
+
+
+class HTTPTransportKwargs(TypedDict, total=False):
+    """Typed dictionary for httpx.HTTPTransport initialization parameters.
+
+    These parameters configure the underlying HTTP transport behavior including
+    SSL verification, proxies, connection limits, and low-level socket options.
+    """
+
+    verify: bool | str | SSLContext
+    cert: NotRequired[CertTypes]
+    trust_env: bool
+    http1: bool
+    http2: bool
+    limits: Limits
+    proxy: NotRequired[ProxyTypes]
+    uds: NotRequired[str]
+    local_address: NotRequired[str]
+    retries: int
+    socket_options: NotRequired[
+        Iterable[
+            tuple[int, int, int]
+            | tuple[int, int, bytes | bytearray]
+            | tuple[int, int, None, int]
+        ]
+    ]
+
+
+class HTTPTransport(_HTTPTransport):
+    """HTTP transport that uses an interceptor for request/response modification.
+
+    This transport is used internally when a user provides a BaseInterceptor.
+    Users should not instantiate this class directly - instead, pass an interceptor
+    to the LLM client and this transport will be created automatically.
+    """
+
+    def __init__(
+        self,
+        interceptor: BaseInterceptor[Request, Response],
+        **kwargs: Unpack[HTTPTransportKwargs],
+    ) -> None:
+        """Initialize transport with interceptor.
+
+        Args:
+            interceptor: HTTP interceptor for modifying raw request/response objects.
+            **kwargs: HTTPTransport configuration parameters (verify, cert, proxy, etc.).
+        """
+        super().__init__(**kwargs)
+        self.interceptor = interceptor
+
+    def handle_request(self, request: Request) -> Response:
+        """Handle request with interception.
+
+        Args:
+            request: The HTTP request to handle.
+
+        Returns:
+            The HTTP response.
+        """
+        request = self.interceptor.on_outbound(request)
+        response = super().handle_request(request)
+        return self.interceptor.on_inbound(response)
+
+
+class AsyncHTTPTransport(_AsyncHTTPTransport):
+    """Async HTTP transport that uses an interceptor for request/response modification.
+
+    This transport is used internally when a user provides a BaseInterceptor.
+    Users should not instantiate this class directly - instead, pass an interceptor
+    to the LLM client and this transport will be created automatically.
+    """
+
+    def __init__(
+        self,
+        interceptor: BaseInterceptor[Request, Response],
+        **kwargs: Unpack[HTTPTransportKwargs],
+    ) -> None:
+        """Initialize async transport with interceptor.
+
+        Args:
+            interceptor: HTTP interceptor for modifying raw request/response objects.
+            **kwargs: HTTPTransport configuration parameters (verify, cert, proxy, etc.).
+        """
+        super().__init__(**kwargs)
+        self.interceptor = interceptor
+
+    async def handle_async_request(self, request: Request) -> Response:
+        """Handle async request with interception.
+
+        Args:
+            request: The HTTP request to handle.
+
+        Returns:
+            The HTTP response.
+        """
+        request = await self.interceptor.aon_outbound(request)
+        response = await super().handle_async_request(request)
+        return await self.interceptor.aon_inbound(response)

lib/crewai/src/crewai/llm/internal/constants.py

@@ -0,0 +1,14 @@
+from crewai.llm.constants import SupportedNativeProviders
+
+
+PROVIDER_MAPPING: dict[str, SupportedNativeProviders] = {
+    "openai": "openai",
+    "anthropic": "anthropic",
+    "claude": "anthropic",
+    "azure": "azure",
+    "azure_openai": "azure",
+    "google": "gemini",
+    "gemini": "gemini",
+    "bedrock": "bedrock",
+    "aws": "bedrock",
+}

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

@@ -0,0 +1,251 @@
+"""Metaclass for LLM provider routing.
+
+This metaclass enables automatic routing to native provider implementations
+based on the model parameter at instantiation time.
+"""
+
+from __future__ import annotations
+
+import logging
+from typing import Any, cast
+
+from pydantic import ConfigDict
+from pydantic._internal._model_construction import ModelMetaclass
+
+from crewai.llm.constants import (
+    ANTHROPIC_MODELS,
+    AZURE_MODELS,
+    BEDROCK_MODELS,
+    GEMINI_MODELS,
+    OPENAI_MODELS,
+    SUPPORTED_NATIVE_PROVIDERS,
+    SupportedModels,
+    SupportedNativeProviders,
+)
+from crewai.llm.internal.constants import PROVIDER_MAPPING
+
+
+class LLMMeta(ModelMetaclass):
+    """Metaclass for LLM that handles provider routing.
+
+    This metaclass intercepts LLM instantiation and routes to the appropriate
+    native provider implementation based on the model parameter.
+    """
+
+    def __new__(
+        mcs,
+        name: str,
+        bases: tuple[type, ...],
+        namespace: dict[str, Any],
+        **kwargs: Any,
+    ) -> type:
+        """Create new LLM class with proper model_config for custom LLMs.
+
+        Args:
+            name: Class name
+            bases: Base classes
+            namespace: Class namespace
+            **kwargs: Additional arguments
+
+        Returns:
+            New class
+        """
+        if name != "BaseLLM" and any(
+            base.__name__ in ("BaseLLM", "LLM") for base in bases
+        ):
+            if "model_config" not in namespace:
+                namespace["model_config"] = ConfigDict(
+                    extra="allow", populate_by_name=True
+                )
+            elif isinstance(namespace["model_config"], dict):
+                config_dict = cast(
+                    ConfigDict, cast(object, dict(namespace["model_config"]))
+                )
+                config_dict.setdefault("extra", "allow")
+                config_dict.setdefault("populate_by_name", True)
+                namespace["model_config"] = ConfigDict(**config_dict)
+
+        return super().__new__(mcs, name, bases, namespace)
+
+    def __call__(cls, *args: Any, **kwargs: Any) -> Any:  # noqa: N805
+        """Route to appropriate provider implementation at instantiation time.
+
+        Args:
+            *args: Positional arguments (model should be first for LLM class)
+            **kwargs: Keyword arguments including model, is_litellm, etc.
+
+        Returns:
+            Instance of the appropriate provider class or LLM class
+
+        Raises:
+            ValueError: If model is not a valid string
+        """
+        if cls.__name__ != "LLM":
+            return super().__call__(*args, **kwargs)
+
+        model = cast(
+            str | SupportedModels | None,
+            (kwargs.get("model") or (args[0] if args else None)),
+        )
+        is_litellm = kwargs.get("is_litellm", False)
+
+        if not model or not isinstance(model, str):
+            raise ValueError("Model must be a non-empty string")
+
+        if args and not kwargs.get("model"):
+            kwargs["model"] = cast(SupportedModels, args[0])
+            _ = args[1:]
+        explicit_provider = cast(SupportedNativeProviders, kwargs.get("provider"))
+
+        if explicit_provider:
+            provider = explicit_provider
+            use_native = True
+            model_string = model
+        elif "/" in model:
+            prefix, _, model_part = cast(
+                tuple[SupportedNativeProviders, Any, SupportedModels],
+                model.partition("/"),
+            )
+
+            canonical_provider = PROVIDER_MAPPING.get(prefix.lower())
+
+            if canonical_provider and cls._validate_model_in_constants(
+                model_part, canonical_provider
+            ):
+                provider = canonical_provider
+                use_native = True
+                model_string = model_part
+            else:
+                provider = prefix
+                use_native = False
+                model_string = model_part
+        else:
+            provider = cls._infer_provider_from_model(model)
+            use_native = True
+            model_string = model
+
+        native_class = cls._get_native_provider(provider) if use_native else None
+        if native_class and not is_litellm and provider in SUPPORTED_NATIVE_PROVIDERS:
+            try:
+                kwargs_copy = {
+                    k: v for k, v in kwargs.items() if k not in ("provider", "model")
+                }
+                return native_class(
+                    model=model_string, provider=provider, **kwargs_copy
+                )
+            except NotImplementedError:
+                raise
+            except Exception as e:
+                raise ImportError(f"Error importing native provider: {e}") from e
+
+        try:
+            import litellm  # noqa: F401
+        except ImportError:
+            logging.error("LiteLLM is not available, falling back to LiteLLM")
+            raise ImportError("Fallback to LiteLLM is not available") from None
+
+        kwargs_copy = {
+            k: v for k, v in kwargs.items() if k not in ("model", "is_litellm")
+        }
+        return super().__call__(model=model, is_litellm=True, **kwargs_copy)
+
+    @staticmethod
+    def _validate_model_in_constants(
+        model: SupportedModels, provider: SupportedNativeProviders | None
+    ) -> bool:
+        """Validate if a model name exists in the provider's constants.
+
+        Args:
+            model: The model name to validate
+            provider: The provider to check against (canonical name)
+
+        Returns:
+            True if the model exists in the provider's constants, False otherwise
+        """
+
+        if provider == "openai":
+            return model in OPENAI_MODELS
+
+        if provider == "anthropic" or provider == "claude":
+            return model in ANTHROPIC_MODELS
+
+        if provider == "gemini":
+            return model in GEMINI_MODELS
+
+        if provider == "bedrock":
+            return model in BEDROCK_MODELS
+
+        if provider == "azure":
+            # azure does not provide a list of available models
+            return True
+
+        return False
+
+    @staticmethod
+    def _infer_provider_from_model(
+        model: SupportedModels | str,
+    ) -> SupportedNativeProviders:
+        """Infer the provider from the model name.
+
+        Args:
+            model: The model name without provider prefix
+
+        Returns:
+            The inferred provider name, defaults to "openai"
+        """
+
+        if model in OPENAI_MODELS:
+            return "openai"
+
+        if model in ANTHROPIC_MODELS:
+            return "anthropic"
+
+        if model in GEMINI_MODELS:
+            return "gemini"
+
+        if model in BEDROCK_MODELS:
+            return "bedrock"
+
+        if model in AZURE_MODELS:
+            return "azure"
+
+        return "openai"
+
+    @staticmethod
+    def _get_native_provider(provider: SupportedNativeProviders | None) -> type | None:
+        """Get native provider class if available.
+
+        Args:
+            provider: The provider name
+
+        Returns:
+            The provider class or None if not available
+        """
+        if provider == "openai":
+            from crewai.llm.providers.openai.completion import OpenAICompletion
+
+            return OpenAICompletion
+
+        if provider == "anthropic" or provider == "claude":
+            from crewai.llm.providers.anthropic.completion import (
+                AnthropicCompletion,
+            )
+
+            return AnthropicCompletion
+
+        if provider == "azure" or provider == "azure_openai":
+            from crewai.llm.providers.azure.completion import AzureCompletion
+
+            return AzureCompletion
+
+        if provider == "google" or provider == "gemini":
+            from crewai.llm.providers.gemini.completion import GeminiCompletion
+
+            return GeminiCompletion
+
+        if provider == "bedrock":
+            from crewai.llm.providers.bedrock.completion import BedrockCompletion
+
+            return BedrockCompletion
+
+        return None

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

@@ -2,14 +2,18 @@ from __future__ import annotations
 
 import json
 import logging
-import os
 from typing import TYPE_CHECKING, Any, cast
 
-from pydantic import BaseModel
+from dotenv import load_dotenv
+import httpx
+from pydantic import BaseModel, Field, PrivateAttr, model_validator
+from typing_extensions import Self
 
 from crewai.events.types.llm_events import LLMCallType
-from crewai.llms.base_llm import BaseLLM
-from crewai.llms.hooks.transport import HTTPTransport
+from crewai.llm.base_llm import BaseLLM
+from crewai.llm.core import CONTEXT_WINDOW_USAGE_RATIO
+from crewai.llm.hooks.transport import HTTPTransport
+from crewai.llm.providers.utils.common import safe_tool_conversion
 from crewai.utilities.agent_utils import is_context_length_exceeded
 from crewai.utilities.exceptions.context_window_exceeding_exception import (
     LLMContextLengthExceededError,
@@ -18,114 +22,85 @@ from crewai.utilities.types import LLMMessage
 
 
 if TYPE_CHECKING:
-    from crewai.llms.hooks.base import BaseInterceptor
+    from anthropic.types import Message
+
+    from crewai.agent.core import Agent
+    from crewai.task import Task
+
 
 try:
     from anthropic import Anthropic
-    from anthropic.types import Message
     from anthropic.types.tool_use_block import ToolUseBlock
-    import httpx
 except ImportError:
     raise ImportError(
         'Anthropic native provider not available, to install: uv add "crewai[anthropic]"'
     ) from None
 
 
+load_dotenv()
+
+
 class AnthropicCompletion(BaseLLM):
     """Anthropic native completion implementation.
 
