Fix test isolation: reset TraceCollectionListener singleton in cleanup_event_handlers fixture

Prevents singleton state pollution (trace_batch_id="debug-trace-batch") from
test_flow_conversation.py leaking into test_trace_enable_disable.py when both
files run in the same pytest-xdist worker under --dist=loadfile. The polluted
singleton caused a background HTTP request to fake.crewai.com that corrupted
VCR cassette state, preventing the OpenAI LLM entry from matching.

.gitignore

@@ -31,3 +31,5 @@ chromadb-*.lock
 blogs/*
 secrets/*
 UNKNOWN.egg-info/
+demos/*
+.crewai/*

conftest.py

@@ -197,6 +197,21 @@ def cleanup_event_handlers() -> Generator[None, Any, None]:
     except Exception:  # noqa: S110
         pass
 
+    try:
+        from crewai.events.listeners.tracing.trace_listener import (
+            TraceCollectionListener,
+        )
+
+        if TraceCollectionListener._instance is not None:
+            instance_dict = TraceCollectionListener._instance.__dict__
+            if "_initialized" in instance_dict:
+                del TraceCollectionListener._instance._initialized
+            if "_listeners_setup" in instance_dict:
+                del TraceCollectionListener._instance._listeners_setup
+        TraceCollectionListener._instance = None
+    except Exception:  # noqa: S110
+        pass
+
 
 @pytest.fixture(autouse=True, scope="function")
 def reset_event_state() -> None:

docs/en/concepts/memory.mdx

@@ -101,7 +101,7 @@ crew = Crew(
 )
 ```
 
-When `memory=True`, the crew creates a default `Memory()` and passes the crew's `embedder` configuration through automatically. All agents in the crew share the crew's memory unless an agent has its own.
+When `memory=True`, the crew creates a default `Memory()` and passes the crew's `embedder` configuration through automatically. All agents in the crew share the crew's memory unless an agent has its own. Without a custom `embedder`, memory uses OpenAI `text-embedding-3-large` embeddings.
 
 After each task, the crew automatically extracts discrete facts from the task output and stores them. Before each task, the agent recalls relevant context from memory and injects it into the task prompt.
 
@@ -515,7 +515,11 @@ memory = Memory(
 
 ## Embedder Configuration
 
-Memory needs an embedding model to convert text into vectors for semantic search. You can configure this in three ways.
+Memory needs an embedding model to convert text into vectors for semantic search. By default, `Memory()` uses OpenAI `text-embedding-3-large` embeddings, which produce 3072-dimensional vectors. Set `OPENAI_API_KEY` for the default path, or configure a custom embedder in one of three ways.
+
+<Warning>
+Existing local memory stores created with 1536-dimensional embeddings, such as `text-embedding-3-small` or `text-embedding-ada-002`, may not be compatible with the `text-embedding-3-large` default. This applies to both the OpenAI and Azure OpenAI providers — Azure's default embedding model also changed from `text-embedding-ada-002` to `text-embedding-3-large`. If local testing fails with an embedding dimension mismatch, reset memory with `crewai reset-memories -m`, delete the local memory storage directory, or explicitly configure the older embedder model until you migrate.
+</Warning>
 
 ### Passing to Memory Directly
 
@@ -523,7 +527,7 @@ Memory needs an embedding model to convert text into vectors for semantic search
 from crewai import Memory
 
 # As a config dict
-memory = Memory(embedder={"provider": "openai", "config": {"model_name": "text-embedding-3-small"}})
+memory = Memory(embedder={"provider": "openai", "config": {"model_name": "text-embedding-3-large"}})
 
 # As a pre-built callable
 from crewai.rag.embeddings.factory import build_embedder
@@ -542,7 +546,7 @@ crew = Crew(
     agents=[...],
     tasks=[...],
     memory=True,
-    embedder={"provider": "openai", "config": {"model_name": "text-embedding-3-small"}},
+    embedder={"provider": "openai", "config": {"model_name": "text-embedding-3-large"}},
 )
 ```
 
@@ -554,7 +558,7 @@ crew = Crew(
 memory = Memory(embedder={
     "provider": "openai",
     "config": {
-        "model_name": "text-embedding-3-small",
+        "model_name": "text-embedding-3-large",
         # "api_key": "sk-...",  # or set OPENAI_API_KEY env var
     },
 })
@@ -701,9 +705,9 @@ memory = Memory(embedder=my_embedder)
 
 | Provider | Key | Typical Model | Notes |
 | :--- | :--- | :--- | :--- |
-| OpenAI | `openai` | `text-embedding-3-small` | Default. Set `OPENAI_API_KEY`. |
+| OpenAI | `openai` | `text-embedding-3-large` | Default. Set `OPENAI_API_KEY`. |
 | Ollama | `ollama` | `mxbai-embed-large` | Local, no API key needed. |
-| Azure OpenAI | `azure` | `text-embedding-ada-002` | Requires `deployment_id`. |
+| Azure OpenAI | `azure` | `text-embedding-3-large` | Default model. Requires `deployment_id`. |
 | Google AI | `google-generativeai` | `gemini-embedding-001` | Set `GOOGLE_API_KEY`. |
 | Google Vertex | `google-vertex` | `gemini-embedding-001` | Requires `project_id`. |
 | Cohere | `cohere` | `embed-english-v3.0` | Strong multilingual support. |
@@ -836,6 +840,9 @@ class MemoryMonitor(BaseEventListener):
 **Background save errors in logs?**
 - Memory saves run in a background thread. Errors are emitted as `MemorySaveFailedEvent` but don't crash the agent. Check logs for the root cause (usually LLM or embedder connection issues).
 
+**Embedding dimension mismatch?**
+- Existing local memory stores may have been created with a different embedding model. The default OpenAI memory embedder is now `text-embedding-3-large` (3072 dimensions), while older stores commonly used 1536-dimensional embeddings. For local testing, run `crewai reset-memories -m`, delete the local memory storage directory, or configure the previous embedder model explicitly.
+
 **Concurrent write conflicts?**
 - LanceDB operations are serialized with a shared lock and retried automatically on conflict. This handles multiple `Memory` instances pointing at the same database (e.g. agent memory + crew memory). No action needed.
 
@@ -862,7 +869,7 @@ All configuration is passed as keyword arguments to `Memory(...)`. Every paramet
 | :--- | :--- | :--- |
 | `llm` | `"gpt-4o-mini"` | LLM for analysis (model name or `BaseLLM` instance). |
 | `storage` | `"lancedb"` | Storage backend (`"lancedb"`, a path string, or a `StorageBackend` instance). |
-| `embedder` | `None` (OpenAI default) | Embedder (config dict, callable, or `None` for default OpenAI). |
+| `embedder` | `None` (OpenAI `text-embedding-3-large`) | Embedder (config dict, callable, or `None` for default OpenAI). |
 | `recency_weight` | `0.3` | Weight for recency in composite score. |
 | `semantic_weight` | `0.5` | Weight for semantic similarity in composite score. |
 | `importance_weight` | `0.2` | Weight for importance in composite score. |

docs/en/guides/migration/upgrading-crewai.mdx

@@ -141,7 +141,7 @@ crew = Crew(
     process=Process.sequential,   # or Process.hierarchical
     memory=True,
     cache=True,
-    embedder={"provider": "openai", "config": {"model": "text-embedding-3-small"}},
+    embedder={"provider": "openai", "config": {"model": "text-embedding-3-large"}},
 )
 ```
 
@@ -173,7 +173,7 @@ write = Task(
 
 ### Memory & embedder config {#memory-embedder-config}
 
-If `memory=True` and you're not using the default OpenAI embeddings, you must pass an `embedder`:
+If `memory=True` and you're not using the default OpenAI `text-embedding-3-large` embeddings, you must pass an `embedder`:
 
 ```python
 crew = Crew(
@@ -187,4 +187,4 @@ crew = Crew(
 )
 ```
 
-Set the relevant provider credentials (`OPENAI_API_KEY`, `OLLAMA_HOST`, etc.) in your `.env` file. Memory storage paths are project-local by default — delete the project's memory directory if you change embedders, since dimensions don't mix.
+Set the relevant provider credentials (`OPENAI_API_KEY`, `OLLAMA_HOST`, etc.) in your `.env` file. Memory storage paths are project-local by default. Existing local memory stores created with 1536-dimensional embeddings may not be compatible with the default OpenAI `text-embedding-3-large` embedder, which uses 3072 dimensions. If you hit a dimension mismatch, delete the project's memory directory, run `crewai reset-memories -m`, or explicitly configure the older embedder model until you migrate.

lib/cli/src/crewai_cli/cli.py

@@ -3,42 +3,94 @@ from __future__ import annotations
 from importlib.metadata import version as get_version
 import os
 import subprocess
-from typing import Any
+from typing import TYPE_CHECKING, Any
 
 import click
 from crewai_core.token_manager import TokenManager
 
-from crewai_cli.add_crew_to_flow import add_crew_to_flow
-from crewai_cli.authentication.main import AuthenticationCommand
 from crewai_cli.config import Settings
-from crewai_cli.create_crew import create_crew
-from crewai_cli.create_flow import create_flow
-from crewai_cli.crew_chat import run_chat
-from crewai_cli.deploy.main import DeployCommand
-from crewai_cli.enterprise.main import EnterpriseConfigureCommand
-from crewai_cli.evaluate_crew import evaluate_crew
-from crewai_cli.experimental.skills.main import SkillCommand
-from crewai_cli.install_crew import install_crew
-from crewai_cli.kickoff_flow import kickoff_flow
-from crewai_cli.organization.main import OrganizationCommand
-from crewai_cli.plot_flow import plot_flow
-from crewai_cli.remote_template.main import TemplateCommand
-from crewai_cli.replay_from_task import replay_task_command
-from crewai_cli.reset_memories_command import reset_memories_command
-from crewai_cli.run_crew import run_crew
-from crewai_cli.run_flow_definition import run_flow_definition
-from crewai_cli.settings.main import SettingsCommand
-from crewai_cli.task_outputs import load_task_outputs
-from crewai_cli.tools.main import ToolCommand
-from crewai_cli.train_crew import train_crew
-from crewai_cli.triggers.main import TriggersCommand
-from crewai_cli.update_crew import update_crew
 from crewai_cli.user_data import (
     _load_user_data,
     is_tracing_enabled,
     update_user_data,
 )
-from crewai_cli.utils import build_env_with_all_tool_credentials, read_toml
+from crewai_cli.utils import (
+    build_env_with_all_tool_credentials,
+    enable_prompt_line_editing,
+    read_toml,
+)
+
+
+def train_crew(*args: Any, **kwargs: Any) -> Any:
+    from crewai_cli.train_crew import train_crew as _train_crew
+
+    return _train_crew(*args, **kwargs)
+
+
+def evaluate_crew(*args: Any, **kwargs: Any) -> Any:
+    from crewai_cli.evaluate_crew import evaluate_crew as _evaluate_crew
+
+    return _evaluate_crew(*args, **kwargs)
+
+
+def replay_task_command(*args: Any, **kwargs: Any) -> Any:
+    from crewai_cli.replay_from_task import replay_task_command as _replay_task_command
+
+    return _replay_task_command(*args, **kwargs)
+
+
+def run_flow_definition(*args: Any, **kwargs: Any) -> Any:
+    from crewai_cli.run_flow_definition import (
+        run_flow_definition as _run_flow_definition,
+    )
+
+    return _run_flow_definition(*args, **kwargs)
+
+
+def run_crew(*args: Any, **kwargs: Any) -> Any:
+    from crewai_cli.run_crew import run_crew as _run_crew
+
+    return _run_crew(*args, **kwargs)
+
+
+if TYPE_CHECKING:
+    # mypy sees the real classes; at runtime the shims below defer the
+    # heavy imports until a command actually instantiates them.
+    from crewai_cli.authentication.main import AuthenticationCommand
+    from crewai_cli.deploy.main import DeployCommand
+    from crewai_cli.organization.main import OrganizationCommand
+    from crewai_cli.remote_template.main import TemplateCommand
+else:
+
+    class AuthenticationCommand:
+        def __new__(cls, *args: Any, **kwargs: Any) -> Any:
+            from crewai_cli.authentication.main import (
+                AuthenticationCommand as _AuthenticationCommand,
+            )
+
+            return _AuthenticationCommand(*args, **kwargs)
+
+    class DeployCommand:
+        def __new__(cls, *args: Any, **kwargs: Any) -> Any:
+            from crewai_cli.deploy.main import DeployCommand as _DeployCommand
+
+            return _DeployCommand(*args, **kwargs)
+
+    class TemplateCommand:
+        def __new__(cls, *args: Any, **kwargs: Any) -> Any:
+            from crewai_cli.remote_template.main import (
+                TemplateCommand as _TemplateCommand,
+            )
+
+            return _TemplateCommand(*args, **kwargs)
+
+    class OrganizationCommand:
+        def __new__(cls, *args: Any, **kwargs: Any) -> Any:
+            from crewai_cli.organization.main import (
+                OrganizationCommand as _OrganizationCommand,
+            )
+
+            return _OrganizationCommand(*args, **kwargs)
 
 
 def _get_cli_version() -> str:
@@ -91,17 +143,57 @@ def uv(uv_args: tuple[str, ...]) -> None:
 
 
 @crewai.command()
-@click.argument("type", type=click.Choice(["crew", "flow"]))
-@click.argument("name")
+@click.argument(
+    "type", required=False, default=None, type=click.Choice(["crew", "flow"])
+)
+@click.argument("name", required=False, default=None)
 @click.option("--provider", type=str, help="The provider to use for the crew")
 @click.option("--skip_provider", is_flag=True, help="Skip provider validation")
+@click.option(
+    "--classic",
+    is_flag=True,
+    help="Use classic Python/YAML project structure instead of JSON",
+)
 def create(
-    type: str, name: str, provider: str | None, skip_provider: bool = False
+    type: str | None,
+    name: str | None,
+    provider: str | None,
+    skip_provider: bool = False,
+    classic: bool = False,
 ) -> None:
     """Create a new crew, or flow."""
+    if not type:
+        from crewai_cli.tui_picker import pick
+
+        options = [
+            ("crew", "A team of AI agents working together"),
+            (
+                "flow",
+                "A deterministic workflow with full control over agents and crews",
+            ),
+        ]
+        type = pick("What would you like to create?", options)
+        if type is None:
+            raise SystemExit(0)
+        click.echo()
+    if not name:
+        enable_prompt_line_editing()
+        name = click.prompt(
+            click.style(f"  Name of your {type}", fg="cyan", bold=True),
+            prompt_suffix=click.style(" › ", fg="bright_white"),  # noqa: RUF001
+        )
     if type == "crew":
-        create_crew(name, provider, skip_provider)
+        if classic:
+            from crewai_cli.create_crew import create_crew
+
+            create_crew(name, provider, skip_provider)
+        else:
+            from crewai_cli.create_json_crew import create_json_crew
+
+            create_json_crew(name, provider, skip_provider)
     elif type == "flow":
+        from crewai_cli.create_flow import create_flow
+
         create_flow(name)
     else:
         click.secho("Error: Invalid type. Must be 'crew' or 'flow'.", fg="red")
@@ -186,6 +278,8 @@ def replay(task_id: str, trained_agents_file: str | None) -> None:
 def log_tasks_outputs() -> None:
     """Retrieve your latest crew.kickoff() task outputs."""
     try:
+        from crewai_cli.task_outputs import load_task_outputs
+
         tasks = load_task_outputs()
 
         if not tasks:
@@ -274,6 +368,8 @@ def reset_memories(
                 "Please specify at least one memory type to reset using the appropriate flags."
             )
             return
+        from crewai_cli.reset_memories_command import reset_memories_command
+
         reset_memories_command(memory, knowledge, agent_knowledge, kickoff_outputs, all)
     except Exception as e:
         click.echo(f"An error occurred while resetting memories: {e}", err=True)
@@ -296,7 +392,7 @@ def reset_memories(
     "--embedder-model",
     type=str,
     default=None,
-    help="Embedder model name (e.g. text-embedding-3-small, gemini-embedding-001).",
+    help="Embedder model name (e.g. text-embedding-3-large, gemini-embedding-001).",
 )
 @click.option(
     "--embedder-config",
@@ -351,7 +447,7 @@ def memory(
     "-m",
     "--model",
     type=str,
-    default="gpt-4o-mini",
+    default="gpt-5.4-mini",
     help="LLM Model to run the tests on the Crew. For now only accepting only OpenAI models.",
 )
 @click.option(
@@ -382,6 +478,8 @@ def test(n_iterations: int, model: str, trained_agents_file: str | None) -> None
 @click.pass_context
 def install(context: click.Context) -> None:
     """Install the Crew."""
+    from crewai_cli.install_crew import install_crew
+
     install_crew(context.args)
 
 
@@ -415,7 +513,9 @@ def install(context: click.Context) -> None:
     help='Experimental: JSON object passed to flow.kickoff(), e.g. \'{"topic":"AI"}\'.',
 )
 def run(
-    trained_agents_file: str | None, definition: str | None, inputs: str | None
+    trained_agents_file: str | None,
+    definition: str | None,
+    inputs: str | None,
 ) -> None:
     """Run the Crew or Flow."""
     if inputs is not None and definition is None:
@@ -435,6 +535,8 @@ def run(
 @crewai.command()
 def update() -> None:
     """Update the pyproject.toml of the Crew project to use uv."""
+    from crewai_cli.update_crew import update_crew
+
     update_crew()
 
 
@@ -544,6 +646,8 @@ def tool() -> None:
 @tool.command(name="create")
 @click.argument("handle")
 def tool_create(handle: str) -> None:
+    from crewai_cli.tools.main import ToolCommand
+
     tool_cmd = ToolCommand()
     tool_cmd.create(handle)
 
@@ -551,6 +655,8 @@ def tool_create(handle: str) -> None:
 @tool.command(name="install")
 @click.argument("handle")
 def tool_install(handle: str) -> None:
+    from crewai_cli.tools.main import ToolCommand
+
     tool_cmd = ToolCommand()
     tool_cmd.login()
     tool_cmd.install(handle)
@@ -567,6 +673,8 @@ def tool_install(handle: str) -> None:
 @click.option("--public", "is_public", flag_value=True, default=False)
 @click.option("--private", "is_public", flag_value=False)
 def tool_publish(is_public: bool, force: bool) -> None:
+    from crewai_cli.tools.main import ToolCommand
+
     tool_cmd = ToolCommand()
     tool_cmd.login()
     tool_cmd.publish(is_public, force)
@@ -599,6 +707,8 @@ def skill() -> None:
     help="Create skill in current dir instead of ./skills/",
 )
 def skill_create(name: str, in_project: bool) -> None:
+    from crewai_cli.experimental.skills.main import SkillCommand
+
     skill_cmd = SkillCommand()
     skill_cmd.create(name, in_project=in_project)
 
@@ -606,6 +716,8 @@ def skill_create(name: str, in_project: bool) -> None:
 @skill.command(name="install")
 @click.argument("ref")
 def skill_install(ref: str) -> None:
+    from crewai_cli.experimental.skills.main import SkillCommand
+
     skill_cmd = SkillCommand()
     skill_cmd.install(ref)
 
@@ -622,6 +734,8 @@ def skill_install(ref: str) -> None:
 @click.option("--private", "is_public", flag_value=False)
 @click.option("--org", default=None, help="Organisation slug (overrides settings).")
 def skill_publish(is_public: bool, org: str | None, force: bool) -> None:
+    from crewai_cli.experimental.skills.main import SkillCommand
+
     skill_cmd = SkillCommand()
     skill_cmd.publish(is_public, org=org, force=force)
 
@@ -629,6 +743,8 @@ def skill_publish(is_public: bool, org: str | None, force: bool) -> None:
 @skill.command(name="list")
 def skill_list() -> None:
     """List locally installed skills."""
+    from crewai_cli.experimental.skills.main import SkillCommand
+
     skill_cmd = SkillCommand()
     skill_cmd.list_cached()
 
@@ -668,6 +784,8 @@ def flow() -> None:
 @flow.command(name="kickoff")
 def flow_run() -> None:
     """Kickoff the Flow."""
+    from crewai_cli.kickoff_flow import kickoff_flow
+
     click.echo("Running the Flow")
     kickoff_flow()
 
@@ -675,6 +793,8 @@ def flow_run() -> None:
 @flow.command(name="plot")
 def flow_plot() -> None:
     """Plot the Flow."""
+    from crewai_cli.plot_flow import plot_flow
+
     click.echo("Plotting the Flow")
     plot_flow()
 
@@ -683,6 +803,8 @@ def flow_plot() -> None:
 @click.argument("crew_name")