     This class provides direct integration with the Anthropic Python SDK,
     offering native tool use, streaming support, and proper message formatting.
+
+    Attributes:
+        model: Anthropic model name (e.g., 'claude-3-5-sonnet-20241022')
+        base_url: Custom base URL for Anthropic API
+        timeout: Request timeout in seconds
+        max_retries: Maximum number of retries
+        max_tokens: Maximum tokens in response (required for Anthropic)
+        top_p: Nucleus sampling parameter
+        stream: Enable streaming responses
+        client_params: Additional parameters for the Anthropic client
+        interceptor: HTTP interceptor for modifying requests/responses at transport level
     """
 
-    def __init__(
-        self,
-        model: str = "claude-3-5-sonnet-20241022",
-        api_key: str | None = None,
-        base_url: str | None = None,
-        timeout: float | None = None,
-        max_retries: int = 2,
-        temperature: float | None = None,
-        max_tokens: int = 4096,  # Required for Anthropic
-        top_p: float | None = None,
-        stop_sequences: list[str] | None = None,
-        stream: bool = False,
-        client_params: dict[str, Any] | None = None,
-        interceptor: BaseInterceptor[httpx.Request, httpx.Response] | None = None,
-        **kwargs: Any,
-    ):
-        """Initialize Anthropic chat completion client.
+    base_url: str | None = Field(
+        default=None, description="Custom base URL for Anthropic API"
+    )
+    timeout: float | None = Field(
+        default=None, description="Request timeout in seconds"
+    )
+    max_retries: int = Field(default=2, description="Maximum number of retries")
+    max_tokens: int = Field(
+        default=4096, description="Maximum tokens in response (required for Anthropic)"
+    )
+    top_p: float | None = Field(default=None, description="Nucleus sampling parameter")
+    stream: bool = Field(default=False, description="Enable streaming responses")
+    client_params: dict[str, Any] | None = Field(
+        default_factory=dict, description="Additional Anthropic client parameters"
+    )
+    _client: Anthropic = PrivateAttr(default=None)  # type: ignore[assignment]
 
-        Args:
-            model: Anthropic model name (e.g., 'claude-3-5-sonnet-20241022')
-            api_key: Anthropic API key (defaults to ANTHROPIC_API_KEY env var)
-            base_url: Custom base URL for Anthropic API
-            timeout: Request timeout in seconds
-            max_retries: Maximum number of retries
-            temperature: Sampling temperature (0-1)
-            max_tokens: Maximum tokens in response (required for Anthropic)
-            top_p: Nucleus sampling parameter
-            stop_sequences: Stop sequences (Anthropic uses stop_sequences, not stop)
-            stream: Enable streaming responses
-            client_params: Additional parameters for the Anthropic client
-            interceptor: HTTP interceptor for modifying requests/responses at transport level.
-            **kwargs: Additional parameters
-        """
-        super().__init__(
-            model=model, temperature=temperature, stop=stop_sequences or [], **kwargs
-        )
+    _is_claude_3: bool = PrivateAttr(default=False)
+    _supports_tools: bool = PrivateAttr(default=False)
 
-        # Client params
-        self.interceptor = interceptor
-        self.client_params = client_params
-        self.base_url = base_url
-        self.timeout = timeout
-        self.max_retries = max_retries
+    @model_validator(mode="after")
+    def setup_client(self) -> Self:
+        """Initialize the Anthropic client and model-specific settings."""
+        self._client = Anthropic(**self._get_client_params())
 
-        self.client = Anthropic(**self._get_client_params())
+        self._is_claude_3 = "claude-3" in self.model.lower()
+        self._supports_tools = self._is_claude_3
 
-        # Store completion parameters
-        self.max_tokens = max_tokens
-        self.top_p = top_p
-        self.stream = stream
-        self.stop_sequences = stop_sequences or []
-
-        # Model-specific settings
-        self.is_claude_3 = "claude-3" in model.lower()
-        self.supports_tools = self.is_claude_3  # Claude 3+ supports tool use
+        return self
 
     @property
-    def stop(self) -> list[str]:
-        """Get stop sequences sent to the API."""
-        return self.stop_sequences
+    def is_claude_3(self) -> bool:
+        """Check if model is Claude 3."""
+        return self._is_claude_3
 
-    @stop.setter
-    def stop(self, value: list[str] | str | None) -> None:
-        """Set stop sequences.
-
-        Synchronizes stop_sequences to ensure values set by CrewAgentExecutor
-        are properly sent to the Anthropic API.
-
-        Args:
-            value: Stop sequences as a list, single string, or None
-        """
-        if value is None:
-            self.stop_sequences = []
-        elif isinstance(value, str):
-            self.stop_sequences = [value]
-        elif isinstance(value, list):
-            self.stop_sequences = value
-        else:
-            self.stop_sequences = []
+    @property
+    def supports_tools(self) -> bool:
+        """Check if model supports tools."""
+        return self._supports_tools
 
     def _get_client_params(self) -> dict[str, Any]:
         """Get client parameters."""
 
-        if self.api_key is None:
-            self.api_key = os.getenv("ANTHROPIC_API_KEY")
-            if self.api_key is None:
-                raise ValueError("ANTHROPIC_API_KEY is required")
-
         client_params = {
             "api_key": self.api_key,
             "base_url": self.base_url,
@@ -149,8 +124,8 @@ class AnthropicCompletion(BaseLLM):
         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,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
         response_model: type[BaseModel] | None = None,
     ) -> str | Any:
         """Call Anthropic messages API.
@@ -245,8 +220,8 @@ class AnthropicCompletion(BaseLLM):
             params["temperature"] = self.temperature
         if self.top_p is not None:
             params["top_p"] = self.top_p
-        if self.stop_sequences:
-            params["stop_sequences"] = self.stop_sequences
+        if self.stop:
+            params["stop_sequences"] = self.stop
 
         # Handle tools for Claude 3+
         if tools and self.supports_tools:
@@ -266,8 +241,6 @@ class AnthropicCompletion(BaseLLM):
                 continue
 
             try:
-                from crewai.llms.providers.utils.common import safe_tool_conversion
-
                 name, description, parameters = safe_tool_conversion(tool, "Anthropic")
             except (ImportError, KeyError, ValueError) as e:
                 logging.error(f"Error converting tool to Anthropic format: {e}")
@@ -341,8 +314,8 @@ class AnthropicCompletion(BaseLLM):
         self,
         params: dict[str, Any],
         available_functions: dict[str, Any] | None = None,
-        from_task: Any | None = None,
-        from_agent: Any | None = None,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
         response_model: type[BaseModel] | None = None,
     ) -> str | Any:
         """Handle non-streaming message completion."""
@@ -357,7 +330,7 @@ class AnthropicCompletion(BaseLLM):
             params["tool_choice"] = {"type": "tool", "name": "structured_output"}
 
         try:
-            response: Message = self.client.messages.create(**params)
+            response: Message = self._client.messages.create(**params)
 
         except Exception as e:
             if is_context_length_exceeded(e):
@@ -429,8 +402,8 @@ class AnthropicCompletion(BaseLLM):
         self,
         params: dict[str, Any],
         available_functions: dict[str, Any] | None = None,
-        from_task: Any | None = None,
-        from_agent: Any | None = None,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
         response_model: type[BaseModel] | None = None,
     ) -> str:
         """Handle streaming message completion."""
@@ -451,7 +424,7 @@ class AnthropicCompletion(BaseLLM):
         stream_params = {k: v for k, v in params.items() if k != "stream"}
 
         # Make streaming API call
-        with self.client.messages.stream(**stream_params) as stream:
+        with self._client.messages.stream(**stream_params) as stream:
             for event in stream:
                 if hasattr(event, "delta") and hasattr(event.delta, "text"):
                     text_delta = event.delta.text
@@ -525,8 +498,8 @@ class AnthropicCompletion(BaseLLM):
         tool_uses: list[ToolUseBlock],
         params: dict[str, Any],
         available_functions: dict[str, Any],
-        from_task: Any | None = None,
-        from_agent: Any | None = None,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
     ) -> str:
         """Handle the complete tool use conversation flow.
 
@@ -579,7 +552,7 @@ class AnthropicCompletion(BaseLLM):
 
         try:
             # Send tool results back to Claude for final response
-            final_response: Message = self.client.messages.create(**follow_up_params)
+            final_response: Message = self._client.messages.create(**follow_up_params)
 
             # Track token usage for follow-up call
             follow_up_usage = self._extract_anthropic_token_usage(final_response)
@@ -636,7 +609,6 @@ class AnthropicCompletion(BaseLLM):
 
     def get_context_window_size(self) -> int:
         """Get the context window size for the model."""
-        from crewai.llm import CONTEXT_WINDOW_USAGE_RATIO
 
         # Context window sizes for Anthropic models
         context_windows = {

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

@@ -5,8 +5,12 @@ import logging
 import os
 from typing import TYPE_CHECKING, Any
 
-from pydantic import BaseModel
+from dotenv import load_dotenv
+from pydantic import BaseModel, Field, PrivateAttr, model_validator
+from typing_extensions import Self
 
+from crewai.llm.core import CONTEXT_WINDOW_USAGE_RATIO, LLM_CONTEXT_WINDOW_SIZES
+from crewai.llm.providers.utils.common import safe_tool_conversion
 from crewai.utilities.agent_utils import is_context_length_exceeded
 from crewai.utilities.exceptions.context_window_exceeding_exception import (
     LLMContextLengthExceededError,
@@ -15,7 +19,8 @@ from crewai.utilities.types import LLMMessage
 
 
 if TYPE_CHECKING:
-    from crewai.llms.hooks.base import BaseInterceptor
+    from crewai.agent.core import Agent
+    from crewai.task import Task
     from crewai.tools.base_tool import BaseTool
 
 
@@ -36,7 +41,7 @@ try:
     )
 
     from crewai.events.types.llm_events import LLMCallType
-    from crewai.llms.base_llm import BaseLLM
+    from crewai.llm.base_llm import BaseLLM
 
 except ImportError:
     raise ImportError(
@@ -44,111 +49,109 @@ except ImportError:
     ) from None
 
 
+load_dotenv()
+
+
 class AzureCompletion(BaseLLM):
     """Azure AI Inference native completion implementation.
 
     This class provides direct integration with the Azure AI Inference Python SDK,
     offering native function calling, streaming support, and proper Azure authentication.
+
+    Attributes:
+        model: Azure deployment name or model name
+        endpoint: Azure endpoint URL
+        api_version: Azure API version
+        timeout: Request timeout in seconds
+        max_retries: Maximum number of retries
+        top_p: Nucleus sampling parameter
+        frequency_penalty: Frequency penalty (-2 to 2)
+        presence_penalty: Presence penalty (-2 to 2)
+        max_tokens: Maximum tokens in response
+        stream: Enable streaming responses
+        interceptor: HTTP interceptor (not yet supported for Azure)
     """
 
-    def __init__(
-        self,
-        model: str,
-        api_key: str | None = None,
-        endpoint: str | None = None,
-        api_version: str | None = None,
-        timeout: float | None = None,
-        max_retries: int = 2,
-        temperature: float | None = None,
-        top_p: float | None = None,
-        frequency_penalty: float | None = None,
-        presence_penalty: float | None = None,
-        max_tokens: int | None = None,
-        stop: list[str] | None = None,
-        stream: bool = False,
-        interceptor: BaseInterceptor[Any, Any] | None = None,
-        **kwargs: Any,
-    ):
-        """Initialize Azure AI Inference chat completion client.
+    endpoint: str = Field(  # type: ignore[assignment]
+        default_factory=lambda: os.getenv("AZURE_ENDPOINT")
+        or os.getenv("AZURE_OPENAI_ENDPOINT")
+        or os.getenv("AZURE_API_BASE"),
+        description="Azure endpoint URL (defaults to AZURE_ENDPOINT env var)",
+    )
+    api_version: str = Field(
+        default_factory=lambda: os.getenv("AZURE_API_VERSION", "2024-06-01"),
+        description="Azure API version (defaults to AZURE_API_VERSION env var or 2024-06-01)",
+    )
+    timeout: float | None = Field(
+        default=None, description="Request timeout in seconds"
+    )
+    max_retries: int = Field(default=2, description="Maximum number of retries")
+    top_p: float | None = Field(default=None, description="Nucleus sampling parameter")
+    frequency_penalty: float | None = Field(
+        default=None, le=2.0, ge=-2.0, description="Frequency penalty (-2 to 2)"
+    )
+    presence_penalty: float | None = Field(
+        default=None, le=2.0, ge=-2.0, description="Presence penalty (-2 to 2)"
+    )
+    max_tokens: int | None = Field(
+        default=None, description="Maximum tokens in response"
+    )
+    stream: bool = Field(default=False, description="Enable streaming responses")
+    _client: ChatCompletionsClient = PrivateAttr(default=None)  # type: ignore[assignment]
 
-        Args:
-            model: Azure deployment name or model name
-            api_key: Azure API key (defaults to AZURE_API_KEY env var)
-            endpoint: Azure endpoint URL (defaults to AZURE_ENDPOINT env var)
-            api_version: Azure API version (defaults to AZURE_API_VERSION env var)
-            timeout: Request timeout in seconds
-            max_retries: Maximum number of retries
-            temperature: Sampling temperature (0-2)
-            top_p: Nucleus sampling parameter
-            frequency_penalty: Frequency penalty (-2 to 2)
-            presence_penalty: Presence penalty (-2 to 2)
-            max_tokens: Maximum tokens in response
-            stop: Stop sequences
-            stream: Enable streaming responses
-            interceptor: HTTP interceptor (not yet supported for Azure).
-            **kwargs: Additional parameters
-        """
-        if interceptor is not None:
+    _is_openai_model: bool = PrivateAttr(default=False)
+    _is_azure_openai_endpoint: bool = PrivateAttr(default=False)
+
+    @model_validator(mode="after")
+    def setup_client(self) -> Self:
+        """Initialize the Azure client and validate configuration."""
+        if self.interceptor is not None:
             raise NotImplementedError(
                 "HTTP interceptors are not yet supported for Azure AI Inference provider. "
                 "Interceptors are currently supported for OpenAI and Anthropic providers only."
             )
 
-        super().__init__(
-            model=model, temperature=temperature, stop=stop or [], **kwargs
-        )
-
-        self.api_key = api_key or os.getenv("AZURE_API_KEY")
-        self.endpoint = (
-            endpoint
-            or os.getenv("AZURE_ENDPOINT")
-            or os.getenv("AZURE_OPENAI_ENDPOINT")
-            or os.getenv("AZURE_API_BASE")
-        )
-        self.api_version = api_version or os.getenv("AZURE_API_VERSION") or "2024-06-01"
-        self.timeout = timeout
-        self.max_retries = max_retries
+        if not self.api_key:
+            self.api_key = os.getenv("AZURE_API_KEY")
 
         if not self.api_key:
             raise ValueError(
                 "Azure API key is required. Set AZURE_API_KEY environment variable or pass api_key parameter."
             )
-        if not self.endpoint:
-            raise ValueError(
-                "Azure endpoint is required. Set AZURE_ENDPOINT environment variable or pass endpoint parameter."
-            )
 
-        # Validate and potentially fix Azure OpenAI endpoint URL
-        self.endpoint = self._validate_and_fix_endpoint(self.endpoint, model)
+        self.endpoint = self._validate_and_fix_endpoint(self.endpoint, self.model)
 
-        # Build client kwargs
-        client_kwargs = {
+        client_kwargs: dict[str, Any] = {
             "endpoint": self.endpoint,
             "credential": AzureKeyCredential(self.api_key),
         }
 
-        # Add api_version if specified (primarily for Azure OpenAI endpoints)
         if self.api_version:
             client_kwargs["api_version"] = self.api_version
 
-        self.client = ChatCompletionsClient(**client_kwargs)  # type: ignore[arg-type]
+        self._client = ChatCompletionsClient(**client_kwargs)
 
-        self.top_p = top_p
-        self.frequency_penalty = frequency_penalty
-        self.presence_penalty = presence_penalty
-        self.max_tokens = max_tokens
-        self.stream = stream
-
-        self.is_openai_model = any(
-            prefix in model.lower() for prefix in ["gpt-", "o1-", "text-"]
+        self._is_openai_model = any(
+            prefix in self.model.lower() for prefix in ["gpt-", "o1-", "text-"]
         )
-
-        self.is_azure_openai_endpoint = (
+        self._is_azure_openai_endpoint = (
             "openai.azure.com" in self.endpoint
             and "/openai/deployments/" in self.endpoint
         )
 
-    def _validate_and_fix_endpoint(self, endpoint: str, model: str) -> str:
+        return self
+
+    @property
+    def is_openai_model(self) -> bool:
+        """Check if model is an OpenAI model."""
+        return self._is_openai_model
+
+    @property
+    def is_azure_openai_endpoint(self) -> bool:
+        """Check if endpoint is an Azure OpenAI endpoint."""
+        return self._is_azure_openai_endpoint
+
+    def _validate_and_fix_endpoint(self, endpoint: str | None, model: str) -> str:
         """Validate and fix Azure endpoint URL format.
 