 def flow_add_crew(crew_name: str) -> None:
     """Add a crew to an existing flow."""
+    from crewai_cli.add_crew_to_flow import add_crew_to_flow
+
     click.echo(f"Adding crew {crew_name} to the flow")
     add_crew_to_flow(crew_name)
 
@@ -695,6 +817,8 @@ def triggers() -> None:
 @triggers.command(name="list")
 def triggers_list() -> None:
     """List all available triggers from integrations."""
+    from crewai_cli.triggers.main import TriggersCommand
+
     triggers_cmd = TriggersCommand()
     triggers_cmd.list_triggers()
 
@@ -703,6 +827,8 @@ def triggers_list() -> None:
 @click.argument("trigger_path")
 def triggers_run(trigger_path: str) -> None:
     """Execute crew with trigger payload. Format: app_slug/trigger_slug"""
+    from crewai_cli.triggers.main import TriggersCommand
+
     triggers_cmd = TriggersCommand()
     triggers_cmd.execute_with_trigger(trigger_path)
 
@@ -715,6 +841,8 @@ def chat() -> None:
     click.secho(
         "\nStarting a conversation with the Crew\nType 'exit' or Ctrl+C to quit.\n",
     )
+    from crewai_cli.crew_chat import run_chat
+
     run_chat()
 
 
@@ -754,6 +882,8 @@ def enterprise() -> None:
 @click.argument("enterprise_url")
 def enterprise_configure(enterprise_url: str) -> None:
     """Configure CrewAI AMP OAuth2 settings from the provided Enterprise URL."""
+    from crewai_cli.enterprise.main import EnterpriseConfigureCommand
+
     enterprise_command = EnterpriseConfigureCommand()
     enterprise_command.configure(enterprise_url)
 
@@ -766,6 +896,8 @@ def config() -> None:
 @config.command("list")
 def config_list() -> None:
     """List all CLI configuration parameters."""
+    from crewai_cli.settings.main import SettingsCommand
+
     config_command = SettingsCommand()
     config_command.list()
 
@@ -775,6 +907,8 @@ def config_list() -> None:
 @click.argument("value")
 def config_set(key: str, value: str) -> None:
     """Set a CLI configuration parameter."""
+    from crewai_cli.settings.main import SettingsCommand
+
     config_command = SettingsCommand()
     config_command.set(key, value)
 
@@ -782,6 +916,8 @@ def config_set(key: str, value: str) -> None:
 @config.command("reset")
 def config_reset() -> None:
     """Reset all CLI configuration parameters to default values."""
+    from crewai_cli.settings.main import SettingsCommand
+
     config_command = SettingsCommand()
     config_command.reset_all_settings()
 

lib/cli/src/crewai_cli/deploy/main.py

@@ -34,6 +34,39 @@ def _run_predeploy_validation(skip_validate: bool) -> bool:
     return True
 
 
+def _display_git_repository_help() -> None:
+    """Explain how to prepare a new project for deployment."""
+    console.print(
+        "Deployment requires a Git repository with an origin remote.",
+        style="bold red",
+    )
+    console.print(
+        "CrewAI AMP deploys from the remote repository URL, so commit and push "
+        "this project first, then run deploy again.",
+        style="yellow",
+    )
+    console.print("\nSuggested setup:")
+    console.print("  git init")
+    console.print("  git add .")
+    console.print('  git commit -m "Initial crew"')
+    console.print("  git branch -M main")
+    console.print("  git remote add origin <your-repo-url>")
+    console.print("  git push -u origin main")
+
+
+def _display_git_remote_help() -> None:
+    """Explain how to add a remote to an existing Git repository."""
+    console.print("No remote repository URL found.", style="bold red")
+    console.print(
+        "CrewAI AMP deploys from the origin remote. Add a remote, push your "
+        "latest commit, then run deploy again.",
+        style="yellow",
+    )
+    console.print("\nSuggested setup:")
+    console.print("  git remote add origin <your-repo-url>")
+    console.print("  git push -u origin HEAD")
+
+
 class DeployCommand(BaseCommand, PlusAPIMixin):
     """
     A class to handle deployment-related operations for CrewAI projects.
@@ -124,14 +157,11 @@ class DeployCommand(BaseCommand, PlusAPIMixin):
         try:
             remote_repo_url = git.Repository().origin_url()
         except ValueError:
-            remote_repo_url = None
+            _display_git_repository_help()
+            return
 
         if remote_repo_url is None:
-            console.print("No remote repository URL found.", style="bold red")
-            console.print(
-                "Please ensure your project has a valid remote repository.",
-                style="yellow",
-            )
+            _display_git_remote_help()
             return
 
         self._confirm_input(env_vars, remote_repo_url, confirm)

lib/cli/src/crewai_cli/deploy/validate.py

@@ -38,6 +38,12 @@ import subprocess
 import sys
 from typing import Any
 
+from crewai.project.json_loader import (
+    JSONProjectValidationError,
+    find_crew_json_file,
+    find_json_project_file,
+    validate_crew_project,
+)
 from rich.console import Console
 
 from crewai_cli.utils import parse_toml
@@ -151,9 +157,33 @@ class DeployValidator:
     def ok(self) -> bool:
         return not self.errors
 
+    @property
+    def _is_json_crew(self) -> bool:
+        """True for JSON crew projects, deferring to the declared type.
+
+        A flow project that also contains a crew.json(c) file validates as
+        the flow it declares in pyproject.toml, not as a JSON crew.
+        """
+        if find_crew_json_file(self.project_root) is None:
+            return False
+        pyproject_path = self.project_root / "pyproject.toml"
+        if not pyproject_path.exists():
+            return True
+        try:
+            data = parse_toml(pyproject_path.read_text())
+        except Exception:
+            return True
+        declared_type: str | None = (
+            (data.get("tool") or {}).get("crewai", {}).get("type")
+        )
+        return declared_type != "flow"
+
     def run(self) -> list[ValidationResult]:
         """Run all checks. Later checks are skipped when earlier ones make
         them impossible (e.g. no pyproject.toml → no lockfile check)."""
+        if self._is_json_crew:
+            return self._run_json_checks()
+
         if not self._check_pyproject():
             return self.results
 
@@ -176,6 +206,110 @@ class DeployValidator:
 
         return self.results
 
+    def _run_json_checks(self) -> list[ValidationResult]:
+        """Validation suite for JSON-defined crew projects."""
+        crew_path = find_crew_json_file(self.project_root)
+        if crew_path is None:
+            return self.results
+
+        try:
+            project = validate_crew_project(crew_path, self.project_root / "agents")
+        except JSONProjectValidationError as e:
+            self._add(
+                Severity.ERROR,
+                "invalid_crew_json",
+                f"{crew_path.name} has invalid JSON crew configuration",
+                detail="\n".join(e.errors),
+                hint="Fix the JSON crew, agent, and task references before deploying.",
+            )
+            return self.results
+        except Exception as e:
+            self._add(
+                Severity.ERROR,
+                "invalid_crew_json",
+                f"Cannot parse {crew_path.name}",
+                detail=str(e),
+            )
+            return self.results
+
+        agents_dir = self.project_root / "agents"
+
+        self._check_pyproject()
+        self._check_lockfile()
+        self._check_env_vars_json(crew_path, agents_dir, project.agent_names)
+        self._check_version_vs_lockfile()
+
+        return self.results
+
+    def _check_env_vars_json(
+        self, crew_path: Path, agents_dir: Path, agent_names: list[str]
+    ) -> None:
+        """Check for env var references in JSON crew files."""
+        referenced: set[str] = set()
+        pattern = re.compile(r"\$\{?([A-Z][A-Z0-9_]+)\}?")
+
+        try:
+            referenced.update(pattern.findall(crew_path.read_text(errors="ignore")))
+        except OSError as exc:
+            logger.debug("Skipping unreadable crew file %s: %s", crew_path, exc)
+
+        for name in agent_names:
+            agent_path = find_json_project_file(agents_dir, name)
+            if agent_path is None:
+                continue
+            try:
+                referenced.update(
+                    pattern.findall(agent_path.read_text(errors="ignore"))
+                )
+            except OSError as exc:
+                logger.debug("Skipping unreadable agent file %s: %s", agent_path, exc)
+
+        for py_path in self.project_root.rglob("*.py"):
+            if ".venv" in py_path.parts:
+                continue
+            try:
+                text = py_path.read_text(encoding="utf-8", errors="ignore")
+            except OSError:
+                continue
+            env_pattern = re.compile(
+                r"""(?x)
+                (?:os\.environ\s*(?:\[\s*|\.get\s*\(\s*)
+                  |os\.getenv\s*\(\s*
+                  |getenv\s*\(\s*)
+                ['"]([A-Z][A-Z0-9_]*)['"]
+                """
+            )
+            referenced.update(env_pattern.findall(text))
+
+        env_file = self.project_root / ".env"
+        env_keys: set[str] = set()
+        if env_file.exists():
+            for line in env_file.read_text(errors="ignore").splitlines():
+                line = line.strip()
+                if not line or line.startswith("#") or "=" not in line:
+                    continue
+                env_keys.add(line.split("=", 1)[0].strip())
+
+        missing_known = sorted(
+            var
+            for var in referenced
+            if var in _KNOWN_API_KEY_HINTS
+            and var not in env_keys
+            and var not in os.environ
+        )
+        if missing_known:
+            self._add(
+                Severity.WARNING,
+                "env_vars_not_in_dotenv",
+                f"{len(missing_known)} referenced API key(s) not in .env",
+                detail=(
+                    "These env vars are referenced in your project but not set "
+                    f"locally: {', '.join(missing_known)}. Deploys will fail "
+                    "unless they are added to the deployment's Environment "
+                    "Variables in the CrewAI dashboard."
+                ),
+            )
+
     def _check_pyproject(self) -> bool:
         pyproject_path = self.project_root / "pyproject.toml"
         if not pyproject_path.exists():

lib/cli/src/crewai_cli/git.py

@@ -48,6 +48,7 @@ class Repository:
                 ["git", "rev-parse", "--is-inside-work-tree"],  # noqa: S607
                 cwd=self.path,
                 encoding="utf-8",
+                stderr=subprocess.DEVNULL,
             )
             return True
         except subprocess.CalledProcessError:

lib/cli/src/crewai_cli/run_crew.py

@@ -1,25 +1,311 @@
+from __future__ import annotations
+
+from contextlib import AbstractContextManager, nullcontext
 from enum import Enum
+import os
+from pathlib import Path
+import re
 import subprocess
+import sys
+from typing import TYPE_CHECKING, Any
 
 import click
+from crewai.project.json_loader import find_crew_json_file
 from crewai_core.constants import CREWAI_TRAINED_AGENTS_FILE_ENV
 from packaging import version
 
-from crewai_cli.utils import build_env_with_all_tool_credentials, read_toml
+from crewai_cli.utils import (
+    build_env_with_all_tool_credentials,
+    enable_prompt_line_editing,
+    read_toml,
+)
 from crewai_cli.version import get_crewai_version
 
 
+if TYPE_CHECKING:
+    from crewai_cli.crew_run_tui import CrewRunApp
+
+
 class CrewType(Enum):
     STANDARD = "standard"
     FLOW = "flow"
 
 
-def run_crew(trained_agents_file: str | None = None) -> None:
-    """Run the crew or flow by running a command in the UV environment.
+# Must accept the same names as the kickoff interpolation pattern in
+# crewai.utilities.string_utils (_VARIABLE_PATTERN), including hyphens —
+# otherwise placeholders are interpolated at runtime but never prompted for.
+_INPUT_PLACEHOLDER_RE = re.compile(r"(?<!{){([A-Za-z_][A-Za-z0-9_\-]*)}(?!})")
 
-    Starting from version 0.103.0, this command can be used to run both
-    standard crews and flows. For flows, it detects the type from pyproject.toml
-    and automatically runs the appropriate command.
+
+def _has_json_crew() -> bool:
+    """Check if this is a JSON-defined crew project.
+
+    The project type declared in pyproject.toml wins: a flow project that
+    happens to contain a crew.json(c) file still runs as a flow. A missing
+    or unreadable pyproject means a bare JSON crew project.
+    """
+    if find_crew_json_file() is None:
+        return False
+    try:
+        pyproject_data = read_toml()
+    except Exception:
+        return True
+    declared_type: str | None = (
+        pyproject_data.get("tool", {}).get("crewai", {}).get("type")
+    )
+    return declared_type != "flow"
+
+
+def _extract_input_placeholders(text: str | None) -> set[str]:
+    if not text:
+        return set()
+    return set(_INPUT_PLACEHOLDER_RE.findall(text))
+
+
+def _missing_input_names(crew: Any, inputs: dict[str, Any]) -> list[str]:
+    """Return input placeholders used by a crew but not provided as defaults."""
+    placeholders: set[str] = set()
+
+    for agent in getattr(crew, "agents", []) or []:
+        placeholders.update(_extract_input_placeholders(getattr(agent, "role", None)))
+        placeholders.update(_extract_input_placeholders(getattr(agent, "goal", None)))
+        placeholders.update(
+            _extract_input_placeholders(getattr(agent, "backstory", None))
+        )
+
+    for task in getattr(crew, "tasks", []) or []:
+        placeholders.update(
+            _extract_input_placeholders(getattr(task, "description", None))
+        )
+        placeholders.update(
+            _extract_input_placeholders(getattr(task, "expected_output", None))
+        )
+        placeholders.update(
+            _extract_input_placeholders(getattr(task, "output_file", None))
+        )
+
+    return sorted(name for name in placeholders if name not in inputs)
+
+
+def _prompt_for_missing_inputs(
+    crew: Any, default_inputs: dict[str, Any]
+) -> dict[str, Any]:
+    """Ask for runtime values for placeholders that lack default inputs."""
+    inputs = dict(default_inputs or {})
+    missing = _missing_input_names(crew, inputs)
+    if not missing:
+        return inputs
+
+    enable_prompt_line_editing()
+
+    click.echo()
+    click.secho("  Runtime inputs", fg="cyan", bold=True)
+    click.secho(
+        "  Values for {placeholder} references in your agents and tasks.",
+        dim=True,
+    )
+
+    for name in missing:
+        inputs[name] = click.prompt(
+            click.style(f"  {name}", fg="cyan"),
+            prompt_suffix=click.style(" > ", fg="bright_white"),
+        )
+
+    return inputs
+
+
+def _json_loading_status(message: str) -> AbstractContextManager[Any]:
+    from rich.console import Console
+    from rich.text import Text
+
+    console = Console()
+    if not console.is_terminal:
+        return nullcontext()
+    return console.status(
+        Text(f"  {message}", style="bold #1F7982"),
+        spinner="dots",
+    )
+
+
+def _load_json_crew(crew_path: Path) -> tuple[Any, dict[str, Any]]:
+    from crewai.project.crew_loader import load_crew
+
+    return load_crew(crew_path)
+
+
+def _load_json_crew_for_tui(
+    crew_path: Path,
+) -> tuple[type[Any], Any, dict[str, Any], list[str], list[str]]:
+    with _json_loading_status("Preparing crew..."):
+        from crewai_cli.crew_run_tui import CrewRunApp
+
+        crew, default_inputs = _load_json_crew(crew_path)
+        _prepare_json_crew_for_tui(crew)
+        task_names = [
+            getattr(task, "name", "") or getattr(task, "description", "")[:40] or "Task"
+            for task in crew.tasks
+        ]
+        agent_names = [
+            getattr(agent, "role", "") or getattr(agent, "name", "") or "Agent"
+            for agent in crew.agents
+        ]
+
+    return CrewRunApp, crew, default_inputs, task_names, agent_names
+
+
+def _prepare_json_crew_for_tui(crew: Any) -> None:
+    """Apply the same quiet/streaming setup used by the TUI JSON loader."""
+    crew.verbose = False
+    for agent in crew.agents:
+        agent.verbose = False
+        if hasattr(agent, "llm") and hasattr(agent.llm, "stream"):
+            agent.llm.stream = True
+
+
+def _run_json_crew(trained_agents_file: str | None = None) -> Any:
+    """Load and run a JSON-defined crew."""
+    from dotenv import load_dotenv
+
+    env_file = Path.cwd() / ".env"
+    if env_file.exists():
+        load_dotenv(env_file, override=True)
+
+    # JSON crews run in-process, so export the trained-agents file directly
+    # instead of forwarding it to a subprocess like classic crews do.
+    if trained_agents_file:
+        os.environ[CREWAI_TRAINED_AGENTS_FILE_ENV] = trained_agents_file
+
+    crew_path = find_crew_json_file()
+    if crew_path is None:
+        raise FileNotFoundError("No crew.jsonc or crew.json found")
+
+    crew_run_app_cls, crew, default_inputs, task_names, agent_names = (
+        _load_json_crew_for_tui(crew_path)
+    )
+    runtime_inputs = _prompt_for_missing_inputs(crew, default_inputs)
+
+    app = crew_run_app_cls(
+        crew_name=crew.name or "Crew",
+        total_tasks=len(crew.tasks),
+        agent_names=agent_names,
+        task_names=task_names,
+    )
+    app._crew = crew
+    app._default_inputs = runtime_inputs
+
+    app.run()
+
+    _print_post_tui_summary(app)
+
+    if app._status == "failed":
+        # Mirror the classic subprocess path: a failed crew must produce a
+        # non-zero exit code so scripts and CI don't treat it as success.
+        raise SystemExit(1)
+
+    if app._status not in ("completed", "failed"):
+        # User quit mid-run. kickoff runs in a thread worker that cannot be
+        # force-cancelled, so end the process to stop in-flight LLM and tool
+        # work instead of letting it burn tokens in the background.
+        click.secho("\n  Run cancelled.", fg="yellow")
+        sys.stdout.flush()
+        os._exit(130)
+
+    if getattr(app, "_want_deploy", False):
+        _chain_deploy()
+
+    return app._crew_result
+
+
+def _chain_deploy() -> None:
+    from rich.console import Console
+
+    console = Console()
+    try:
+        from crewai_cli.deploy.main import DeployCommand
+
+        console.print("\nStarting deployment…\n", style="bold #FF5A50")
+        DeployCommand().create_crew(confirm=False, skip_validate=True)
+    except SystemExit:
+        from crewai_cli.authentication.main import AuthenticationCommand
+
+        console.print()
+        AuthenticationCommand().login()
+        try:
+            DeployCommand().create_crew(confirm=False, skip_validate=True)
+        except Exception as e:
+            console.print(f"\nDeploy failed: {e}\n", style="bold red")
+    except Exception as e:
+        console.print(f"\nDeploy failed: {e}\n", style="bold red")
+
+
+def _print_post_tui_summary(app: CrewRunApp) -> None:
+    """Print a summary to the terminal after the Textual TUI exits."""
+    import time
+
+    from rich.console import Console
+    from rich.markdown import Markdown
+    from rich.padding import Padding
+    from rich.panel import Panel
+    from rich.text import Text
+
+    console = Console()
+    elapsed = time.time() - app._start_time
+
+    out_tokens = app._output_tokens + app._live_out_tokens
+    token_parts = []
+    if app._input_tokens:
+        token_parts.append(f"↑{app._input_tokens:,}")
+    if out_tokens:
+        token_parts.append(f"↓{out_tokens:,}")
+    token_str = "  ".join(token_parts)
+    if token_str:
+        token_str += " tokens"
+
+    crewai_red = "#FF5A50"
+    crewai_teal = "#1F7982"
+
+    if app._status == "completed":
+        summary = Text()
+        summary.append(
+            f"  ✔ Completed {app._total_tasks} tasks",
+            style=f"bold {crewai_teal}",
+        )
+        summary.append(f" in {elapsed:.1f}s", style="dim")
+        if token_str:
+            summary.append(f"  {token_str}", style="dim")
+        console.print(
+            Panel(
+                summary,
+                title=f" {app._crew_name} ",
+                title_align="left",
+                border_style=crewai_teal,
+                padding=(0, 1),
+            )
+        )
+        if app._final_output:
+            console.print()
+            console.print(Text("  Final Result", style=f"bold {crewai_teal}"))
+            console.print()
+            console.print(Padding(Markdown(app._final_output), (0, 2)))
+    elif app._status == "failed":
+        content = Text()
+        content.append("  ✘ Failed", style=f"bold {crewai_red}")
+        content.append(f" after {elapsed:.1f}s\n", style="dim")
+        if app._error:
+            content.append(f"\n  {app._error}\n", style=crewai_red)
+        console.print(
+            Panel(
+                content,
+                title=f" {app._crew_name} ",
+                title_align="left",
+                border_style=crewai_red,
+                padding=(0, 1),
+            )
+        )
+
+
+def run_crew(trained_agents_file: str | None = None) -> None:
+    """Run the crew or flow.
 
     Args:
         trained_agents_file: Optional path to a trained-agents pickle produced
@@ -27,6 +313,11 @@ def run_crew(trained_agents_file: str | None = None) -> None:
             ``CREWAI_TRAINED_AGENTS_FILE`` so agents load suggestions from this
             file instead of the default ``trained_agents_data.pkl``.
     """
+    # JSON crew projects take precedence
+    if _has_json_crew():
+        _run_json_crew(trained_agents_file=trained_agents_file)
+        return
+
     crewai_version = get_crewai_version()
     min_required_version = "0.71.0"
     pyproject_data = read_toml()

lib/cli/src/crewai_cli/tui_picker.py

@@ -0,0 +1,419 @@
+"""Arrow-key interactive pickers for CLI prompts."""
+
+from __future__ import annotations
+
+from contextlib import suppress
+import sys
+from typing import overload
+
+import click
+
+
+# CrewAI brand: primary=#FF5A50 (coral), teal=#1F7982
+_CORAL = "\033[38;2;255;90;80m"  # #FF5A50
+_TEAL = "\033[38;2;31;121;130m"  # #1F7982
+_BOLD = "\033[1m"
+_DIM = "\033[2m"
+_RESET = "\033[0m"
+_HIDE_CURSOR = "\033[?25l"
+_SHOW_CURSOR = "\033[?25h"
+
+
+def _is_interactive() -> bool:
+    try:
+        return sys.stdin.isatty() and sys.stdout.isatty()
+    except Exception:
+        return False
+
+
+def _read_key() -> str:
+    if sys.platform == "win32":
+        import msvcrt
+
+        ch = msvcrt.getwch()
+        if ch in ("\x00", "\xe0"):
+            ch2 = msvcrt.getwch()
+            return {"H": "up", "P": "down"}.get(ch2, "")
+        if ch == "\r":
+            return "enter"
+        if ch == " ":
+            return "space"
+        if ch == "\x03":
+            raise KeyboardInterrupt
+        return ch
+
+    import termios
+    import tty
+
+    fd = sys.stdin.fileno()
+    old = termios.tcgetattr(fd)
+    try:
+        tty.setcbreak(fd)
+        ch = sys.stdin.read(1)
+        if ch == "\x1b":
+            seq = sys.stdin.read(2)
+            if seq == "[A":
+                return "up"
+            if seq == "[B":
+                return "down"
+            return "esc"
+        if ch in ("\r", "\n"):
+            return "enter"
+        if ch == " ":
+            return "space"
+        if ch == "\x03":
+            raise KeyboardInterrupt
+        return ch
+    finally:
+        termios.tcsetattr(fd, termios.TCSADRAIN, old)
+
+
+def _clear_lines(n: int) -> None:
+    sys.stdout.write(f"\033[{n}A")
+    for _ in range(n):
+        sys.stdout.write("\033[2K\n")
+    sys.stdout.write(f"\033[{n}A")
+    sys.stdout.flush()
+
+
+def _draw_single(labels: list[str], cursor: int, *, clear: bool = False) -> None:
+    total = len(labels)
+    if clear:
+        sys.stdout.write(f"\033[{total}A")
+    for i, label in enumerate(labels):
+        if i == cursor:
+            sys.stdout.write(f"\033[2K  {_CORAL}→{_RESET} {_BOLD}{label}{_RESET}\n")
+        else:
+            sys.stdout.write(f"\033[2K    {label}\n")
+    sys.stdout.flush()
+
+
+def _draw_multi(
+    labels: list[str],
+    cursor: int,
+    selected: set[int],
+    *,
+    action_indices: set[int] | None = None,
+    separator_indices: set[int] | None = None,
+    clear: bool = False,
+) -> None:
+    action_indices = action_indices or set()
+    separator_indices = separator_indices or set()
+    hint_text = "↑↓ navigate, space toggle, enter confirm"
+    if action_indices:
+        hint_text = "↑↓ navigate, space toggle, enter confirm, ▸ rows expand/collapse"
+    hint = f"  {_DIM}{hint_text}{_RESET}"
+    total = len(labels) + 1
+    if clear:
+        sys.stdout.write(f"\033[{total}A")
+    sys.stdout.write(f"\033[2K{hint}\n")
+    for i, label in enumerate(labels):
+        if i in separator_indices:
+            sys.stdout.write(f"\033[2K      {_TEAL}{label}{_RESET}\n")
+            continue