         Azure OpenAI endpoints should be in the format:
@@ -160,7 +163,15 @@ class AzureCompletion(BaseLLM):
 
         Returns:
             Validated and potentially corrected endpoint URL
+
+        Raises:
+            ValueError: If endpoint is None or empty
         """
+        if not endpoint:
+            raise ValueError(
+                "Azure endpoint is required. Set AZURE_ENDPOINT environment variable or pass endpoint parameter."
+            )
+
         if "openai.azure.com" in endpoint and "/openai/deployments/" not in endpoint:
             endpoint = endpoint.rstrip("/")
 
@@ -177,8 +188,8 @@ class AzureCompletion(BaseLLM):
         tools: list[dict[str, BaseTool]] | None = None,
         callbacks: list[Any] | None = None,
         available_functions: dict[str, Any] | None = None,
-        from_task: Any | None = None,
-        from_agent: Any | None = None,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
         response_model: type[BaseModel] | None = None,
     ) -> str | Any:
         """Call Azure AI Inference chat completions API.
@@ -317,8 +328,6 @@ class AzureCompletion(BaseLLM):
     ) -> list[dict[str, Any]]:
         """Convert CrewAI tool format to Azure OpenAI function calling format."""
 
-        from crewai.llms.providers.utils.common import safe_tool_conversion
-
         azure_tools = []
 
         for tool in tools:
@@ -371,14 +380,14 @@ class AzureCompletion(BaseLLM):
         self,
         params: dict[str, Any],
         available_functions: dict[str, Any] | None = None,
-        from_task: Any | None = None,
-        from_agent: Any | None = None,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
         response_model: type[BaseModel] | None = None,
     ) -> str | Any:
         """Handle non-streaming chat completion."""
         # Make API call
         try:
-            response: ChatCompletions = self.client.complete(**params)
+            response: ChatCompletions = self._client.complete(**params)
 
             if not response.choices:
                 raise ValueError("No choices returned from Azure API")
@@ -467,8 +476,8 @@ class AzureCompletion(BaseLLM):
         self,
         params: dict[str, Any],
         available_functions: dict[str, Any] | None = None,
-        from_task: Any | None = None,
-        from_agent: Any | None = None,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
         response_model: type[BaseModel] | None = None,
     ) -> str:
         """Handle streaming chat completion."""
@@ -476,7 +485,7 @@ class AzureCompletion(BaseLLM):
         tool_calls = {}
 
         # Make streaming API call
-        for update in self.client.complete(**params):
+        for update in self._client.complete(**params):
             if isinstance(update, StreamingChatCompletionsUpdate):
                 if update.choices:
                     choice = update.choices[0]
@@ -554,7 +563,6 @@ class AzureCompletion(BaseLLM):
 
     def get_context_window_size(self) -> int:
         """Get the context window size for the model."""
-        from crewai.llm import CONTEXT_WINDOW_USAGE_RATIO, LLM_CONTEXT_WINDOW_SIZES
 
         min_context = 1024
         max_context = 2097152

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

@@ -5,11 +5,15 @@ import logging
 import os
 from typing import TYPE_CHECKING, Any, TypedDict, cast
 
-from pydantic import BaseModel
-from typing_extensions import Required
+from dotenv import load_dotenv
+from mypy_boto3_bedrock_runtime.client import BedrockRuntimeClient
+from pydantic import BaseModel, Field, PrivateAttr, model_validator
+from typing_extensions import Required, Self
 
 from crewai.events.types.llm_events import LLMCallType
-from crewai.llms.base_llm import BaseLLM
+from crewai.llm.base_llm import BaseLLM
+from crewai.llm.core import CONTEXT_WINDOW_USAGE_RATIO
+from crewai.llm.providers.utils.common import safe_tool_conversion
 from crewai.utilities.agent_utils import is_context_length_exceeded
 from crewai.utilities.exceptions.context_window_exceeding_exception import (
     LLMContextLengthExceededError,
@@ -30,7 +34,8 @@ if TYPE_CHECKING:
         ToolTypeDef,
     )
 
-    from crewai.llms.hooks.base import BaseInterceptor
+    from crewai.agent.core import Agent
+    from crewai.task import Task
 
 
 try:
@@ -72,6 +77,9 @@ else:
         topK: int
 
 
+load_dotenv()
+
+
 class ToolInputSchema(TypedDict):
     """Type definition for tool input schema in Converse API."""
 
@@ -141,74 +149,84 @@ class BedrockCompletion(BaseLLM):
     - Complete streaming event handling (messageStart, contentBlockStart, etc.)
     - Response metadata and trace information capture
     - Model-specific conversation format handling (e.g., Cohere requirements)
+
+    Attributes:
+        model: The Bedrock model ID to use
+        aws_access_key_id: AWS access key (defaults to environment variable)
+        aws_secret_access_key: AWS secret key (defaults to environment variable)
+        aws_session_token: AWS session token for temporary credentials
+        region_name: AWS region name
+        max_tokens: Maximum tokens to generate
+        top_p: Nucleus sampling parameter
+        top_k: Top-k sampling parameter (Claude models only)
+        stream: Whether to use streaming responses
+        guardrail_config: Guardrail configuration for content filtering
+        additional_model_request_fields: Model-specific request parameters
+        additional_model_response_field_paths: Custom response field paths
+        interceptor: HTTP interceptor (not yet supported for Bedrock)
     """
 
-    def __init__(
-        self,
-        model: str = "anthropic.claude-3-5-sonnet-20241022-v2:0",
-        aws_access_key_id: str | None = None,
-        aws_secret_access_key: str | None = None,
-        aws_session_token: str | None = None,
-        region_name: str = "us-east-1",
-        temperature: float | None = None,
-        max_tokens: int | None = None,
-        top_p: float | None = None,
-        top_k: int | None = None,
-        stop_sequences: Sequence[str] | None = None,
-        stream: bool = False,
-        guardrail_config: dict[str, Any] | None = None,
-        additional_model_request_fields: dict[str, Any] | None = None,
-        additional_model_response_field_paths: list[str] | None = None,
-        interceptor: BaseInterceptor[Any, Any] | None = None,
-        **kwargs: Any,
-    ) -> None:
-        """Initialize AWS Bedrock completion client.
+    aws_access_key_id: str = Field(  # type: ignore[assignment]
+        default_factory=lambda: os.getenv("AWS_ACCESS_KEY_ID"),
+        description="AWS access key (defaults to environment variable)",
+    )
+    aws_secret_access_key: str = Field(  # type: ignore[assignment]
+        default_factory=lambda: os.getenv("AWS_SECRET_ACCESS_KEY"),
+        description="AWS secret key (defaults to environment variable)",
+    )
+    aws_session_token: str = Field(  # type: ignore[assignment]
+        default_factory=lambda: os.getenv("AWS_SESSION_TOKEN"),
+        description="AWS session token for temporary credentials",
+    )
+    region_name: str = Field(
+        default_factory=lambda: os.getenv("AWS_REGION", "us-east-1"),
+        description="AWS region name",
+    )
+    max_tokens: int | None = Field(
+        default=None, description="Maximum tokens to generate"
+    )
+    top_p: float | None = Field(default=None, description="Nucleus sampling parameter")
+    top_k: int | None = Field(
+        default=None, description="Top-k sampling parameter (Claude models only)"
+    )
+    stream: bool = Field(
+        default=False, description="Whether to use streaming responses"
+    )
+    guardrail_config: dict[str, Any] = Field(
+        default_factory=dict,
+        description="Guardrail configuration for content filtering",
+    )
+    additional_model_request_fields: dict[str, Any] = Field(
+        default_factory=dict, description="Model-specific request parameters"
+    )
+    additional_model_response_field_paths: list[str] = Field(
+        default_factory=list, description="Custom response field paths"
+    )
+    _client: BedrockRuntimeClient = PrivateAttr(  # type: ignore[assignment]
+        default_factory=lambda: Session().client,
+    )
 
-        Args:
-            model: The Bedrock model ID to use
-            aws_access_key_id: AWS access key (defaults to environment variable)
-            aws_secret_access_key: AWS secret key (defaults to environment variable)
-            aws_session_token: AWS session token for temporary credentials
-            region_name: AWS region name
-            temperature: Sampling temperature for response generation
-            max_tokens: Maximum tokens to generate
-            top_p: Nucleus sampling parameter
-            top_k: Top-k sampling parameter (Claude models only)
-            stop_sequences: List of sequences that stop generation
-            stream: Whether to use streaming responses
-            guardrail_config: Guardrail configuration for content filtering
-            additional_model_request_fields: Model-specific request parameters
-            additional_model_response_field_paths: Custom response field paths
-            interceptor: HTTP interceptor (not yet supported for Bedrock).
-            **kwargs: Additional parameters
-        """
-        if interceptor is not None:
+    _is_claude_model: bool = PrivateAttr(default=False)
+    _supports_tools: bool = PrivateAttr(default=True)
+    _supports_streaming: bool = PrivateAttr(default=True)
+    _model_id: str = PrivateAttr()
+
+    @model_validator(mode="after")
+    def setup_client(self) -> Self:
+        """Initialize the Bedrock client and validate configuration."""
+        if self.interceptor is not None:
             raise NotImplementedError(
                 "HTTP interceptors are not yet supported for AWS Bedrock provider. "
                 "Interceptors are currently supported for OpenAI and Anthropic providers only."
             )
 
-        # Extract provider from kwargs to avoid duplicate argument
-        kwargs.pop("provider", None)
-
-        super().__init__(
-            model=model,
-            temperature=temperature,
-            stop=stop_sequences or [],
-            provider="bedrock",
-            **kwargs,
-        )
-
-        # Initialize Bedrock client with proper configuration
         session = Session(
-            aws_access_key_id=aws_access_key_id or os.getenv("AWS_ACCESS_KEY_ID"),
-            aws_secret_access_key=aws_secret_access_key
-            or os.getenv("AWS_SECRET_ACCESS_KEY"),
-            aws_session_token=aws_session_token or os.getenv("AWS_SESSION_TOKEN"),
-            region_name=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,
+            region_name=self.region_name,
         )
 
-        # Configure client with timeouts and retries following AWS best practices
         config = Config(
             read_timeout=300,
             retries={
@@ -218,54 +236,34 @@ class BedrockCompletion(BaseLLM):
             tcp_keepalive=True,
         )
 
-        self.client = session.client("bedrock-runtime", config=config)
-        self.region_name = region_name
+        self._client = session.client("bedrock-runtime", config=config)
 
-        # Store completion parameters
-        self.max_tokens = max_tokens
-        self.top_p = top_p
-        self.top_k = top_k
-        self.stream = stream
-        self.stop_sequences = stop_sequences or []
+        self._is_claude_model = "claude" in self.model.lower()
+        self._supports_tools = True
+        self._supports_streaming = True
+        self._model_id = self.model
 
-        # Store advanced features (optional)
-        self.guardrail_config = guardrail_config
-        self.additional_model_request_fields = additional_model_request_fields
-        self.additional_model_response_field_paths = (
-            additional_model_response_field_paths
-        )
-
-        # Model-specific settings
-        self.is_claude_model = "claude" in model.lower()
-        self.supports_tools = True  # Converse API supports tools for most models
-        self.supports_streaming = True
-
-        # Handle inference profiles for newer models
-        self.model_id = model
+        return self
 
     @property
-    def stop(self) -> list[str]:
-        """Get stop sequences sent to the API."""
-        return list(self.stop_sequences)
+    def is_claude_model(self) -> bool:
+        """Check if model is a Claude model."""
+        return self._is_claude_model
 
-    @stop.setter
-    def stop(self, value: Sequence[str] | str | None) -> None:
-        """Set stop sequences.
+    @property
+    def supports_tools(self) -> bool:
+        """Check if model supports tools."""
+        return self._supports_tools
 
-        Synchronizes stop_sequences to ensure values set by CrewAgentExecutor
-        are properly sent to the Bedrock API.
+    @property
+    def supports_streaming(self) -> bool:
+        """Check if model supports streaming."""
+        return self._supports_streaming
 
-        Args:
-            value: Stop sequences as a Sequence, single string, or None
-        """
-        if value is None:
-            self.stop_sequences = []
-        elif isinstance(value, str):
-            self.stop_sequences = [value]
-        elif isinstance(value, Sequence):
-            self.stop_sequences = list(value)
-        else:
-            self.stop_sequences = []
+    @property
+    def model_id(self) -> str:
+        """Get the model ID."""
+        return self._model_id
 
     def call(
         self,
@@ -273,8 +271,8 @@ class BedrockCompletion(BaseLLM):
         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,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
         response_model: type[BaseModel] | None = None,
     ) -> str | Any:
         """Call AWS Bedrock Converse API."""
@@ -359,8 +357,8 @@ class BedrockCompletion(BaseLLM):
         messages: list[dict[str, Any]],
         body: BedrockConverseRequestBody,
         available_functions: Mapping[str, Any] | None = None,
-        from_task: Any | None = None,
-        from_agent: Any | None = None,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
     ) -> str:
         """Handle non-streaming converse API call following AWS best practices."""
         try:
@@ -378,7 +376,7 @@ class BedrockCompletion(BaseLLM):
                     raise ValueError(f"Invalid message format at index {i}")
 
             # Call Bedrock Converse API with proper error handling
-            response = self.client.converse(
+            response = self._client.converse(
                 modelId=self.model_id,
                 messages=cast(
                     "Sequence[MessageTypeDef | MessageOutputTypeDef]",
@@ -540,8 +538,8 @@ class BedrockCompletion(BaseLLM):
         messages: list[dict[str, Any]],
         body: BedrockConverseRequestBody,
         available_functions: dict[str, Any] | None = None,
-        from_task: Any | None = None,
-        from_agent: Any | None = None,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
     ) -> str:
         """Handle streaming converse API call with comprehensive event handling."""
         full_response = ""
@@ -549,7 +547,7 @@ class BedrockCompletion(BaseLLM):
         tool_use_id = None
 
         try:
-            response = self.client.converse_stream(
+            response = self._client.converse_stream(
                 modelId=self.model_id,
                 messages=cast(
                     "Sequence[MessageTypeDef | MessageOutputTypeDef]",
@@ -778,7 +776,6 @@ class BedrockCompletion(BaseLLM):
         tools: list[dict[str, Any]],
     ) -> list[ConverseToolTypeDef]:
         """Convert CrewAI tools to Converse API format following AWS specification."""
-        from crewai.llms.providers.utils.common import safe_tool_conversion
 
         converse_tools: list[ConverseToolTypeDef] = []
 
@@ -818,8 +815,8 @@ class BedrockCompletion(BaseLLM):
             config["temperature"] = float(self.temperature)
         if self.top_p is not None:
             config["topP"] = float(self.top_p)
-        if self.stop_sequences:
-            config["stopSequences"] = self.stop_sequences
+        if self.stop:
+            config["stopSequences"] = self.stop
 
         if self.is_claude_model and self.top_k is not None:
             # top_k is supported by Claude models
@@ -871,7 +868,6 @@ class BedrockCompletion(BaseLLM):
 
     def get_context_window_size(self) -> int:
         """Get the context window size for the model."""
-        from crewai.llm import CONTEXT_WINDOW_USAGE_RATIO
 
         # Context window sizes for common Bedrock models
         context_windows = {

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

@@ -1,12 +1,17 @@
+from __future__ import annotations
+
 import logging
 import os
-from typing import Any, cast
+from typing import TYPE_CHECKING, Any, cast
 
-from pydantic import BaseModel
+from dotenv import load_dotenv
+from pydantic import BaseModel, Field, PrivateAttr, model_validator
+from typing_extensions import Self
 
 from crewai.events.types.llm_events import LLMCallType
-from crewai.llms.base_llm import BaseLLM
-from crewai.llms.hooks.base import BaseInterceptor
+from crewai.llm.base_llm import BaseLLM
+from crewai.llm.core import CONTEXT_WINDOW_USAGE_RATIO, LLM_CONTEXT_WINDOW_SIZES
+from crewai.llm.providers.utils.common import safe_tool_conversion
 from crewai.utilities.agent_utils import is_context_length_exceeded
 from crewai.utilities.exceptions.context_window_exceeding_exception import (
     LLMContextLengthExceededError,
@@ -14,6 +19,11 @@ from crewai.utilities.exceptions.context_window_exceeding_exception import (
 from crewai.utilities.types import LLMMessage
 
 
+if TYPE_CHECKING:
+    from crewai.agent.core import Agent
+    from crewai.task import Task
+
+
 try:
     from google import genai  # type: ignore[import-untyped]
     from google.genai import types  # type: ignore[import-untyped]
@@ -24,111 +34,93 @@ except ImportError:
     ) from None
 
 
+load_dotenv()
+
+
 class GeminiCompletion(BaseLLM):
     """Google Gemini native completion implementation.
 
     This class provides direct integration with the Google Gen AI Python SDK,
     offering native function calling, streaming support, and proper Gemini formatting.
+
+    Attributes:
+        model: Gemini model name (e.g., 'gemini-2.0-flash-001', 'gemini-1.5-pro')
+        project: Google Cloud project ID (for Vertex AI)
+        location: Google Cloud location (for Vertex AI, defaults to 'us-central1')
+        top_p: Nucleus sampling parameter
+        top_k: Top-k sampling parameter
+        max_output_tokens: Maximum tokens in response
+        stream: Enable streaming responses
+        safety_settings: Safety filter settings
+        client_params: Additional parameters for Google Gen AI Client constructor
+        interceptor: HTTP interceptor (not yet supported for Gemini)
     """
 
-    def __init__(
-        self,
-        model: str = "gemini-2.0-flash-001",
-        api_key: str | None = None,
-        project: str | None = None,
-        location: str | None = None,
-        temperature: float | None = None,
-        top_p: float | None = None,
-        top_k: int | None = None,
-        max_output_tokens: int | None = None,
-        stop_sequences: list[str] | None = None,
-        stream: bool = False,
-        safety_settings: dict[str, Any] | None = None,
-        client_params: dict[str, Any] | None = None,
-        interceptor: BaseInterceptor[Any, Any] | None = None,
-        **kwargs: Any,
-    ):
-        """Initialize Google Gemini chat completion client.
+    project: str | None = Field(
+        default_factory=lambda: os.getenv("GOOGLE_CLOUD_PROJECT"),
+        description="Google Cloud project ID (for Vertex AI)",
+    )
+    location: str = Field(
+        default_factory=lambda: os.getenv("GOOGLE_CLOUD_LOCATION", "us-central1"),
+        description="Google Cloud location (for Vertex AI, defaults to 'us-central1')",
+    )
+    top_p: float | None = Field(default=None, description="Nucleus sampling parameter")
+    top_k: int | None = Field(default=None, description="Top-k sampling parameter")
+    max_output_tokens: int | None = Field(
+        default=None, description="Maximum tokens in response"
+    )
+    stream: bool = Field(default=False, description="Enable streaming responses")
+    safety_settings: dict[str, Any] = Field(
+        default_factory=dict, description="Safety filter settings"
+    )
+    client_params: dict[str, Any] = Field(
+        default_factory=dict,
+        description="Additional parameters for Google Gen AI Client constructor",
+    )
+    _client: Any = PrivateAttr(default=None)
 
-        Args:
-            model: Gemini model name (e.g., 'gemini-2.0-flash-001', 'gemini-1.5-pro')
-            api_key: Google API key (defaults to GOOGLE_API_KEY or GEMINI_API_KEY env var)
-            project: Google Cloud project ID (for Vertex AI)
-            location: Google Cloud location (for Vertex AI, defaults to 'us-central1')
-            temperature: Sampling temperature (0-2)
-            top_p: Nucleus sampling parameter
-            top_k: Top-k sampling parameter
-            max_output_tokens: Maximum tokens in response
-            stop_sequences: Stop sequences
-            stream: Enable streaming responses
-            safety_settings: Safety filter settings
-            client_params: Additional parameters to pass to the Google Gen AI Client constructor.
-                          Supports parameters like http_options, credentials, debug_config, etc.
-            interceptor: HTTP interceptor (not yet supported for Gemini).
-            **kwargs: Additional parameters
-        """
-        if interceptor is not None:
+    _is_gemini_2: bool = PrivateAttr(default=False)
+    _is_gemini_1_5: bool = PrivateAttr(default=False)
+    _supports_tools: bool = PrivateAttr(default=False)
+
+    @model_validator(mode="after")
+    def setup_client(self) -> Self:
+        """Initialize the Gemini client and validate configuration."""
+        if self.interceptor is not None:
             raise NotImplementedError(
                 "HTTP interceptors are not yet supported for Google Gemini provider. "
                 "Interceptors are currently supported for OpenAI and Anthropic providers only."
             )
 
-        super().__init__(
-            model=model, temperature=temperature, stop=stop_sequences or [], **kwargs
-        )
-
-        # Store client params for later use
-        self.client_params = client_params or {}
-
-        # Get API configuration with environment variable fallbacks
-        self.api_key = (
-            api_key or os.getenv("GOOGLE_API_KEY") or os.getenv("GEMINI_API_KEY")
-        )
-        self.project = project or os.getenv("GOOGLE_CLOUD_PROJECT")
-        self.location = location or os.getenv("GOOGLE_CLOUD_LOCATION") or "us-central1"
+        if self.api_key is None:
+            self.api_key = os.getenv("GOOGLE_API_KEY") or os.getenv("GEMINI_API_KEY")
 
         use_vertexai = os.getenv("GOOGLE_GENAI_USE_VERTEXAI", "").lower() == "true"
 
-        self.client = self._initialize_client(use_vertexai)
+        self._client = self._initialize_client(use_vertexai)
 
-        # Store completion parameters
-        self.top_p = top_p
-        self.top_k = top_k
-        self.max_output_tokens = max_output_tokens
-        self.stream = stream
-        self.safety_settings = safety_settings or {}
-        self.stop_sequences = stop_sequences or []
+        self._is_gemini_2 = "gemini-2" in self.model.lower()
+        self._is_gemini_1_5 = "gemini-1.5" in self.model.lower()
+        self._supports_tools = self._is_gemini_1_5 or self._is_gemini_2
 
-        # Model-specific settings
-        self.is_gemini_2 = "gemini-2" in model.lower()
-        self.is_gemini_1_5 = "gemini-1.5" in model.lower()
-        self.supports_tools = self.is_gemini_1_5 or self.is_gemini_2
+        return self
 
     @property
-    def stop(self) -> list[str]:
-        """Get stop sequences sent to the API."""
-        return self.stop_sequences
+    def is_gemini_2(self) -> bool:
+        """Check if model is Gemini 2."""
+        return self._is_gemini_2
 
-    @stop.setter
-    def stop(self, value: list[str] | str | None) -> None:
-        """Set stop sequences.
+    @property
+    def is_gemini_1_5(self) -> bool:
+        """Check if model is Gemini 1.5."""
+        return self._is_gemini_1_5
 
-        Synchronizes stop_sequences to ensure values set by CrewAgentExecutor
-        are properly sent to the Gemini API.
+    @property
+    def supports_tools(self) -> bool:
+        """Check if model supports tools."""
+        return self._supports_tools
 
-        Args:
-            value: Stop sequences as a list, single string, or None
-        """
-        if value is None:
-            self.stop_sequences = []
-        elif isinstance(value, str):
-            self.stop_sequences = [value]
-        elif isinstance(value, list):
-            self.stop_sequences = value
-        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) -> Any:
         """Initialize the Google Gen AI client with proper parameter handling.
 
         Args:
@@ -150,12 +142,9 @@ class GeminiCompletion(BaseLLM):
                     "location": self.location,
                 }
             )
-
             client_params.pop("api_key", None)
-
         elif self.api_key:
             client_params["api_key"] = self.api_key
-
             client_params.pop("vertexai", None)
             client_params.pop("project", None)
             client_params.pop("location", None)
@@ -180,11 +169,10 @@ class GeminiCompletion(BaseLLM):
         params = {}
 
         if (
-            hasattr(self, "client")
-            and hasattr(self.client, "vertexai")
-            and self.client.vertexai
+            hasattr(self, "_client")
+            and hasattr(self._client, "vertexai")
+            and self._client.vertexai
         ):
-            # Vertex AI configuration
             params.update(
                 {
                     "vertexai": True,
@@ -206,8 +194,8 @@ class GeminiCompletion(BaseLLM):
         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,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
         response_model: type[BaseModel] | None = None,
     ) -> str | Any:
         """Call Google Gemini generate content API.
@@ -296,15 +284,12 @@ class GeminiCompletion(BaseLLM):
         self.tools = tools
         config_params = {}
 
-        # Add system instruction if present
         if system_instruction:
-            # Convert system instruction to Content format
             system_content = types.Content(
                 role="user", parts=[types.Part.from_text(text=system_instruction)]
             )
             config_params["system_instruction"] = system_content
 
-        # Add generation config parameters
         if self.temperature is not None:
             config_params["temperature"] = self.temperature
         if self.top_p is not None:
@@ -313,14 +298,13 @@ class GeminiCompletion(BaseLLM):
             config_params["top_k"] = self.top_k
         if self.max_output_tokens is not None:
             config_params["max_output_tokens"] = self.max_output_tokens
-        if self.stop_sequences:
-            config_params["stop_sequences"] = self.stop_sequences
+        if self.stop:
+            config_params["stop_sequences"] = self.stop
 
         if response_model:
             config_params["response_mime_type"] = "application/json"
             config_params["response_schema"] = response_model.model_json_schema()
 
-        # Handle tools for supported models
         if tools and self.supports_tools:
             config_params["tools"] = self._convert_tools_for_interference(tools)
 