+        if i in action_indices:
+            check = "  "
+        elif i in selected:
+            check = f"{_CORAL}[x]{_RESET}"
+        else:
+            check = "[ ]"
+        arrow = f"{_CORAL}→{_RESET} " if i == cursor else "  "
+        bold = f"{_BOLD}{label}{_RESET}" if i == cursor else label
+        sys.stdout.write(f"\033[2K    {arrow}{check} {bold}\n")
+    sys.stdout.flush()
+
+
+def _arrow_select_one(labels: list[str]) -> int:
+    cursor = 0
+    total = len(labels)
+    sys.stdout.write(_HIDE_CURSOR)
+    sys.stdout.flush()
+    try:
+        _draw_single(labels, cursor)
+        while True:
+            key = _read_key()
+            if key == "up" and cursor > 0:
+                cursor -= 1
+                _draw_single(labels, cursor, clear=True)
+            elif key == "down" and cursor < total - 1:
+                cursor += 1
+                _draw_single(labels, cursor, clear=True)
+            elif key == "enter":
+                _clear_lines(total)
+                return cursor
+            elif key in ("esc", "q"):
+                _clear_lines(total)
+                return -1
+    finally:
+        sys.stdout.write(_SHOW_CURSOR)
+        sys.stdout.flush()
+
+
+def _arrow_select_multi(
+    labels: list[str],
+    *,
+    action_indices: set[int] | None = None,
+    separator_indices: set[int] | None = None,
+    preselected: set[int] | None = None,
+    initial_cursor: int | None = None,
+) -> tuple[list[int], int | None]:
+    total = len(labels)
+    selected: set[int] = set(preselected or ())
+    action_indices = action_indices or set()
+    separator_indices = separator_indices or set()
+    if initial_cursor is not None and 0 <= initial_cursor < total:
+        cursor = initial_cursor
+    else:
+        cursor = _first_selectable_index(total, separator_indices)
+    sys.stdout.write(_HIDE_CURSOR)
+    sys.stdout.flush()
+    try:
+        _draw_multi(
+            labels,
+            cursor,
+            selected,
+            action_indices=action_indices,
+            separator_indices=separator_indices,
+        )
+        while True:
+            key = _read_key()
+            if key == "up":
+                cursor = _next_selectable_index(cursor, -1, total, separator_indices)
+                _draw_multi(
+                    labels,
+                    cursor,
+                    selected,
+                    action_indices=action_indices,
+                    separator_indices=separator_indices,
+                    clear=True,
+                )
+            elif key == "down":
+                cursor = _next_selectable_index(cursor, 1, total, separator_indices)
+                _draw_multi(
+                    labels,
+                    cursor,
+                    selected,
+                    action_indices=action_indices,
+                    separator_indices=separator_indices,
+                    clear=True,
+                )
+            elif key == "space":
+                if cursor in action_indices:
+                    _clear_lines(total + 1)
+                    return sorted(selected), cursor
+                selected ^= {cursor}
+                _draw_multi(
+                    labels,
+                    cursor,
+                    selected,
+                    action_indices=action_indices,
+                    separator_indices=separator_indices,
+                    clear=True,
+                )
+            elif key == "enter":
+                _clear_lines(total + 1)
+                if cursor in action_indices:
+                    return sorted(selected), cursor
+                return sorted(selected), None
+            elif key in ("esc", "q"):
+                _clear_lines(total + 1)
+                return sorted(selected), None
+    finally:
+        sys.stdout.write(_SHOW_CURSOR)
+        sys.stdout.flush()
+
+
+def _numbered_select(labels: list[str]) -> int:
+    for idx, label in enumerate(labels, 1):
+        click.echo(f"    {idx}. {label}")
+    click.echo()
+    while True:
+        choice = click.prompt("  Select", type=str, default="1")
+        if choice.lower() == "q":
+            return -1
+        try:
+            num = int(choice)
+            if 1 <= num <= len(labels):
+                return num - 1
+        except ValueError:
+            # Non-numeric input falls through to the shared error message.
+            pass
+        click.secho(f"  Invalid choice. Enter 1-{len(labels)}.", fg="red")
+
+
+def _numbered_select_multi(
+    labels: list[str],
+    *,
+    action_indices: set[int] | None = None,
+    separator_indices: set[int] | None = None,
+    preselected: set[int] | None = None,
+) -> tuple[list[int], int | None]:
+    action_indices = action_indices or set()
+    separator_indices = separator_indices or set()
+    numbered_indices: list[int] = []
+    for idx, label in enumerate(labels):
+        if idx in separator_indices:
+            click.secho(f"    {label}", fg="cyan")
+            continue
+        numbered_indices.append(idx)
+        click.echo(f"    {len(numbered_indices)}. {label}")
+    click.echo()
+    raw = click.prompt(
+        "  Select (comma-separated numbers, or empty to skip)",
+        default="",
+        show_default=False,
+    )
+    if not raw.strip():
+        return sorted(preselected or ()), None
+    indices: list[int] = list(preselected or ())
+    for part in raw.split(","):
+        with suppress(ValueError):
+            num = int(part.strip())
+            if 1 <= num <= len(numbered_indices):
+                idx = numbered_indices[num - 1]
+                if idx in action_indices:
+                    return sorted(set(indices)), idx
+                indices.append(idx)
+    return sorted(set(indices)), None
+
+
+def _first_selectable_index(total: int, separator_indices: set[int]) -> int:
+    for idx in range(total):
+        if idx not in separator_indices:
+            return idx
+    return 0
+
+
+def _next_selectable_index(
+    cursor: int,
+    direction: int,
+    total: int,
+    separator_indices: set[int],
+) -> int:
+    next_cursor = cursor + direction
+    while 0 <= next_cursor < total:
+        if next_cursor not in separator_indices:
+            return next_cursor
+        next_cursor += direction
+    return cursor
+
+
+# ── Public API ──────────────────────────────────────────────────
+
+
+def pick(title: str, options: list[tuple[str, str]]) -> str | None:
+    """Arrow-key single-select picker.
+
+    Args:
+        title: Header text.
+        options: List of ``(value, description)`` tuples.
+
+    Returns:
+        The *value* of the selected option, or ``None`` if cancelled.
+    """
+    labels = [f"{value:<12s} {desc}" for value, desc in options]
+
+    click.echo()
+    click.secho(f"  {title}", fg="cyan", bold=True)
+    click.echo()
+
+    if _is_interactive():
+        try:
+            idx = _arrow_select_one(labels)
+        except Exception:
+            idx = _numbered_select(labels)
+    else:
+        idx = _numbered_select(labels)
+
+    if idx < 0:
+        return None
+
+    value, _desc = options[idx]
+    click.secho(f"  ✔ {value}", fg="green")
+    return value
+
+
+def pick_one(title: str, labels: list[str]) -> int:
+    """Arrow-key single-select from plain labels.
+
+    Returns:
+        Selected index, or ``-1`` if cancelled.
+    """
+    click.echo()
+    click.secho(f"  {title}", fg="cyan")
+
+    if _is_interactive():
+        try:
+            return _arrow_select_one(labels)
+        except Exception:
+            return _numbered_select(labels)
+    return _numbered_select(labels)
+
+
+@overload
+def pick_many(
+    title: str,
+    labels: list[str],
+    *,
+    separator_indices: set[int] | None = None,
+    preselected: set[int] | None = None,
+    initial_cursor: int | None = None,
+) -> list[int]: ...
+
+
+@overload
+def pick_many(
+    title: str,
+    labels: list[str],
+    *,
+    action_indices: set[int],
+    separator_indices: set[int] | None = None,
+    preselected: set[int] | None = None,
+    initial_cursor: int | None = None,
+) -> tuple[list[int], int | None]: ...
+
+
+def pick_many(
+    title: str,
+    labels: list[str],
+    *,
+    action_indices: set[int] | None = None,
+    separator_indices: set[int] | None = None,
+    preselected: set[int] | None = None,
+    initial_cursor: int | None = None,
+) -> list[int] | tuple[list[int], int | None]:
+    """Arrow-key multi-select with checkboxes.
+
+    Returns:
+        Sorted list of selected indices, or ``(indices, action_index)`` when
+        ``action_indices`` is provided.
+    """
+    click.echo()
+    click.secho(f"  {title}", fg="cyan")
+
+    if _is_interactive():
+        try:
+            selected, action = _arrow_select_multi(
+                labels,
+                action_indices=action_indices,
+                separator_indices=separator_indices,
+                preselected=preselected,
+                initial_cursor=initial_cursor,
+            )
+        except Exception:
+            selected, action = _numbered_select_multi(
+                labels,
+                action_indices=action_indices,
+                separator_indices=separator_indices,
+                preselected=preselected,
+            )
+    else:
+        selected, action = _numbered_select_multi(
+            labels,
+            action_indices=action_indices,
+            separator_indices=separator_indices,
+            preselected=preselected,
+        )
+    if action_indices is None:
+        return selected
+    return selected, action

lib/cli/src/crewai_cli/utils.py

@@ -24,6 +24,7 @@ __all__ = [
     "build_env_with_all_tool_credentials",
     "build_env_with_tool_repository_credentials",
     "copy_template",
+    "enable_prompt_line_editing",
     "fetch_and_json_env_file",
     "get_project_description",
     "get_project_name",
@@ -40,6 +41,19 @@ __all__ = [
 console = Console()
 
 
+def enable_prompt_line_editing() -> None:
+    """Enable cursor movement/history editing for Click text prompts when available."""
+    try:
+        import readline
+    except ImportError:
+        return
+
+    try:
+        readline.parse_and_bind("set editing-mode emacs")
+    except Exception:  # pragma: no cover - readline backends vary by platform
+        return
+
+
 def copy_template(
     src: Path, dst: Path, name: str, class_name: str, folder_name: str
 ) -> None:

lib/cli/tests/deploy/test_deploy_main.py

@@ -150,6 +150,7 @@ class TestDeployCommand(unittest.TestCase):
     @patch("crewai_cli.deploy.main.fetch_and_json_env_file")
     @patch("crewai_cli.deploy.main.git.Repository.origin_url")
     @patch("builtins.input")
+    @pytest.mark.timeout(180)
     def test_create_crew(self, mock_input, mock_git_origin_url, mock_fetch_env):
         mock_fetch_env.return_value = {"ENV_VAR": "value"}
         mock_git_origin_url.return_value = "https://github.com/test/repo.git"
@@ -165,6 +166,40 @@ class TestDeployCommand(unittest.TestCase):
             self.assertIn("Deployment created successfully!", fake_out.getvalue())
             self.assertIn("new-uuid", fake_out.getvalue())
 
+    @patch("crewai_cli.deploy.main.fetch_and_json_env_file")
+    @patch("crewai_cli.deploy.main.git.Repository")
+    def test_create_crew_without_git_repo_shows_setup_help(
+        self, mock_repository, mock_fetch_env
+    ):
+        mock_fetch_env.return_value = {"ENV_VAR": "value"}
+        mock_repository.side_effect = ValueError("not a Git repository")
+
+        with patch("sys.stdout", new=StringIO()) as fake_out:
+            self.deploy_command.create_crew(skip_validate=True)
+            output = fake_out.getvalue()
+
+        self.assertIn("Deployment requires a Git repository", output)
+        self.assertIn("git init", output)
+        self.assertIn("git remote add origin <your-repo-url>", output)
+        self.mock_client.create_crew.assert_not_called()
+
+    @patch("crewai_cli.deploy.main.fetch_and_json_env_file")
+    @patch("crewai_cli.deploy.main.git.Repository")
+    def test_create_crew_without_remote_shows_remote_help(
+        self, mock_repository, mock_fetch_env
+    ):
+        mock_fetch_env.return_value = {"ENV_VAR": "value"}
+        mock_repository.return_value.origin_url.return_value = None
+
+        with patch("sys.stdout", new=StringIO()) as fake_out:
+            self.deploy_command.create_crew(skip_validate=True)
+            output = fake_out.getvalue()
+
+        self.assertIn("No remote repository URL found.", output)
+        self.assertIn("git remote add origin <your-repo-url>", output)
+        self.assertIn("git push -u origin HEAD", output)
+        self.mock_client.create_crew.assert_not_called()
+
     def test_list_crews(self):
         mock_response = MagicMock()
         mock_response.status_code = 200

lib/cli/tests/deploy/test_validate.py

@@ -110,6 +110,45 @@ def _run_without_import_check(root: Path) -> DeployValidator:
     return v
 
 
+def _scaffold_json_crew(root: Path, *, task_agent: str = "researcher") -> None:
+    (root / "pyproject.toml").write_text(_make_pyproject(name="json_crew"))
+    (root / "uv.lock").write_text("# dummy uv lockfile\n")
+    agents_dir = root / "agents"
+    agents_dir.mkdir()
+    (agents_dir / "researcher.jsonc").write_text(
+        dedent(
+            """
+            {
+              "role": "Researcher",
+              "goal": "Research things",
+              "backstory": "Experienced researcher",
+              "llm": "openai/gpt-4o-mini"
+            }
+            """
+        ).strip()
+        + "\n"
+    )
+    (root / "crew.jsonc").write_text(
+        dedent(
+            f"""
+            {{
+              "name": "json_crew",
+              "agents": ["researcher"],
+              "tasks": [
+                {{
+                  "name": "research",
+                  "description": "Research https://example.com/a//b",
+                  "expected_output": "Findings",
+                  "agent": "{task_agent}"
+                }}
+              ]
+            }}
+            """
+        ).strip()
+        + "\n"
+    )
+
+
 @pytest.mark.parametrize(
     "project_name, expected",
     [
@@ -129,6 +168,38 @@ def test_valid_standard_crew_project_passes(tmp_path: Path) -> None:
     assert v.ok, f"expected clean run, got {v.results}"
 
 
+def test_valid_json_crew_project_passes(tmp_path: Path) -> None:
+    _scaffold_json_crew(tmp_path)
+    v = DeployValidator(project_root=tmp_path)
+    v.run()
+    assert "invalid_crew_json" not in _codes(v)
+
+
+def test_json_task_agent_mismatch_is_error(tmp_path: Path) -> None:
+    _scaffold_json_crew(tmp_path, task_agent="missing_agent")
+    v = DeployValidator(project_root=tmp_path)
+    v.run()
+    finding = next(r for r in v.results if r.code == "invalid_crew_json")
+    assert finding.severity is Severity.ERROR
+    assert "missing_agent" in finding.detail
+
+
+def test_json_runtime_fields_are_deploy_errors(tmp_path: Path) -> None:
+    _scaffold_json_crew(tmp_path)
+    crew_path = tmp_path / "crew.jsonc"
+    crew_path.write_text(
+        crew_path.read_text().replace(
+            '"name": "json_crew",',
+            '"name": "json_crew",\n  "id": "00000000-0000-4000-8000-000000000000",',
+        )
+    )
+    v = DeployValidator(project_root=tmp_path)
+    v.run()
+    finding = next(r for r in v.results if r.code == "invalid_crew_json")
+    assert finding.severity is Severity.ERROR
+    assert "runtime-only" in finding.detail
+
+
 def test_missing_pyproject_errors(tmp_path: Path) -> None:
     v = _run_without_import_check(tmp_path)
     assert "missing_pyproject" in _codes(v)
@@ -426,4 +497,31 @@ def test_create_crew_aborts_on_validation_error(tmp_path: Path) -> None:
         cmd = DeployCommand()
         cmd.create_crew()
         assert not cmd.plus_api_client.create_crew.called
-        del mock_api  # silence unused-var lint
+        del mock_api  # silence unused-var lint
+
+
+def test_is_json_crew_defers_to_declared_flow_type(tmp_path):
+    """A flow project with a stray crew.jsonc must validate as a flow."""
+    (tmp_path / "crew.jsonc").write_text("{}")
+    (tmp_path / "pyproject.toml").write_text(
+        '[project]\nname = "demo"\nversion = "0.1.0"\n\n'
+        '[tool.crewai]\ntype = "flow"\n'
+    )
+
+    assert DeployValidator(project_root=tmp_path)._is_json_crew is False
+
+
+def test_is_json_crew_true_for_declared_crew_type(tmp_path):
+    (tmp_path / "crew.jsonc").write_text("{}")
+    (tmp_path / "pyproject.toml").write_text(
+        '[project]\nname = "demo"\nversion = "0.1.0"\n\n'
+        '[tool.crewai]\ntype = "crew"\n'
+    )
+
+    assert DeployValidator(project_root=tmp_path)._is_json_crew is True
+
+
+def test_is_json_crew_true_without_pyproject(tmp_path):
+    (tmp_path / "crew.jsonc").write_text("{}")
+
+    assert DeployValidator(project_root=tmp_path)._is_json_crew is True

lib/cli/tests/test_cli.py

@@ -94,9 +94,9 @@ def test_version_command_with_tools(runner):
 def test_test_default_iterations(evaluate_crew, runner):
     result = runner.invoke(test)
 
-    evaluate_crew.assert_called_once_with(3, "gpt-4o-mini", trained_agents_file=None)
+    evaluate_crew.assert_called_once_with(3, "gpt-5.4-mini", trained_agents_file=None)
     assert result.exit_code == 0
-    assert "Testing the crew for 3 iterations with model gpt-4o-mini" in result.output
+    assert "Testing the crew for 3 iterations with model gpt-5.4-mini" in result.output
 
 
 @mock.patch("crewai_cli.cli.evaluate_crew")

lib/cli/tests/test_create_crew.py

@@ -6,6 +6,8 @@ from unittest import mock
 
 import pytest
 from click.testing import CliRunner
+import crewai_cli.create_json_crew as json_crew
+import crewai_cli.tui_picker as tui_picker
 from crewai_cli.create_crew import create_crew, create_folder_structure
 
 
@@ -345,3 +347,441 @@ def test_env_vars_are_uppercased_in_env_file(
     env_file_path = crew_path / ".env"
     content = env_file_path.read_text()
     assert "MODEL=" in content
+
+
+def test_json_wizard_defaults_to_sequential_and_memory_enabled(monkeypatch):
+    monkeypatch.setattr(
+        json_crew,
+        "_wizard_agent",
+        lambda **_: {
+            "name": "researcher",
+            "role": "Researcher",
+            "goal": "Research",
+            "backstory": "Researcher",
+            "llm": "openai/gpt-5.5",
+            "tools": [],
+            "planning": False,
+            "allow_delegation": False,
+        },
+    )
+    monkeypatch.setattr(
+        json_crew,
+        "_wizard_task",
+        lambda **_: {
+            "name": "research_task",
+            "description": "Research",
+            "expected_output": "Findings",
+            "agent": "researcher",
+            "context": [],
+        },
+    )
+
+    def confirm(label: str, default: bool = False) -> bool:
+        if label == "Enable crew memory?":
+            return default
+        return False
+
+    monkeypatch.setattr(json_crew, "_confirm", confirm)
+    monkeypatch.setattr(json_crew.click, "prompt", lambda *_, **__: "")
+    monkeypatch.setattr(
+        json_crew,
+        "pick_one",
+        lambda *_args, **_kwargs: pytest.fail("process should not be prompted"),
+    )
+
+    _agents, _tasks, settings = json_crew._wizard_agents_and_tasks(
+        skip_provider=True,
+        default_llm="openai/gpt-5.5",
+    )
+
+    assert settings == {"process": "sequential", "memory": True, "inputs": {}}
+
+
+def test_json_wizard_shows_interpolation_hint(capsys):
+    json_crew._show_interpolation_hint("tasks")
+
+    output = capsys.readouterr().out
+    assert "{placeholder}" in output
+    assert "dynamic values" in output
+    assert "{topic}" not in output
+    assert "Description >" not in output
+    assert '"description"' not in output
+
+
+def test_json_wizard_text_prompt_uses_full_prompt_for_readline(monkeypatch):
+    prompts: list[str] = []
+
+    monkeypatch.setattr(
+        json_crew, "_readline_safe_prompt", lambda prompt: f"safe:{prompt}"
+    )
+    monkeypatch.setattr(
+        "builtins.input", lambda prompt: prompts.append(prompt) or "Draft content"
+    )
+
+    assert json_crew._prompt_text("Goal", spacing_before=False) == "Draft content"
+    assert len(prompts) == 1
+    assert prompts[0].startswith("safe:")
+    assert "Goal" in prompts[0]
+    assert " > " in prompts[0]
+
+
+def test_json_wizard_tool_picker_prioritizes_common_tools(monkeypatch):
+    picker_calls: list[tuple[str, list[str], dict[str, object]]] = []
+
+    def pick_many(title: str, labels: list[str], **kwargs):
+        picker_calls.append((title, labels, kwargs))
+        return [1, 3], None
+
+    monkeypatch.setattr(json_crew, "pick_many", pick_many)
+
+    tools = json_crew._select_tools()
+
+    assert tools == ["SerperDevTool", "DirectoryReadTool"]
+    assert len(picker_calls) == 1
+    labels = picker_calls[0][1]
+    assert 0 in picker_calls[0][2]["separator_indices"]
+    assert labels[0] == "── Common tools ──"
+    assert labels[1].strip().endswith("SerperDevTool")
+    assert labels[2].strip().endswith("ScrapeWebsiteTool")
+    assert labels[3].strip().endswith("DirectoryReadTool")
+    assert labels[4].strip().endswith("FileReadTool")
+    assert labels[5].strip().endswith("FileWriterTool")
+    assert labels[1].index("Google search") < labels[1].index("SerperDevTool")
+    assert "More tools" not in labels
+
+
+def test_json_wizard_tool_picker_collapses_categories_by_default(monkeypatch):
+    picker_calls: list[tuple[str, list[str], dict[str, object]]] = []
+
+    def pick_many(title: str, labels: list[str], **kwargs):
+        picker_calls.append((title, labels, kwargs))
+        return [], None
+
+    monkeypatch.setattr(json_crew, "pick_many", pick_many)
+
+    json_crew._select_tools()
+
+    labels = picker_calls[0][1]
+    action_indices = picker_calls[0][2]["action_indices"]
+    # Categories show as collapsed action rows, not separators with tools
+    assert any(label.startswith("▸ Search & Research") for label in labels)
+    assert any(label.startswith("▸ Web Scraping") for label in labels)
+    assert not any(label.strip().endswith("BraveSearchTool") for label in labels)
+    assert len(action_indices) >= 4
+    # Only the common tools section is visible beyond the category rows
+    assert len(labels) == 1 + 5 + len(action_indices)
+
+
+def test_json_wizard_tool_picker_expands_one_category_at_a_time(monkeypatch):
+    picker_calls: list[tuple[str, list[str], dict[str, object]]] = []
+
+    def find_category_row(labels: list[str], category: str) -> int:
+        return next(
+            idx for idx, label in enumerate(labels) if category in label
+        )
+
+    def pick_many(title: str, labels: list[str], **kwargs):
+        picker_calls.append((title, labels, kwargs))
+        call_num = len(picker_calls)
+        if call_num == 1:
+            return [], find_category_row(labels, "Search & Research")
+        if call_num == 2:
+            # Search & Research is expanded; select BraveSearchTool and
+            # expand Web Scraping instead
+            brave = next(
+                idx
+                for idx, label in enumerate(labels)
+                if label.strip().endswith("BraveSearchTool")
+            )
+            return [brave], find_category_row(labels, "Web Scraping")
+        return [], None
+
+    monkeypatch.setattr(json_crew, "pick_many", pick_many)
+
+    tools = json_crew._select_tools()
+
+    assert tools == ["BraveSearchTool"]
+    assert len(picker_calls) == 3
+    # Second render: Search & Research expanded, others collapsed
+    labels2 = picker_calls[1][1]
+    assert any(label.startswith("▾ Search & Research") for label in labels2)
+    assert any(label.strip().endswith("BraveSearchTool") for label in labels2)
+    assert any(label.startswith("▸ Web Scraping") for label in labels2)
+    # Third render: Web Scraping expanded, Search & Research collapsed again
+    labels3 = picker_calls[2][1]
+    assert any(label.startswith("▸ Search & Research") for label in labels3)
+    assert any(label.startswith("▾ Web Scraping") for label in labels3)
+    assert not any(label.strip().endswith("BraveSearchTool") for label in labels3)
+    # The collapsed Search & Research row reports its selection count
+    assert any(
+        "Search & Research" in label and "1 selected" in label for label in labels3
+    )
+    # Cursor returns to the toggled category row
+    assert picker_calls[2][2]["initial_cursor"] == next(
+        idx for idx, label in enumerate(labels3) if "Web Scraping" in label
+    )
+
+
+def test_json_wizard_tool_picker_preserves_selection_across_renders(monkeypatch):
+    picker_calls: list[tuple[str, list[str], dict[str, object]]] = []
+
+    def pick_many(title: str, labels: list[str], **kwargs):
+        picker_calls.append((title, labels, kwargs))
+        call_num = len(picker_calls)
+        if call_num == 1:
+            # Select a common tool, then expand a category
+            category_row = next(