@@ -335,8 +319,6 @@ class GeminiCompletion(BaseLLM):
         """Convert CrewAI tool format to Gemini function declaration format."""
         gemini_tools = []
 
-        from crewai.llms.providers.utils.common import safe_tool_conversion
-
         for tool in tools:
             name, description, parameters = safe_tool_conversion(tool, "Gemini")
 
@@ -345,7 +327,6 @@ class GeminiCompletion(BaseLLM):
                 description=description,
             )
 
-            # Add parameters if present - ensure parameters is a dict
             if parameters and isinstance(parameters, dict):
                 function_declaration.parameters = parameters
 
@@ -381,16 +362,12 @@ class GeminiCompletion(BaseLLM):
             content = message.get("content", "")
 
             if role == "system":
-                # Extract system instruction - Gemini handles it separately
                 if system_instruction:
                     system_instruction += f"\n\n{content}"
                 else:
                     system_instruction = cast(str, 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)]
                 )
@@ -404,8 +381,8 @@ class GeminiCompletion(BaseLLM):
         system_instruction: str | None,
         config: types.GenerateContentConfig,
         available_functions: dict[str, Any] | None = None,
-        from_task: Any | None = None,
-        from_agent: Any | None = None,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
         response_model: type[BaseModel] | None = None,
     ) -> str | Any:
         """Handle non-streaming content generation."""
@@ -416,7 +393,7 @@ class GeminiCompletion(BaseLLM):
         }
 
         try:
-            response = self.client.models.generate_content(**api_params)
+            response = self._client.models.generate_content(**api_params)
 
             usage = self._extract_token_usage(response)
         except Exception as e:
@@ -470,8 +447,8 @@ class GeminiCompletion(BaseLLM):
         contents: list[types.Content],
         config: types.GenerateContentConfig,
         available_functions: dict[str, Any] | None = None,
-        from_task: Any | None = None,
-        from_agent: Any | None = None,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
         response_model: type[BaseModel] | None = None,
     ) -> str:
         """Handle streaming content generation."""
@@ -484,7 +461,7 @@ class GeminiCompletion(BaseLLM):
             "config": config,
         }
 
-        for chunk in self.client.models.generate_content_stream(**api_params):
+        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(
@@ -507,13 +484,11 @@ class GeminiCompletion(BaseLLM):
                                     else {},
                                 }
 
-        # Handle completed function calls
         if function_calls and available_functions:
             for call_data in function_calls.values():
                 function_name = call_data["name"]
                 function_args = call_data["args"]
 
-                # Execute tool
                 result = self._handle_tool_execution(
                     function_name=function_name,
                     function_args=function_args,
@@ -547,7 +522,6 @@ class GeminiCompletion(BaseLLM):
 
     def get_context_window_size(self) -> int:
         """Get the context window size for the model."""
-        from crewai.llm import CONTEXT_WINDOW_USAGE_RATIO, LLM_CONTEXT_WINDOW_SIZES
 
         min_context = 1024
         max_context = 2097152
@@ -574,13 +548,11 @@ class GeminiCompletion(BaseLLM):
             "gemma-3-27b": 128000,
         }
 
-        # Find the best match for the model name
         for model_prefix, size in context_windows.items():
             if self.model.startswith(model_prefix):
                 return int(size * CONTEXT_WINDOW_USAGE_RATIO)
 
-        # Default context window size for Gemini models
-        return int(1048576 * CONTEXT_WINDOW_USAGE_RATIO)  # 1M tokens
+        return int(1048576 * CONTEXT_WINDOW_USAGE_RATIO)
 
     def _extract_token_usage(self, response: dict[str, Any]) -> dict[str, Any]:
         """Extract token usage from Gemini response."""

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

@@ -6,16 +6,20 @@ import logging
 import os
 from typing import TYPE_CHECKING, Any
 
+from dotenv import load_dotenv
 import httpx
 from openai import APIConnectionError, NotFoundError, OpenAI
 from openai.types.chat import ChatCompletion, ChatCompletionChunk
 from openai.types.chat.chat_completion import Choice
 from openai.types.chat.chat_completion_chunk import ChoiceDelta
-from pydantic import BaseModel
+from pydantic import BaseModel, Field, PrivateAttr, model_validator
+from typing_extensions import Self
 
 from crewai.events.types.llm_events import LLMCallType
-from crewai.llms.base_llm import BaseLLM
-from crewai.llms.hooks.transport import HTTPTransport
+from crewai.llm.base_llm import BaseLLM
+from crewai.llm.core import CONTEXT_WINDOW_USAGE_RATIO, LLM_CONTEXT_WINDOW_SIZES
+from crewai.llm.hooks.transport import HTTPTransport
+from crewai.llm.providers.utils.common import safe_tool_conversion
 from crewai.utilities.agent_utils import is_context_length_exceeded
 from crewai.utilities.exceptions.context_window_exceeding_exception import (
     LLMContextLengthExceededError,
@@ -25,11 +29,13 @@ from crewai.utilities.types import LLMMessage
 
 if TYPE_CHECKING:
     from crewai.agent.core import Agent
-    from crewai.llms.hooks.base import BaseInterceptor
     from crewai.task import Task
     from crewai.tools.base_tool import BaseTool
 
 
+load_dotenv()
+
+
 class OpenAICompletion(BaseLLM):
     """OpenAI native completion implementation.
 
@@ -37,60 +43,56 @@ class OpenAICompletion(BaseLLM):
     offering native structured outputs, function calling, and streaming support.
     """
 
-    def __init__(
-        self,
-        model: str = "gpt-4o",
-        api_key: str | None = None,
-        base_url: str | None = None,
-        organization: str | None = None,
-        project: str | None = None,
-        timeout: float | None = None,
-        max_retries: int = 2,
-        default_headers: dict[str, str] | None = None,
-        default_query: dict[str, Any] | None = None,
-        client_params: dict[str, Any] | None = None,
-        temperature: float | None = None,
-        top_p: float | None = None,
-        frequency_penalty: float | None = None,
-        presence_penalty: float | None = None,
-        max_tokens: int | None = None,
-        max_completion_tokens: int | None = None,
-        seed: int | None = None,
-        stream: bool = False,
-        response_format: dict[str, Any] | type[BaseModel] | None = None,
-        logprobs: bool | None = None,
-        top_logprobs: int | None = None,
-        reasoning_effort: str | None = None,
-        provider: str | None = None,
-        interceptor: BaseInterceptor[httpx.Request, httpx.Response] | None = None,
-        **kwargs: Any,
-    ) -> None:
-        """Initialize OpenAI chat completion client."""
+    # Client configuration fields
+    organization: str | None = Field(default=None, description="OpenAI organization ID")
+    project: str | None = Field(default=None, description="OpenAI project ID")
+    max_retries: int = Field(default=2, description="Maximum number of retries")
+    default_headers: dict[str, str] = Field(
+        default_factory=dict, description="Default headers for requests"
+    )
+    default_query: dict[str, Any] = Field(
+        default_factory=dict, description="Default query parameters"
+    )
+    client_params: dict[str, Any] = Field(
+        default_factory=dict, description="Additional client parameters"
+    )
+    timeout: float | None = Field(default=None, description="Request timeout")
+    api_base: str | None = Field(
+        default=None, description="API base URL", deprecated=True
+    )
 
-        if provider is None:
-            provider = kwargs.pop("provider", "openai")
+    # Completion parameters
+    top_p: float | None = Field(default=None, description="Top-p sampling parameter")
+    frequency_penalty: float | None = Field(
+        default=None, description="Frequency penalty"
+    )
+    presence_penalty: float | None = Field(default=None, description="Presence penalty")
+    max_tokens: int | None = Field(default=None, description="Maximum tokens")
+    max_completion_tokens: int | None = Field(
+        None, description="Maximum completion tokens"
+    )
+    seed: int | None = Field(default=None, description="Random seed")
+    stream: bool = Field(default=False, description="Enable streaming")
+    response_format: dict[str, Any] | type[BaseModel] | None = Field(
+        default=None, description="Response format"
+    )
+    logprobs: bool | None = Field(default=None, description="Return log probabilities")
+    top_logprobs: int | None = Field(
+        default=None, description="Number of top log probabilities"
+    )
+    reasoning_effort: str | None = Field(
+        default=None, description="Reasoning effort level"
+    )
 
-        self.interceptor = interceptor
-        # Client configuration attributes
-        self.organization = organization
-        self.project = project
-        self.max_retries = max_retries
-        self.default_headers = default_headers
-        self.default_query = default_query
-        self.client_params = client_params
-        self.timeout = timeout
-        self.base_url = base_url
-        self.api_base = kwargs.pop("api_base", None)
+    _client: OpenAI = PrivateAttr(default=None)  # type: ignore[assignment]
+    is_o1_model: bool = Field(default=False, description="Whether this is an O1 model")
+    is_gpt4_model: bool = Field(
+        default=False, description="Whether this is a GPT-4 model"
+    )
 
-        super().__init__(
-            model=model,
-            temperature=temperature,
-            api_key=api_key or os.getenv("OPENAI_API_KEY"),
-            base_url=base_url,
-            timeout=timeout,
-            provider=provider,
-            **kwargs,
-        )
+    @model_validator(mode="after")
+    def setup_client(self) -> Self:
+        """Initialize OpenAI client after model validation."""
 
         client_config = self._get_client_params()
         if self.interceptor:
@@ -98,31 +100,15 @@ class OpenAICompletion(BaseLLM):
             http_client = httpx.Client(transport=transport)
             client_config["http_client"] = http_client
 
-        self.client = OpenAI(**client_config)
+        self._client = OpenAI(**client_config)
 
-        # Completion parameters
-        self.top_p = top_p
-        self.frequency_penalty = frequency_penalty
-        self.presence_penalty = presence_penalty
-        self.max_tokens = max_tokens
-        self.max_completion_tokens = max_completion_tokens
-        self.seed = seed
-        self.stream = stream
-        self.response_format = response_format
-        self.logprobs = logprobs
-        self.top_logprobs = top_logprobs
-        self.reasoning_effort = reasoning_effort
-        self.is_o1_model = "o1" in model.lower()
-        self.is_gpt4_model = "gpt-4" in model.lower()
+        self.is_o1_model = "o1" in self.model.lower()
+        self.is_gpt4_model = "gpt-4" in self.model.lower()
+
+        return self
 
     def _get_client_params(self) -> dict[str, Any]:
         """Get OpenAI client parameters."""
-
-        if self.api_key is None:
-            self.api_key = os.getenv("OPENAI_API_KEY")
-            if self.api_key is None:
-                raise ValueError("OPENAI_API_KEY is required")
-
         base_params = {
             "api_key": self.api_key,
             "organization": self.organization,
@@ -268,7 +254,6 @@ class OpenAICompletion(BaseLLM):
         self, tools: list[dict[str, BaseTool]]
     ) -> list[dict[str, Any]]:
         """Convert CrewAI tool format to OpenAI function calling format."""
-        from crewai.llms.providers.utils.common import safe_tool_conversion
 
         openai_tools = []
 
@@ -296,14 +281,14 @@ class OpenAICompletion(BaseLLM):
         self,
         params: dict[str, Any],
         available_functions: dict[str, Any] | None = None,
-        from_task: Any | None = None,
-        from_agent: Any | None = None,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
         response_model: type[BaseModel] | None = None,
     ) -> str | Any:
         """Handle non-streaming chat completion."""
         try:
             if response_model:
-                parsed_response = self.client.beta.chat.completions.parse(
+                parsed_response = self._client.beta.chat.completions.parse(
                     **params,
                     response_format=response_model,
                 )
@@ -327,7 +312,7 @@ class OpenAICompletion(BaseLLM):
                     )
                     return structured_json
 
-            response: ChatCompletion = self.client.chat.completions.create(**params)
+            response: ChatCompletion = self._client.chat.completions.create(**params)
 
             usage = self._extract_openai_token_usage(response)
 
@@ -419,8 +404,8 @@ class OpenAICompletion(BaseLLM):
         self,
         params: dict[str, Any],
         available_functions: dict[str, Any] | None = None,
-        from_task: Any | None = None,
-        from_agent: Any | None = None,
+        from_task: Task | None = None,
+        from_agent: Agent | None = None,
         response_model: type[BaseModel] | None = None,
     ) -> str:
         """Handle streaming chat completion."""
@@ -429,7 +414,7 @@ class OpenAICompletion(BaseLLM):
 
         if response_model:
             completion_stream: Iterator[ChatCompletionChunk] = (
-                self.client.chat.completions.create(**params)
+                self._client.chat.completions.create(**params)
             )
 
             accumulated_content = ""
@@ -472,7 +457,7 @@ class OpenAICompletion(BaseLLM):
                 )
                 return accumulated_content
 
-        stream: Iterator[ChatCompletionChunk] = self.client.chat.completions.create(
+        stream: Iterator[ChatCompletionChunk] = self._client.chat.completions.create(
             **params
         )
 
@@ -560,7 +545,6 @@ class OpenAICompletion(BaseLLM):
 
     def get_context_window_size(self) -> int:
         """Get the context window size for the model."""
-        from crewai.llm import CONTEXT_WINDOW_USAGE_RATIO, LLM_CONTEXT_WINDOW_SIZES
 
         min_context = 1024
         max_context = 2097152

lib/crewai/src/crewai/llms/__init__.py

@@ -1 +1,38 @@
-"""LLM implementations for crewAI."""
+"""LLM implementations for crewAI.
+
+.. deprecated:: 1.4.0
+    The `crewai.llms` package is deprecated. Use `crewai.llm` instead.
+
+    This package was reorganized from `crewai.llms.*` to `crewai.llm.*`.
+    All submodules are redirected to their new locations in `crewai.llm.*`.
+
+    Migration guide:
+        Old: from crewai.llms.base_llm import BaseLLM
+        New: from crewai.llm.base_llm import BaseLLM
+
+        Old: from crewai.llms.hooks.base import BaseInterceptor
+        New: from crewai.llm.hooks.base import BaseInterceptor
+
+        Old: from crewai.llms.constants import OPENAI_MODELS
+        New: from crewai.llm.constants import OPENAI_MODELS
+
+        Or use top-level imports:
+        from crewai import LLM, BaseLLM
+"""
+
+import warnings
+
+from crewai.llm import LLM
+from crewai.llm.base_llm import BaseLLM
+
+
+# Issue deprecation warning when this module is imported
+warnings.warn(
+    "The 'crewai.llms' package is deprecated and will be removed in a future version. "
+    "Please use 'crewai.llm' (singular) instead. "
+    "All submodules have been reorganized from 'crewai.llms.*' to 'crewai.llm.*'.",
+    DeprecationWarning,
+    stacklevel=2,
+)
+
+__all__ = ["LLM", "BaseLLM"]

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

@@ -1,550 +1,15 @@
-"""Base LLM abstract class for CrewAI.
+"""Deprecated: Use crewai.llm.base_llm instead.
 
-This module provides the abstract base class for all LLM implementations
-in CrewAI, including common functionality for native SDK implementations.
+.. deprecated:: 1.4.0
 """
 
-from __future__ import annotations
+import warnings
 
-from abc import ABC, abstractmethod
-from datetime import datetime
-import json
-import logging
-import re
-from typing import TYPE_CHECKING, Any, Final
 
-from pydantic import BaseModel
-
-from crewai.events.event_bus import crewai_event_bus
-from crewai.events.types.llm_events import (
-    LLMCallCompletedEvent,
-    LLMCallFailedEvent,
-    LLMCallStartedEvent,
-    LLMCallType,
-    LLMStreamChunkEvent,
+warnings.warn(
+    "crewai.llms.base_llm is deprecated. Use crewai.llm.base_llm instead.",
+    DeprecationWarning,
+    stacklevel=2,
 )
-from crewai.events.types.tool_usage_events import (
-    ToolUsageErrorEvent,
-    ToolUsageFinishedEvent,
-    ToolUsageStartedEvent,
-)
-from crewai.types.usage_metrics import UsageMetrics
 
-
-if TYPE_CHECKING:
-    from crewai.agent.core import Agent
-    from crewai.task import Task
-    from crewai.tools.base_tool import BaseTool
-    from crewai.utilities.types import LLMMessage
-
-
-DEFAULT_CONTEXT_WINDOW_SIZE: Final[int] = 4096
-DEFAULT_SUPPORTS_STOP_WORDS: Final[bool] = True
-_JSON_EXTRACTION_PATTERN: Final[re.Pattern[str]] = re.compile(r"\{.*}", re.DOTALL)
-
-
-class BaseLLM(ABC):
-    """Abstract base class for LLM implementations.
-
-    This class defines the interface that all LLM implementations must follow.
-    Users can extend this class to create custom LLM implementations that don't
-    rely on litellm's authentication mechanism.
-
-    Custom LLM implementations should handle error cases gracefully, including
-    timeouts, authentication failures, and malformed responses. They should also
-    implement proper validation for input parameters and provide clear error
-    messages when things go wrong.
-
-    Attributes:
-        model: The model identifier/name.
-        temperature: Optional temperature setting for response generation.
-        stop: A list of stop sequences that the LLM should use to stop generation.
-        additional_params: Additional provider-specific parameters.
-    """
-
-    is_litellm: bool = False
-
-    def __init__(
-        self,
-        model: str,
-        temperature: float | None = None,
-        api_key: str | None = None,
-        base_url: str | None = None,
-        provider: str | None = None,
-        **kwargs: Any,
-    ) -> None:
-        """Initialize the BaseLLM with default attributes.
-
-        Args:
-            model: The model identifier/name.
-            temperature: Optional temperature setting for response generation.
-            stop: Optional list of stop sequences for generation.
-            **kwargs: Additional provider-specific parameters.
-        """
-        if not model:
-            raise ValueError("Model name is required and cannot be empty")
-
-        self.model = model
-        self.temperature = temperature
-        self.api_key = api_key
-        self.base_url = base_url
-        # Store additional parameters for provider-specific use
-        self.additional_params = kwargs
-        self._provider = provider or "openai"
-
-        stop = kwargs.pop("stop", None)
-        if stop is None:
-            self.stop: list[str] = []
-        elif isinstance(stop, str):
-            self.stop = [stop]
-        elif isinstance(stop, list):
-            self.stop = stop
-        else:
-            self.stop = []
-
-        self._token_usage = {
-            "total_tokens": 0,
-            "prompt_tokens": 0,
-            "completion_tokens": 0,
-            "successful_requests": 0,
-            "cached_prompt_tokens": 0,
-        }
-
-    @property
-    def provider(self) -> str:
-        """Get the provider of the LLM."""
-        return self._provider
-
-    @provider.setter
-    def provider(self, value: str) -> None:
-        """Set the provider of the LLM."""
-        self._provider = value
-
-    @abstractmethod
-    def call(
-        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.
-        """
-
-    def _convert_tools_for_interference(
-        self, tools: list[dict[str, BaseTool]]
-    ) -> list[dict[str, BaseTool]]:
-        """Convert tools to a format that can be used for interference.
-
-        Args:
-            tools: List of tools to convert.
-
-        Returns:
-            List of converted tools (default implementation returns as-is)
-        """
-        return tools
-
-    def supports_stop_words(self) -> bool:
-        """Check if the LLM supports stop words.
-
-        Returns:
-            True if the LLM supports stop words, False otherwise.
-        """
-        return DEFAULT_SUPPORTS_STOP_WORDS
-
-    def _supports_stop_words_implementation(self) -> bool:
-        """Check if stop words are configured for this LLM instance.
-
-        Native providers can override supports_stop_words() to return this value
-        to ensure consistent behavior based on whether stop words are actually configured.
-
-        Returns:
-            True if stop words are configured and can be applied
-        """
-        return bool(self.stop)
-
-    def _apply_stop_words(self, content: str) -> str:
-        """Apply stop words to truncate response content.
-
-        This method provides consistent stop word behavior across all native SDK providers.
-        Native providers should call this method to post-process their responses.
-
-        Args:
-            content: The raw response content from the LLM
-
-        Returns:
-            Content truncated at the first occurrence of any stop word
-
-        Example:
-            >>> llm = MyNativeLLM(stop=["Observation:", "Final Answer:"])
-            >>> response = (
-            ...     "I need to search.\\n\\nAction: search\\nObservation: Found results"
-            ... )
-            >>> llm._apply_stop_words(response)
-            "I need to search.\\n\\nAction: search"
-        """
-        if not self.stop or not content:
-            return content
-
-        # Find the earliest occurrence of any stop word
-        earliest_stop_pos = len(content)
-        found_stop_word = None
-
-        for stop_word in self.stop:
-            stop_pos = content.find(stop_word)
-            if stop_pos != -1 and stop_pos < earliest_stop_pos:
-                earliest_stop_pos = stop_pos
-                found_stop_word = stop_word
-
-        # Truncate at the stop word if found
-        if found_stop_word is not None:
-            truncated = content[:earliest_stop_pos].strip()
-            logging.debug(
-                f"Applied stop word '{found_stop_word}' at position {earliest_stop_pos}"
-            )
-            return truncated
-
-        return content
-
-    def get_context_window_size(self) -> int:
-        """Get the context window size for the LLM.
-
-        Returns:
-            The number of tokens/characters the model can handle.
-        """
-        # Default implementation - subclasses should override with model-specific values
-        return DEFAULT_CONTEXT_WINDOW_SIZE
-
-    # Common helper methods for native SDK implementations
-
-    def _emit_call_started_event(
-        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,
-    ) -> None:
-        """Emit LLM call started event."""
-        if not hasattr(crewai_event_bus, "emit"):
-            raise ValueError("crewai_event_bus does not have an emit method") from None
-
-        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,
-            ),
-        )
-
-    def _emit_call_completed_event(
-        self,
-        response: Any,
-        call_type: LLMCallType,
-        from_task: Task | None = None,
-        from_agent: Agent | None = None,
-        messages: str | list[dict[str, Any]] | None = None,
-    ) -> None:
-        """Emit LLM call completed event."""
-        crewai_event_bus.emit(
-            self,
-            event=LLMCallCompletedEvent(
-                messages=messages,
-                response=response,
-                call_type=call_type,
-                from_task=from_task,
-                from_agent=from_agent,
-                model=self.model,
-            ),
-        )
-
-    def _emit_call_failed_event(
-        self,
-        error: str,
-        from_task: Task | None = None,
-        from_agent: Agent | None = None,
-    ) -> None:
-        """Emit LLM call failed event."""
-        if not hasattr(crewai_event_bus, "emit"):
-            raise ValueError("crewai_event_bus does not have an emit method") from None
-
-        crewai_event_bus.emit(
-            self,
-            event=LLMCallFailedEvent(
-                error=error,
-                from_task=from_task,
-                from_agent=from_agent,
-            ),
-        )
-
-    def _emit_stream_chunk_event(
-        self,
-        chunk: str,
-        from_task: Task | None = None,
-        from_agent: Agent | None = None,
-        tool_call: dict[str, Any] | None = None,
-    ) -> None:
-        """Emit stream chunk event."""
-        if not hasattr(crewai_event_bus, "emit"):
-            raise ValueError("crewai_event_bus does not have an emit method") from None
-
-        crewai_event_bus.emit(
-            self,
-            event=LLMStreamChunkEvent(
-                chunk=chunk,
-                tool_call=tool_call,
-                from_task=from_task,
-                from_agent=from_agent,
-            ),
-        )
-
-    def _handle_tool_execution(
-        self,
-        function_name: str,
-        function_args: dict[str, Any],
-        available_functions: dict[str, Any],
-        from_task: Task | None = None,
-        from_agent: Agent | None = None,
-    ) -> str | None:
-        """Handle tool execution with proper event emission.
-
-        Args:
-            function_name: Name of the function to execute
-            function_args: Arguments to pass to the function
-            available_functions: Dict of available functions
-            from_task: Optional task object
-            from_agent: Optional agent object
-
-        Returns:
-            Result of function execution or None if function not found
-        """
-        if function_name not in available_functions:
-            logging.warning(
-                f"Function '{function_name}' not found in available functions"
-            )
-            return None
-
-        try:
-            # Emit tool usage started event
-            started_at = datetime.now()
-
-            crewai_event_bus.emit(
-                self,
-                event=ToolUsageStartedEvent(
-                    tool_name=function_name,
-                    tool_args=function_args,
-                    from_agent=from_agent,
-                    from_task=from_task,
-                ),
-            )
-
-            # Execute the function
-            fn = available_functions[function_name]
-            result = fn(**function_args)
-
-            # Emit tool usage finished event
-            crewai_event_bus.emit(
-                self,
-                event=ToolUsageFinishedEvent(
-                    output=result,
-                    tool_name=function_name,
-                    tool_args=function_args,
-                    started_at=started_at,
-                    finished_at=datetime.now(),
-                    from_task=from_task,
-                    from_agent=from_agent,
-                ),
-            )
-
-            # Emit LLM call completed event for tool call
-            self._emit_call_completed_event(
-                response=result,
-                call_type=LLMCallType.TOOL_CALL,
-                from_task=from_task,
-                from_agent=from_agent,
-            )
-
-            return str(result)
-
-        except Exception as e:
-            error_msg = f"Error executing function '{function_name}': {e!s}"
-            logging.error(error_msg)
-
-            # Emit tool usage error event
-            if not hasattr(crewai_event_bus, "emit"):
-                raise ValueError(
-                    "crewai_event_bus does not have an emit method"
-                ) from None
-
-            crewai_event_bus.emit(
-                self,
-                event=ToolUsageErrorEvent(
-                    tool_name=function_name,
-                    tool_args=function_args,
-                    error=error_msg,
-                    from_task=from_task,
-                    from_agent=from_agent,
-                ),
-            )
-
-            # Emit LLM call failed event
-            self._emit_call_failed_event(
-                error=error_msg,
-                from_task=from_task,
-                from_agent=from_agent,
-            )
-
-            return None
-
-    def _format_messages(self, messages: str | list[LLMMessage]) -> list[LLMMessage]:
-        """Convert messages to standard format.
-
-        Args:
-            messages: Input messages (string or list of message dicts)
-
-        Returns:
-            List of message dictionaries with 'role' and 'content' keys
-
-        Raises:
-            ValueError: If message format is invalid
-        """
-        if isinstance(messages, str):
-            return [{"role": "user", "content": messages}]
-
-        # Validate message format
-        for i, msg in enumerate(messages):
-            if not isinstance(msg, dict):
-                raise ValueError(f"Message at index {i} must be a dictionary")
-            if "role" not in msg or "content" not in msg:
-                raise ValueError(
-                    f"Message at index {i} must have 'role' and 'content' keys"
-                )
-
-        return messages
-
-    @staticmethod
-    def _validate_structured_output(
-        response: str,
-        response_format: type[BaseModel] | None,
-    ) -> str | BaseModel:
-        """Validate and parse structured output.
-
-        Args:
-            response: Raw response string
-            response_format: Optional Pydantic model for structured output
-
-        Returns:
-            Parsed response (BaseModel instance if response_format provided, otherwise string)
-
-        Raises:
-            ValueError: If structured output validation fails
-        """
-        if response_format is None:
-            return response
-
-        try:
-            # Try to parse as JSON first
-            if response.strip().startswith("{") or response.strip().startswith("["):
-                data = json.loads(response)
-                return response_format.model_validate(data)
-
-            json_match = _JSON_EXTRACTION_PATTERN.search(response)
-            if json_match:
-                data = json.loads(json_match.group())
-                return response_format.model_validate(data)
-
-            raise ValueError("No JSON found in response")
-
-        except (json.JSONDecodeError, ValueError) as e:
-            logging.warning(f"Failed to parse structured output: {e}")
-            raise ValueError(
-                f"Failed to parse response into {response_format.__name__}: {e}"
-            ) from e
-
-    @staticmethod
-    def _extract_provider(model: str) -> str:
-        """Extract provider from model string.
-
-        Args:
-            model: Model string (e.g., 'openai/gpt-4' or 'gpt-4')
-
-        Returns:
-            Provider name (e.g., 'openai')
-        """
-        if "/" in model:
-            return model.partition("/")[0]
-        return "openai"  # Default provider
-
-    def _track_token_usage_internal(self, usage_data: dict[str, Any]) -> None:
-        """Track token usage internally in the LLM instance.
-
-        Args:
-            usage_data: Token usage data from the API response
-        """
-        # Extract tokens in a provider-agnostic way
-        prompt_tokens = (
-            usage_data.get("prompt_tokens")
-            or usage_data.get("prompt_token_count")
-            or usage_data.get("input_tokens")
-            or 0
-        )
-
-        completion_tokens = (
-            usage_data.get("completion_tokens")
-            or usage_data.get("candidates_token_count")
-            or usage_data.get("output_tokens")
-            or 0
-        )
-
-        cached_tokens = (
-            usage_data.get("cached_tokens")
-            or usage_data.get("cached_prompt_tokens")
-            or 0
-        )
-
-        self._token_usage["prompt_tokens"] += prompt_tokens
-        self._token_usage["completion_tokens"] += completion_tokens
-        self._token_usage["total_tokens"] += prompt_tokens + completion_tokens
-        self._token_usage["successful_requests"] += 1
-        self._token_usage["cached_prompt_tokens"] += cached_tokens
-
-    def get_token_usage_summary(self) -> UsageMetrics:
-        """Get summary of token usage for this LLM instance.
-
-        Returns:
-            Dictionary with token usage totals
-        """
-        return UsageMetrics(**self._token_usage)
+from crewai.llm.base_llm import *  # noqa: E402, F403