+                idx for idx, label in enumerate(labels) if "Web Scraping" in label
+            )
+            return [1], category_row
+        # Confirm without touching anything else
+        return sorted(kwargs["preselected"]), None
+
+    monkeypatch.setattr(json_crew, "pick_many", pick_many)
+
+    tools = json_crew._select_tools()
+
+    # The common-tool selection survived the expand re-render via preselected
+    assert tools == ["SerperDevTool"]
+    assert 1 in picker_calls[1][2]["preselected"]
+
+
+def test_json_wizard_tool_picker_lists_builtin_tools_across_categories(monkeypatch):
+    picker_calls: list[tuple[str, list[str], dict[str, object]]] = []
+    expanded_labels: list[str] = []
+
+    def pick_many(title: str, labels: list[str], **kwargs):
+        picker_calls.append((title, labels, kwargs))
+        expanded_labels.extend(labels)
+        action_indices = sorted(kwargs["action_indices"])
+        call_num = len(picker_calls)
+        if call_num <= len(action_indices):
+            # Expand the n-th category (indices shift between renders, so
+            # recompute from this render's action rows)
+            return [], action_indices[call_num - 1]
+        return [], None
+
+    monkeypatch.setattr(json_crew, "pick_many", pick_many)
+
+    json_crew._select_tools()
+
+    tool_names = {
+        label.rsplit(maxsplit=1)[-1]
+        for label in expanded_labels
+        if not label.startswith(("▸", "▾", "──"))
+    }
+
+    assert {
+        "DirectorySearchTool",
+        "MDXSearchTool",
+        "XMLSearchTool",
+        "YoutubeVideoSearchTool",
+        "S3ReaderTool",
+        "E2BExecTool",
+        "TavilyResearchTool",
+        "SerplyNewsSearchTool",
+        "BrowserbaseLoadTool",
+        "PatronusEvalTool",
+    }.issubset(tool_names)
+    assert {
+        "MCPServerAdapter",
+        "MongoDBVectorSearchConfig",
+        "ScrapegraphScrapeToolSchema",
+        "SnowflakeConfig",
+    }.isdisjoint(tool_names)
+
+
+def test_multi_picker_skips_separator_on_initial_cursor(monkeypatch):
+    cursors: list[int] = []
+
+    monkeypatch.setattr(tui_picker, "_read_key", lambda: "enter")
+    monkeypatch.setattr(
+        tui_picker,
+        "_draw_multi",
+        lambda _labels, cursor, *_args, **_kwargs: cursors.append(cursor),
+    )
+    monkeypatch.setattr(tui_picker, "_clear_lines", lambda *_args, **_kwargs: None)
+
+    assert tui_picker._arrow_select_multi(
+        ["── Common tools ──", "Google search via Serper API SerperDevTool"],
+        separator_indices={0},
+    ) == ([], None)
+    assert cursors == [1]
+
+
+def test_json_wizard_agent_attribute_prompts_are_compact(monkeypatch):
+    prompt_calls: list[tuple[str, bool]] = []
+    prompt_values = {
+        "Role": "Senior Dev Rel",
+        "Goal": "Draft content",
+        "Backstory": "Knows developer communities",
+    }
+
+    def prompt_text(
+        label: str,
+        default: str = "",
+        *,
+        spacing_before: bool = True,
+    ) -> str:
+        prompt_calls.append((label, spacing_before))
+        return prompt_values[label]
+
+    monkeypatch.setattr(json_crew, "_prompt_text", prompt_text)
+    monkeypatch.setattr(json_crew, "_select_model", lambda: "openai/gpt-5.5")
+    monkeypatch.setattr(json_crew, "pick_many", lambda *_args, **_kwargs: ([], None))
+    monkeypatch.setattr(json_crew, "_confirm", lambda *_args, **_kwargs: False)
+
+    agent = json_crew._wizard_agent(agent_num=1, existing_names=[])
+
+    assert agent is not None
+    assert prompt_calls == [
+        ("Role", False),
+        ("Goal", False),
+        ("Backstory", False),
+    ]
+
+
+def test_json_wizard_task_attribute_prompts_are_compact(monkeypatch):
+    prompt_calls: list[tuple[str, bool]] = []
+    prompt_values = {
+        "Description": "Research latest release",
+        "Expected output": "Release summary",
+    }
+
+    def prompt_text(
+        label: str,
+        default: str = "",
+        *,
+        spacing_before: bool = True,
+    ) -> str:
+        prompt_calls.append((label, spacing_before))
+        return prompt_values[label]
+
+    monkeypatch.setattr(json_crew, "_prompt_text", prompt_text)
+
+    task = json_crew._wizard_task(
+        task_num=1,
+        agent_names=["senior_dev_rel"],
+        prior_task_names=[],
+    )
+
+    assert task is not None
+    assert prompt_calls == [
+        ("Description", False),
+        ("Expected output", False),
+    ]
+
+
+def test_json_create_provider_preselects_default_model(tmp_path, monkeypatch):
+    monkeypatch.chdir(tmp_path)
+    with mock.patch(
+        "crewai_cli.create_json_crew._wizard_agents_and_tasks"
+    ) as mock_wizard:
+        mock_wizard.return_value = (
+            [
+                {
+                    "name": "researcher",
+                    "role": "Researcher",
+                    "goal": "Research",
+                    "backstory": "Researcher",
+                    "llm": "openai/gpt-5.5",
+                    "tools": [],
+                    "planning": False,
+                    "allow_delegation": False,
+                }
+            ],
+            [
+                {
+                    "name": "research_task",
+                    "description": "Research",
+                    "expected_output": "Findings",
+                    "agent": "researcher",
+                    "context": [],
+                }
+            ],
+            {"process": "sequential", "memory": False, "inputs": {}},
+        )
+
+        json_crew.create_json_crew("JSON Crew", provider="openai", skip_provider=True)
+
+    mock_wizard.assert_called_once_with(
+        skip_provider=True,
+        default_llm="openai/gpt-5.5",
+    )
+    assert (tmp_path / "json_crew" / "crew.jsonc").exists()
+    assert not (tmp_path / "json_crew" / "tests").exists()
+    assert not (tmp_path / "json_crew" / "config.jsonc").exists()
+
+    crew_template = (tmp_path / "json_crew" / "crew.jsonc").read_text()
+    assert (
+        '"guardrail": "Every factual claim needs context support."'
+        in crew_template
+    )
+    assert '"guardrails": [' in crew_template
+    assert '"guardrail_max_retries": 2' in crew_template
+    assert "Docs: https://docs.crewai.com/concepts/tasks" in crew_template
+    assert '"output_pydantic": null' in crew_template
+    assert '"markdown": false' in crew_template
+    assert "Docs: https://docs.crewai.com/concepts/crews" in crew_template
+    assert '"manager_agent": "researcher"' in crew_template
+    assert '"output_log_file": "crew.log"' in crew_template
+    assert "Crew-level LLM fields also accept object form" in crew_template
+    assert '"chat_llm": {"model": "llama3", "provider": "ollama"' in (
+        crew_template
+    )
+    assert "Use {placeholder} in agent or task text" in crew_template
+    assert "`crewai run` prompts for any placeholders" in crew_template
+    assert "Use {placeholder} inputs here" in crew_template
+
+    agent_template = (
+        tmp_path / "json_crew" / "agents" / "researcher.jsonc"
+    ).read_text()
+    assert "You can use {placeholder} inputs in role, goal, or backstory" in (
+        agent_template
+    )
+    assert '"role": "Senior {industry} Researcher"' in agent_template
+    assert "Optional agent-level guardrail" in agent_template
+    assert '"guardrail_max_retries": 2' in agent_template
+    assert "Docs: https://docs.crewai.com/concepts/agents" in agent_template
+    assert '"reasoning": true' in agent_template
+    assert "For custom endpoints or deployment-based providers" in agent_template
+    assert '"deployment_name": "my-deployment", "provider": "azure"' in (
+        agent_template
+    )
+    assert '"planning_config": {' in agent_template
+    assert '"llm": {"model": "deepseek-chat", "provider": "deepseek"}' in (
+        agent_template
+    )
+    assert '"knowledge_sources": []' in agent_template
+
+
+def test_json_provider_default_model_helper():
+    assert json_crew._default_model_for_provider("openai") == "openai/gpt-5.5"
+    assert json_crew._default_model_for_provider("anthropic/claude-custom") == (
+        "anthropic/claude-custom"
+    )
+    assert json_crew._default_model_for_provider("unknown") is None
+
+
+def test_json_wizard_task_reprompts_on_cancelled_agent_pick(monkeypatch):
+    """Esc on the agent picker must reprompt, not silently assign agent 0."""
+    prompts = iter(["Do the research", "A report"])
+    monkeypatch.setattr(json_crew, "_prompt_text", lambda *a, **k: next(prompts))
+
+    pick_calls: list[str] = []
+    picks = iter([-1, 1])
+
+    def fake_pick_one(title: str, labels: list[str]) -> int:
+        pick_calls.append(title)
+        return next(picks)
+
+    monkeypatch.setattr(json_crew, "pick_one", fake_pick_one)
+
+    task = json_crew._wizard_task(
+        task_num=1,
+        agent_names=["first_agent", "second_agent"],
+        prior_task_names=[],
+    )
+
+    assert len(pick_calls) == 2
+    assert task["agent"] == "second_agent"

lib/cli/tests/test_crew_run_tui.py

@@ -0,0 +1,796 @@
+from datetime import datetime
+import time
+
+import pytest
+
+from crewai.events.event_bus import crewai_event_bus
+from crewai.events.types.observation_events import (
+    GoalAchievedEarlyEvent,
+    PlanRefinementEvent,
+    PlanReplanTriggeredEvent,
+    PlanStepCompletedEvent,
+    PlanStepStartedEvent,
+    StepObservationCompletedEvent,
+    StepObservationFailedEvent,
+    StepObservationStartedEvent,
+)
+from crewai.events.types.tool_usage_events import (
+    ToolUsageErrorEvent,
+    ToolUsageFinishedEvent,
+    ToolUsageStartedEvent,
+)
+from crewai_cli import run_crew
+from crewai_cli.crew_run_tui import CrewRunApp
+
+
+def _app_with_plan() -> CrewRunApp:
+    app = CrewRunApp()
+    app._plan = {
+        "plan": "Demo plan",
+        "steps": [
+            {"step_number": 1, "description": "First"},
+            {"step_number": 2, "description": "Second"},
+            {"step_number": 3, "description": "Third"},
+        ],
+    }
+    app._plan_step_status = {1: "pending", 2: "pending", 3: "pending"}
+    return app
+
+
+def _log_entry(name: str) -> dict:
+    now = time.time()
+    return {
+        "tool_name": name,
+        "status": "success",
+        "args": None,
+        "result": f"{name} result",
+        "error": None,
+        "start_time": now,
+        "duration": 1.0,
+        "task_idx": 1,
+    }
+
+
+def _emit_event(event: object) -> None:
+    future = crewai_event_bus.emit(None, event)
+    if future:
+        future.result(timeout=5)
+
+
+def test_chain_deploy_skips_validation_after_auth_retry(monkeypatch) -> None:
+    create_calls: list[dict[str, object]] = []
+    login_calls: list[bool] = []
+
+    class FakeDeployCommand:
+        attempts = 0
+
+        def create_crew(self, **kwargs) -> None:
+            create_calls.append(kwargs)
+            FakeDeployCommand.attempts += 1
+            if FakeDeployCommand.attempts == 1:
+                raise SystemExit(1)
+
+    class FakeAuthenticationCommand:
+        def login(self) -> None:
+            login_calls.append(True)
+
+    monkeypatch.setattr("crewai_cli.deploy.main.DeployCommand", FakeDeployCommand)
+    monkeypatch.setattr(
+        "crewai_cli.authentication.main.AuthenticationCommand",
+        FakeAuthenticationCommand,
+    )
+
+    run_crew._chain_deploy()
+
+    assert create_calls == [
+        {"confirm": False, "skip_validate": True},
+        {"confirm": False, "skip_validate": True},
+    ]
+    assert login_calls == [True]
+
+
+def test_plan_step_status_updates_only_the_explicit_step() -> None:
+    app = _app_with_plan()
+
+    app._set_plan_step_status(2, "done")
+
+    assert app._plan_step_status == {
+        1: "pending",
+        2: "done",
+        3: "pending",
+    }
+
+
+def test_step_observation_events_update_the_explicit_step() -> None:
+    app = _app_with_plan()
+    app._subscribe()
+    try:
+        future = crewai_event_bus.emit(
+            None,
+            StepObservationStartedEvent(
+                agent_role="Agent",
+                step_number=2,
+                step_description="Second",
+            ),
+        )
+        if future:
+            future.result(timeout=5)
+
+        assert app._plan_step_status == {
+            1: "pending",
+            2: "active",
+            3: "pending",
+        }
+
+        future = crewai_event_bus.emit(
+            None,
+            StepObservationCompletedEvent(
+                agent_role="Agent",