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

@@ -1,558 +1,15 @@
-from typing import Literal, TypeAlias
+"""Deprecated: Use crewai.llm.constants instead.
+
+.. deprecated:: 1.4.0
+"""
+
+import warnings
 
 
-OpenAIModels: TypeAlias = Literal[
-    "gpt-3.5-turbo",
-    "gpt-3.5-turbo-0125",
-    "gpt-3.5-turbo-0301",
-    "gpt-3.5-turbo-0613",
-    "gpt-3.5-turbo-1106",
-    "gpt-3.5-turbo-16k",
-    "gpt-3.5-turbo-16k-0613",
-    "gpt-3.5-turbo-instruct",
-    "gpt-3.5-turbo-instruct-0914",
-    "gpt-4",
-    "gpt-4-0125-preview",
-    "gpt-4-0314",
-    "gpt-4-0613",
-    "gpt-4-1106-preview",
-    "gpt-4-32k",
-    "gpt-4-32k-0314",
-    "gpt-4-32k-0613",
-    "gpt-4-turbo",
-    "gpt-4-turbo-2024-04-09",
-    "gpt-4-turbo-preview",
-    "gpt-4-vision-preview",
-    "gpt-4.1",
-    "gpt-4.1-2025-04-14",
-    "gpt-4.1-mini",
-    "gpt-4.1-mini-2025-04-14",
-    "gpt-4.1-nano",
-    "gpt-4.1-nano-2025-04-14",
-    "gpt-4o",
-    "gpt-4o-2024-05-13",
-    "gpt-4o-2024-08-06",
-    "gpt-4o-2024-11-20",
-    "gpt-4o-audio-preview",
-    "gpt-4o-audio-preview-2024-10-01",
-    "gpt-4o-audio-preview-2024-12-17",
-    "gpt-4o-audio-preview-2025-06-03",
-    "gpt-4o-mini",
-    "gpt-4o-mini-2024-07-18",
-    "gpt-4o-mini-audio-preview",
-    "gpt-4o-mini-audio-preview-2024-12-17",
-    "gpt-4o-mini-realtime-preview",
-    "gpt-4o-mini-realtime-preview-2024-12-17",
-    "gpt-4o-mini-search-preview",
-    "gpt-4o-mini-search-preview-2025-03-11",
-    "gpt-4o-mini-transcribe",
-    "gpt-4o-mini-tts",
-    "gpt-4o-realtime-preview",
-    "gpt-4o-realtime-preview-2024-10-01",
-    "gpt-4o-realtime-preview-2024-12-17",
-    "gpt-4o-realtime-preview-2025-06-03",
-    "gpt-4o-search-preview",
-    "gpt-4o-search-preview-2025-03-11",
-    "gpt-4o-transcribe",
-    "gpt-4o-transcribe-diarize",
-    "gpt-5",
-    "gpt-5-2025-08-07",
-    "gpt-5-chat",
-    "gpt-5-chat-latest",
-    "gpt-5-codex",
-    "gpt-5-mini",
-    "gpt-5-mini-2025-08-07",
-    "gpt-5-nano",
-    "gpt-5-nano-2025-08-07",
-    "gpt-5-pro",
-    "gpt-5-pro-2025-10-06",
-    "gpt-5-search-api",
-    "gpt-5-search-api-2025-10-14",
-    "gpt-audio",
-    "gpt-audio-2025-08-28",
-    "gpt-audio-mini",
-    "gpt-audio-mini-2025-10-06",
-    "gpt-image-1",
-    "gpt-image-1-mini",
-    "gpt-realtime",
-    "gpt-realtime-2025-08-28",
-    "gpt-realtime-mini",
-    "gpt-realtime-mini-2025-10-06",
-    "o1",
-    "o1-preview",
-    "o1-2024-12-17",
-    "o1-mini",
-    "o1-mini-2024-09-12",
-    "o1-pro",
-    "o1-pro-2025-03-19",
-    "o3-mini",
-    "o3",
-    "o4-mini",
-    "whisper-1",
-]
-OPENAI_MODELS: list[OpenAIModels] = [
-    "gpt-3.5-turbo",
-    "gpt-3.5-turbo-0125",
-    "gpt-3.5-turbo-0301",
-    "gpt-3.5-turbo-0613",
-    "gpt-3.5-turbo-1106",
-    "gpt-3.5-turbo-16k",
-    "gpt-3.5-turbo-16k-0613",
-    "gpt-3.5-turbo-instruct",
-    "gpt-3.5-turbo-instruct-0914",
-    "gpt-4",
-    "gpt-4-0125-preview",
-    "gpt-4-0314",
-    "gpt-4-0613",
-    "gpt-4-1106-preview",
-    "gpt-4-32k",
-    "gpt-4-32k-0314",
-    "gpt-4-32k-0613",
-    "gpt-4-turbo",
-    "gpt-4-turbo-2024-04-09",
-    "gpt-4-turbo-preview",
-    "gpt-4-vision-preview",
-    "gpt-4.1",
-    "gpt-4.1-2025-04-14",
-    "gpt-4.1-mini",
-    "gpt-4.1-mini-2025-04-14",
-    "gpt-4.1-nano",
-    "gpt-4.1-nano-2025-04-14",
-    "gpt-4o",
-    "gpt-4o-2024-05-13",
-    "gpt-4o-2024-08-06",
-    "gpt-4o-2024-11-20",
-    "gpt-4o-audio-preview",
-    "gpt-4o-audio-preview-2024-10-01",
-    "gpt-4o-audio-preview-2024-12-17",
-    "gpt-4o-audio-preview-2025-06-03",
-    "gpt-4o-mini",
-    "gpt-4o-mini-2024-07-18",
-    "gpt-4o-mini-audio-preview",
-    "gpt-4o-mini-audio-preview-2024-12-17",
-    "gpt-4o-mini-realtime-preview",
-    "gpt-4o-mini-realtime-preview-2024-12-17",
-    "gpt-4o-mini-search-preview",
-    "gpt-4o-mini-search-preview-2025-03-11",
-    "gpt-4o-mini-transcribe",
-    "gpt-4o-mini-tts",
-    "gpt-4o-realtime-preview",
-    "gpt-4o-realtime-preview-2024-10-01",
-    "gpt-4o-realtime-preview-2024-12-17",
-    "gpt-4o-realtime-preview-2025-06-03",
-    "gpt-4o-search-preview",
-    "gpt-4o-search-preview-2025-03-11",
-    "gpt-4o-transcribe",
-    "gpt-4o-transcribe-diarize",
-    "gpt-5",
-    "gpt-5-2025-08-07",
-    "gpt-5-chat",
-    "gpt-5-chat-latest",
-    "gpt-5-codex",
-    "gpt-5-mini",
-    "gpt-5-mini-2025-08-07",
-    "gpt-5-nano",
-    "gpt-5-nano-2025-08-07",
-    "gpt-5-pro",
-    "gpt-5-pro-2025-10-06",
-    "gpt-5-search-api",
-    "gpt-5-search-api-2025-10-14",
-    "gpt-audio",
-    "gpt-audio-2025-08-28",
-    "gpt-audio-mini",
-    "gpt-audio-mini-2025-10-06",
-    "gpt-image-1",
-    "gpt-image-1-mini",
-    "gpt-realtime",
-    "gpt-realtime-2025-08-28",
-    "gpt-realtime-mini",
-    "gpt-realtime-mini-2025-10-06",
-    "o1",
-    "o1-preview",
-    "o1-2024-12-17",
-    "o1-mini",
-    "o1-mini-2024-09-12",
-    "o1-pro",
-    "o1-pro-2025-03-19",
-    "o3-mini",
-    "o3",
-    "o4-mini",
-    "whisper-1",
-]
+warnings.warn(
+    "crewai.llms.constants is deprecated. Use crewai.llm.constants instead.",
+    DeprecationWarning,
+    stacklevel=2,
+)
 
-
-AnthropicModels: TypeAlias = Literal[
-    "claude-3-7-sonnet-latest",
-    "claude-3-7-sonnet-20250219",
-    "claude-3-5-haiku-latest",
-    "claude-3-5-haiku-20241022",
-    "claude-haiku-4-5",
-    "claude-haiku-4-5-20251001",
-    "claude-sonnet-4-20250514",
-    "claude-sonnet-4-0",
-    "claude-4-sonnet-20250514",
-    "claude-sonnet-4-5",
-    "claude-sonnet-4-5-20250929",
-    "claude-3-5-sonnet-latest",
-    "claude-3-5-sonnet-20241022",
-    "claude-3-5-sonnet-20240620",
-    "claude-opus-4-0",
-    "claude-opus-4-20250514",
-    "claude-4-opus-20250514",
-    "claude-opus-4-1",
-    "claude-opus-4-1-20250805",
-    "claude-3-opus-latest",
-    "claude-3-opus-20240229",
-    "claude-3-sonnet-20240229",
-    "claude-3-haiku-latest",
-    "claude-3-haiku-20240307",
-]
-ANTHROPIC_MODELS: list[AnthropicModels] = [
-    "claude-3-7-sonnet-latest",
-    "claude-3-7-sonnet-20250219",
-    "claude-3-5-haiku-latest",
-    "claude-3-5-haiku-20241022",
-    "claude-haiku-4-5",
-    "claude-haiku-4-5-20251001",
-    "claude-sonnet-4-20250514",
-    "claude-sonnet-4-0",
-    "claude-4-sonnet-20250514",
-    "claude-sonnet-4-5",
-    "claude-sonnet-4-5-20250929",
-    "claude-3-5-sonnet-latest",
-    "claude-3-5-sonnet-20241022",
-    "claude-3-5-sonnet-20240620",
-    "claude-opus-4-0",
-    "claude-opus-4-20250514",
-    "claude-4-opus-20250514",
-    "claude-opus-4-1",
-    "claude-opus-4-1-20250805",
-    "claude-3-opus-latest",
-    "claude-3-opus-20240229",
-    "claude-3-sonnet-20240229",
-    "claude-3-haiku-latest",
-    "claude-3-haiku-20240307",
-]
-
-GeminiModels: TypeAlias = Literal[
-    "gemini-2.5-pro",
-    "gemini-2.5-pro-preview-03-25",
-    "gemini-2.5-pro-preview-05-06",
-    "gemini-2.5-pro-preview-06-05",
-    "gemini-2.5-flash",
-    "gemini-2.5-flash-preview-05-20",
-    "gemini-2.5-flash-preview-04-17",
-    "gemini-2.5-flash-image",
-    "gemini-2.5-flash-image-preview",
-    "gemini-2.5-flash-lite",
-    "gemini-2.5-flash-lite-preview-06-17",
-    "gemini-2.5-flash-preview-09-2025",
-    "gemini-2.5-flash-lite-preview-09-2025",
-    "gemini-2.5-flash-preview-tts",
-    "gemini-2.5-pro-preview-tts",
-    "gemini-2.5-computer-use-preview-10-2025",
-    "gemini-2.0-flash",
-    "gemini-2.0-flash-001",
-    "gemini-2.0-flash-exp",
-    "gemini-2.0-flash-exp-image-generation",
-    "gemini-2.0-flash-lite",
-    "gemini-2.0-flash-lite-001",
-    "gemini-2.0-flash-lite-preview",
-    "gemini-2.0-flash-lite-preview-02-05",
-    "gemini-2.0-flash-preview-image-generation",
-    "gemini-2.0-flash-thinking-exp",
-    "gemini-2.0-flash-thinking-exp-01-21",
-    "gemini-2.0-flash-thinking-exp-1219",
-    "gemini-2.0-pro-exp",
-    "gemini-2.0-pro-exp-02-05",
-    "gemini-exp-1206",
-    "gemini-1.5-pro",
-    "gemini-1.5-flash",
-    "gemini-1.5-flash-8b",
-    "gemini-flash-latest",
-    "gemini-flash-lite-latest",
-    "gemini-pro-latest",
-    "gemini-2.0-flash-live-001",
-    "gemini-live-2.5-flash-preview",
-    "gemini-2.5-flash-live-preview",
-    "gemini-robotics-er-1.5-preview",
-    "gemini-gemma-2-27b-it",
-    "gemini-gemma-2-9b-it",
-    "gemma-3-1b-it",
-    "gemma-3-4b-it",
-    "gemma-3-12b-it",
-    "gemma-3-27b-it",
-    "gemma-3n-e2b-it",
-    "gemma-3n-e4b-it",
-    "learnlm-2.0-flash-experimental",
-]
-GEMINI_MODELS: list[GeminiModels] = [
-    "gemini-2.5-pro",
-    "gemini-2.5-pro-preview-03-25",
-    "gemini-2.5-pro-preview-05-06",
-    "gemini-2.5-pro-preview-06-05",
-    "gemini-2.5-flash",
-    "gemini-2.5-flash-preview-05-20",
-    "gemini-2.5-flash-preview-04-17",
-    "gemini-2.5-flash-image",
-    "gemini-2.5-flash-image-preview",
-    "gemini-2.5-flash-lite",
-    "gemini-2.5-flash-lite-preview-06-17",
-    "gemini-2.5-flash-preview-09-2025",
-    "gemini-2.5-flash-lite-preview-09-2025",
-    "gemini-2.5-flash-preview-tts",
-    "gemini-2.5-pro-preview-tts",
-    "gemini-2.5-computer-use-preview-10-2025",
-    "gemini-2.0-flash",
-    "gemini-2.0-flash-001",
-    "gemini-2.0-flash-exp",
-    "gemini-2.0-flash-exp-image-generation",
-    "gemini-2.0-flash-lite",
-    "gemini-2.0-flash-lite-001",
-    "gemini-2.0-flash-lite-preview",
-    "gemini-2.0-flash-lite-preview-02-05",
-    "gemini-2.0-flash-preview-image-generation",
-    "gemini-2.0-flash-thinking-exp",
-    "gemini-2.0-flash-thinking-exp-01-21",
-    "gemini-2.0-flash-thinking-exp-1219",
-    "gemini-2.0-pro-exp",
-    "gemini-2.0-pro-exp-02-05",
-    "gemini-exp-1206",
-    "gemini-1.5-pro",
-    "gemini-1.5-flash",
-    "gemini-1.5-flash-8b",
-    "gemini-flash-latest",
-    "gemini-flash-lite-latest",
-    "gemini-pro-latest",
-    "gemini-2.0-flash-live-001",
-    "gemini-live-2.5-flash-preview",
-    "gemini-2.5-flash-live-preview",
-    "gemini-robotics-er-1.5-preview",
-    "gemini-gemma-2-27b-it",
-    "gemini-gemma-2-9b-it",
-    "gemma-3-1b-it",
-    "gemma-3-4b-it",
-    "gemma-3-12b-it",
-    "gemma-3-27b-it",
-    "gemma-3n-e2b-it",
-    "gemma-3n-e4b-it",
-    "learnlm-2.0-flash-experimental",
-]
-
-
-AzureModels: TypeAlias = Literal[
-    "gpt-3.5-turbo",
-    "gpt-3.5-turbo-0301",
-    "gpt-3.5-turbo-0613",
-    "gpt-3.5-turbo-16k",
-    "gpt-3.5-turbo-16k-0613",
-    "gpt-35-turbo",
-    "gpt-35-turbo-0125",
-    "gpt-35-turbo-1106",
-    "gpt-35-turbo-16k-0613",
-    "gpt-35-turbo-instruct-0914",
-    "gpt-4",
-    "gpt-4-0314",
-    "gpt-4-0613",
-    "gpt-4-1106-preview",
-    "gpt-4-0125-preview",
-    "gpt-4-32k",
-    "gpt-4-32k-0314",
-    "gpt-4-32k-0613",
-    "gpt-4-turbo",
-    "gpt-4-turbo-2024-04-09",
-    "gpt-4-vision",
-    "gpt-4o",
-    "gpt-4o-2024-05-13",
-    "gpt-4o-2024-08-06",
-    "gpt-4o-2024-11-20",
-    "gpt-4o-mini",
-    "gpt-5",
-    "o1",
-    "o1-mini",
-    "o1-preview",
-    "o3-mini",
-    "o3",
-    "o4-mini",
-]
-AZURE_MODELS: list[AzureModels] = [
-    "gpt-3.5-turbo",
-    "gpt-3.5-turbo-0301",
-    "gpt-3.5-turbo-0613",
-    "gpt-3.5-turbo-16k",
-    "gpt-3.5-turbo-16k-0613",
-    "gpt-35-turbo",
-    "gpt-35-turbo-0125",
-    "gpt-35-turbo-1106",
-    "gpt-35-turbo-16k-0613",
-    "gpt-35-turbo-instruct-0914",
-    "gpt-4",
-    "gpt-4-0314",
-    "gpt-4-0613",
-    "gpt-4-1106-preview",
-    "gpt-4-0125-preview",
-    "gpt-4-32k",
-    "gpt-4-32k-0314",
-    "gpt-4-32k-0613",
-    "gpt-4-turbo",
-    "gpt-4-turbo-2024-04-09",
-    "gpt-4-vision",
-    "gpt-4o",
-    "gpt-4o-2024-05-13",
-    "gpt-4o-2024-08-06",
-    "gpt-4o-2024-11-20",
-    "gpt-4o-mini",
-    "gpt-5",
-    "o1",
-    "o1-mini",
-    "o1-preview",
-    "o3-mini",
-    "o3",
-    "o4-mini",
-]
-
-
-BedrockModels: TypeAlias = Literal[
-    "ai21.jamba-1-5-large-v1:0",
-    "ai21.jamba-1-5-mini-v1:0",
-    "amazon.nova-lite-v1:0",
-    "amazon.nova-lite-v1:0:24k",
-    "amazon.nova-lite-v1:0:300k",
-    "amazon.nova-micro-v1:0",
-    "amazon.nova-micro-v1:0:128k",
-    "amazon.nova-micro-v1:0:24k",
-    "amazon.nova-premier-v1:0",
-    "amazon.nova-premier-v1:0:1000k",
-    "amazon.nova-premier-v1:0:20k",
-    "amazon.nova-premier-v1:0:8k",
-    "amazon.nova-premier-v1:0:mm",
-    "amazon.nova-pro-v1:0",
-    "amazon.nova-pro-v1:0:24k",
-    "amazon.nova-pro-v1:0:300k",
-    "amazon.titan-text-express-v1",
-    "amazon.titan-text-express-v1:0:8k",
-    "amazon.titan-text-lite-v1",
-    "amazon.titan-text-lite-v1:0:4k",
-    "amazon.titan-tg1-large",
-    "anthropic.claude-3-5-haiku-20241022-v1:0",
-    "anthropic.claude-3-5-sonnet-20240620-v1:0",
-    "anthropic.claude-3-5-sonnet-20241022-v2:0",
-    "anthropic.claude-3-7-sonnet-20250219-v1:0",
-    "anthropic.claude-3-haiku-20240307-v1:0",
-    "anthropic.claude-3-haiku-20240307-v1:0:200k",
-    "anthropic.claude-3-haiku-20240307-v1:0:48k",
-    "anthropic.claude-3-opus-20240229-v1:0",
-    "anthropic.claude-3-opus-20240229-v1:0:12k",
-    "anthropic.claude-3-opus-20240229-v1:0:200k",
-    "anthropic.claude-3-opus-20240229-v1:0:28k",
-    "anthropic.claude-3-sonnet-20240229-v1:0",
-    "anthropic.claude-3-sonnet-20240229-v1:0:200k",
-    "anthropic.claude-3-sonnet-20240229-v1:0:28k",
-    "anthropic.claude-haiku-4-5-20251001-v1:0",
-    "anthropic.claude-instant-v1:2:100k",
-    "anthropic.claude-opus-4-1-20250805-v1:0",
-    "anthropic.claude-opus-4-20250514-v1:0",
-    "anthropic.claude-sonnet-4-20250514-v1:0",
-    "anthropic.claude-sonnet-4-5-20250929-v1:0",
-    "anthropic.claude-v2:0:100k",
-    "anthropic.claude-v2:0:18k",
-    "anthropic.claude-v2:1:18k",
-    "anthropic.claude-v2:1:200k",
-    "cohere.command-r-plus-v1:0",
-    "cohere.command-r-v1:0",
-    "cohere.rerank-v3-5:0",
-    "deepseek.r1-v1:0",
-    "meta.llama3-1-70b-instruct-v1:0",
-    "meta.llama3-1-8b-instruct-v1:0",
-    "meta.llama3-2-11b-instruct-v1:0",
-    "meta.llama3-2-1b-instruct-v1:0",
-    "meta.llama3-2-3b-instruct-v1:0",
-    "meta.llama3-2-90b-instruct-v1:0",
-    "meta.llama3-3-70b-instruct-v1:0",
-    "meta.llama3-70b-instruct-v1:0",
-    "meta.llama3-8b-instruct-v1:0",
-    "meta.llama4-maverick-17b-instruct-v1:0",
-    "meta.llama4-scout-17b-instruct-v1:0",
-    "mistral.mistral-7b-instruct-v0:2",
-    "mistral.mistral-large-2402-v1:0",
-    "mistral.mistral-small-2402-v1:0",
-    "mistral.mixtral-8x7b-instruct-v0:1",
-    "mistral.pixtral-large-2502-v1:0",
-    "openai.gpt-oss-120b-1:0",
-    "openai.gpt-oss-20b-1:0",
-    "qwen.qwen3-32b-v1:0",
-    "qwen.qwen3-coder-30b-a3b-v1:0",
-    "twelvelabs.pegasus-1-2-v1:0",
-]
-BEDROCK_MODELS: list[BedrockModels] = [
-    "ai21.jamba-1-5-large-v1:0",
-    "ai21.jamba-1-5-mini-v1:0",
-    "amazon.nova-lite-v1:0",
-    "amazon.nova-lite-v1:0:24k",
-    "amazon.nova-lite-v1:0:300k",
-    "amazon.nova-micro-v1:0",
-    "amazon.nova-micro-v1:0:128k",
-    "amazon.nova-micro-v1:0:24k",
-    "amazon.nova-premier-v1:0",
-    "amazon.nova-premier-v1:0:1000k",
-    "amazon.nova-premier-v1:0:20k",
-    "amazon.nova-premier-v1:0:8k",
-    "amazon.nova-premier-v1:0:mm",
-    "amazon.nova-pro-v1:0",
-    "amazon.nova-pro-v1:0:24k",
-    "amazon.nova-pro-v1:0:300k",
-    "amazon.titan-text-express-v1",
-    "amazon.titan-text-express-v1:0:8k",
-    "amazon.titan-text-lite-v1",
-    "amazon.titan-text-lite-v1:0:4k",
-    "amazon.titan-tg1-large",
-    "anthropic.claude-3-5-haiku-20241022-v1:0",
-    "anthropic.claude-3-5-sonnet-20240620-v1:0",
-    "anthropic.claude-3-5-sonnet-20241022-v2:0",
-    "anthropic.claude-3-7-sonnet-20250219-v1:0",
-    "anthropic.claude-3-haiku-20240307-v1:0",
-    "anthropic.claude-3-haiku-20240307-v1:0:200k",
-    "anthropic.claude-3-haiku-20240307-v1:0:48k",
-    "anthropic.claude-3-opus-20240229-v1:0",
-    "anthropic.claude-3-opus-20240229-v1:0:12k",
-    "anthropic.claude-3-opus-20240229-v1:0:200k",
-    "anthropic.claude-3-opus-20240229-v1:0:28k",
-    "anthropic.claude-3-sonnet-20240229-v1:0",
-    "anthropic.claude-3-sonnet-20240229-v1:0:200k",
-    "anthropic.claude-3-sonnet-20240229-v1:0:28k",
-    "anthropic.claude-haiku-4-5-20251001-v1:0",
-    "anthropic.claude-instant-v1:2:100k",
-    "anthropic.claude-opus-4-1-20250805-v1:0",
-    "anthropic.claude-opus-4-20250514-v1:0",
-    "anthropic.claude-sonnet-4-20250514-v1:0",
-    "anthropic.claude-sonnet-4-5-20250929-v1:0",
-    "anthropic.claude-v2:0:100k",
-    "anthropic.claude-v2:0:18k",
-    "anthropic.claude-v2:1:18k",
-    "anthropic.claude-v2:1:200k",
-    "cohere.command-r-plus-v1:0",
-    "cohere.command-r-v1:0",
-    "cohere.rerank-v3-5:0",
-    "deepseek.r1-v1:0",
-    "meta.llama3-1-70b-instruct-v1:0",
-    "meta.llama3-1-8b-instruct-v1:0",
-    "meta.llama3-2-11b-instruct-v1:0",
-    "meta.llama3-2-1b-instruct-v1:0",
-    "meta.llama3-2-3b-instruct-v1:0",
-    "meta.llama3-2-90b-instruct-v1:0",
-    "meta.llama3-3-70b-instruct-v1:0",
-    "meta.llama3-70b-instruct-v1:0",
-    "meta.llama3-8b-instruct-v1:0",
-    "meta.llama4-maverick-17b-instruct-v1:0",
-    "meta.llama4-scout-17b-instruct-v1:0",
-    "mistral.mistral-7b-instruct-v0:2",
-    "mistral.mistral-large-2402-v1:0",
-    "mistral.mistral-small-2402-v1:0",
-    "mistral.mixtral-8x7b-instruct-v0:1",
-    "mistral.pixtral-large-2502-v1:0",
-    "openai.gpt-oss-120b-1:0",
-    "openai.gpt-oss-20b-1:0",
-    "qwen.qwen3-32b-v1:0",
-    "qwen.qwen3-coder-30b-a3b-v1:0",
-    "twelvelabs.pegasus-1-2-v1:0",
-]
+from crewai.llm.constants import *  # noqa: E402, F403