+                step_number=2,
+                step_description="Second",
+                step_completed_successfully=True,
+            ),
+        )
+        if future:
+            future.result(timeout=5)
+    finally:
+        app._unsubscribe()
+
+    assert app._plan_step_status == {
+        1: "pending",
+        2: "done",
+        3: "pending",
+    }
+
+
+def test_plan_step_lifecycle_events_update_the_explicit_step() -> None:
+    app = _app_with_plan()
+    app._subscribe()
+    try:
+        _emit_event(
+            PlanStepStartedEvent(
+                agent_role="Agent",
+                step_number=2,
+                step_description="Second",
+            )
+        )
+
+        assert app._plan_step_status == {
+            1: "pending",
+            2: "active",
+            3: "pending",
+        }
+
+        _emit_event(
+            PlanStepCompletedEvent(
+                agent_role="Agent",
+                step_number=2,
+                step_description="Second",
+                success=True,
+                result="done",
+            )
+        )
+    finally:
+        app._unsubscribe()
+
+    assert app._plan_step_status == {
+        1: "pending",
+        2: "done",
+        3: "pending",
+    }
+
+
+def test_failed_plan_step_lifecycle_event_marks_exact_step_failed() -> None:
+    app = _app_with_plan()
+    app._subscribe()
+    try:
+        _emit_event(
+            PlanStepCompletedEvent(
+                agent_role="Agent",
+                step_number=2,
+                step_description="Second",
+                success=False,
+                error="Step failed",
+            )
+        )
+    finally:
+        app._unsubscribe()
+
+    assert app._plan_step_status == {
+        1: "pending",
+        2: "failed",
+        3: "pending",
+    }
+
+
+def test_tool_usage_events_do_not_advance_plan_steps() -> None:
+    app = _app_with_plan()
+    app._subscribe()
+    try:
+        future = crewai_event_bus.emit(
+            None,
+            ToolUsageStartedEvent(tool_name="search", tool_args={"query": "CrewAI"}),
+        )
+        if future:
+            future.result(timeout=5)
+
+        now = datetime.now()
+        future = crewai_event_bus.emit(
+            None,
+            ToolUsageFinishedEvent(
+                tool_name="search",
+                tool_args={"query": "CrewAI"},
+                started_at=now,
+                finished_at=now,
+                output="result",
+            ),
+        )
+        if future:
+            future.result(timeout=5)
+    finally:
+        app._unsubscribe()
+
+    assert app._plan_step_status == {
+        1: "pending",
+        2: "pending",
+        3: "pending",
+    }
+
+
+def test_next_tool_does_not_mark_unfinished_tool_successful() -> None:
+    app = _app_with_plan()
+    app._subscribe()
+    try:
+        _emit_event(
+            ToolUsageStartedEvent(tool_name="search", tool_args={"query": "CrewAI"}),
+        )
+        _emit_event(
+            ToolUsageStartedEvent(tool_name="scrape", tool_args={"url": "https://x"}),
+        )
+    finally:
+        app._unsubscribe()
+
+    assert app._log_entries[0]["status"] == "timeout"
+    assert app._log_entries[0]["result"] is None
+    assert app._log_entries[0]["error"] == (
+        "No result received before the next tool started"
+    )
+    assert app._log_entries[1]["status"] == "running"
+    assert app._plan_step_status == {
+        1: "pending",
+        2: "pending",
+        3: "pending",
+    }
+
+
+def test_internal_reasoning_function_call_is_hidden_from_activity_log() -> None:
+    app = _app_with_plan()
+    app._subscribe()
+    try:
+        future = crewai_event_bus.emit(
+            None,
+            ToolUsageStartedEvent(
+                tool_name="create_reasoning_plan",
+                tool_args={"plan": "Plan", "steps": [], "ready": True},
+            ),
+        )
+        if future:
+            future.result(timeout=5)
+
+        now = datetime.now()
+        future = crewai_event_bus.emit(
+            None,
+            ToolUsageFinishedEvent(
+                tool_name="create_reasoning_plan",
+                tool_args={"plan": "Plan", "steps": [], "ready": True},
+                started_at=now,
+                finished_at=now,
+                output='{"plan":"Plan","steps":[],"ready":true}',
+            ),
+        )
+        if future:
+            future.result(timeout=5)
+
+        future = crewai_event_bus.emit(
+            None,
+            ToolUsageErrorEvent(
+                tool_name="create_reasoning_plan",
+                tool_args={"plan": "Plan", "steps": [], "ready": True},
+                error="internal planning fallback",
+            ),
+        )
+        if future:
+            future.result(timeout=5)
+    finally:
+        app._unsubscribe()
+
+    assert app._log_entries == []
+    assert app._current_task_steps == []
+
+
+def test_tool_failure_does_not_override_successful_plan_step_completion() -> None:
+    app = _app_with_plan()
+    app._subscribe()
+    try:
+        _emit_event(
+            PlanStepStartedEvent(
+                agent_role="Agent",
+                step_number=1,
+                step_description="First",
+            )
+        )
+        _emit_event(
+            ToolUsageStartedEvent(
+                tool_name="search_the_internet_with_serper",
+                tool_args={"search_query": "CrewAI release"},
+                plan_step_number=1,
+                plan_step_description="First",
+            )
+        )
+        _emit_event(
+            ToolUsageErrorEvent(
+                tool_name="search_the_internet_with_serper",
+                tool_args={"search_query": "CrewAI release"},
+                plan_step_number=1,
+                plan_step_description="First",
+                error="No results",
+            )
+        )
+        _emit_event(
+            PlanStepCompletedEvent(
+                agent_role="Agent",
+                step_number=1,
+                step_description="First",
+                success=True,
+                result="Recovered with another source",
+            )
+        )
+    finally:
+        app._unsubscribe()
+
+    assert app._plan_step_status == {
+        1: "done",
+        2: "pending",
+        3: "pending",
+    }
+
+
+def test_tool_event_step_metadata_is_stored_in_activity_log() -> None:
+    app = _app_with_plan()
+    app._subscribe()
+    try:
+        _emit_event(
+            ToolUsageStartedEvent(
+                tool_name="search_the_internet_with_serper",
+                tool_args={"search_query": "CrewAI release"},
+                plan_step_number=2,
+                plan_step_description="Second",
+            )
+        )
+        now = datetime.now()
+        _emit_event(
+            ToolUsageFinishedEvent(
+                tool_name="search_the_internet_with_serper",
+                tool_args={"search_query": "CrewAI release"},
+                plan_step_number=2,
+                plan_step_description="Second",
+                started_at=now,
+                finished_at=now,
+                output="Found official source",
+            )
+        )
+    finally:
+        app._unsubscribe()
+
+    assert app._log_entries[0]["plan_step_number"] == 2
+    assert app._plan_step_status == {
+        1: "pending",
+        2: "pending",
+        3: "pending",
+    }
+
+
+def test_starting_next_tool_does_not_infer_plan_step_progress() -> None:
+    app = _app_with_plan()
+    app._subscribe()
+    try:
+        _emit_event(
+            ToolUsageStartedEvent(
+                tool_name="search_the_internet_with_serper",
+                tool_args={"search_query": "CrewAI release"},
+            )
+        )
+        _emit_event(
+            ToolUsageErrorEvent(
+                tool_name="search_the_internet_with_serper",
+                tool_args={"search_query": "CrewAI release"},
+                error="No results",
+            )
+        )
+        _emit_event(
+            ToolUsageStartedEvent(
+                tool_name="read_website_content",
+                tool_args={"url": "https://example.com"},
+            )
+        )
+    finally:
+        app._unsubscribe()
+
+    assert app._log_entries[0]["status"] == "error"
+    assert app._log_entries[1]["status"] == "running"
+    assert app._plan_step_status == {
+        1: "pending",
+        2: "pending",
+        3: "pending",
+    }
+
+
+@pytest.mark.asyncio
+async def test_crew_done_does_not_mark_unfinished_tool_successful() -> None:
+    app = _app_with_plan()
+
+    async with app.run_test(size=(100, 40)) as pilot:
+        app._plan_step_status = {1: "failed", 2: "done", 3: "pending"}
+        app._log_entries = [
+            {
+                "tool_name": "search",
+                "status": "running",
+                "args": '{"query": "CrewAI"}',
+                "result": None,
+                "error": None,
+                "start_time": time.time() - 2,
+                "duration": None,
+                "task_idx": 1,
+            }
+        ]
+
+        app._on_crew_done("final output")
+        await pilot.pause()
+
+    assert app._log_entries[0]["status"] == "timeout"
+    assert app._log_entries[0]["result"] is None
+    assert app._log_entries[0]["error"] == "No result received before crew completed"
+    assert app._plan_step_status == {1: "failed", 2: "done", 3: "done"}
+
+
+def test_streamed_step_observation_updates_named_step_only() -> None:
+    app = _app_with_plan()
+
+    updated = app._try_parse_step_observation(
+        '{"step_completed_successfully":true,'
+        '"key_information_learned":"Step 2 succeeded with the official source."}'
+    )
+
+    assert updated is True
+    assert app._plan_step_status == {
+        1: "pending",
+        2: "done",
+        3: "pending",
+    }
+
+
+def test_failed_streamed_step_observation_marks_named_step_failed() -> None:
+    app = _app_with_plan()
+
+    updated = app._try_parse_step_observation(
+        '{"step_completed_successfully":false,'
+        '"key_information_learned":"Step 2 failed because the tool failed."}'
+    )
+
+    assert updated is True
+    assert app._plan_step_status == {
+        1: "pending",
+        2: "failed",
+        3: "pending",
+    }
+
+
+def test_streamed_goal_achieved_observation_collapses_remaining_steps_done() -> None:
+    app = _app_with_plan()
+
+    updated = app._try_parse_step_observation(
+        '{"step_number":2,'
+        '"step_completed_successfully":true,'
+        '"key_information_learned":"Goal is already satisfied.",'
+        '"goal_already_achieved":true}'
+    )
+
+    assert updated is True
+    assert app._plan_step_status == {
+        1: "done",
+        2: "done",
+        3: "done",
+    }
+
+
+def test_task_completion_collapses_pending_plan_steps_but_preserves_failed() -> None:
+    app = _app_with_plan()
+    app._plan_step_status = {1: "failed", 2: "done", 3: "pending"}
+
+    app._collapse_plan_on_task_done()
+
+    assert app._plan_step_status == {1: "failed", 2: "done", 3: "done"}
+
+
+def test_observation_failure_collapses_to_done_because_executor_continues() -> None:
+    app = _app_with_plan()
+    app._plan_step_status = {1: "done", 2: "active", 3: "pending"}
+    app._subscribe()
+    try:
+        future = crewai_event_bus.emit(
+            None,
+            StepObservationFailedEvent(
+                agent_role="Agent",
+                step_number=2,
+                step_description="Second",
+                error="observer timeout",
+            ),
+        )
+        if future:
+            future.result(timeout=5)
+    finally:
+        app._unsubscribe()
+
+    assert app._plan_step_status == {
+        1: "done",
+        2: "done",
+        3: "pending",
+    }
+
+
+def test_goal_achieved_event_collapses_remaining_steps_done() -> None:
+    app = _app_with_plan()
+    app._plan_step_status = {1: "done", 2: "active", 3: "pending"}
+    app._subscribe()
+    try:
+        future = crewai_event_bus.emit(
+            None,
+            GoalAchievedEarlyEvent(
+                agent_role="Agent",
+                step_number=2,
+                steps_completed=2,
+                steps_remaining=1,
+            ),
+        )
+        if future:
+            future.result(timeout=5)
+    finally:
+        app._unsubscribe()
+
+    assert app._plan_step_status == {
+        1: "done",
+        2: "done",
+        3: "done",
+    }
+
+
+def test_replan_event_keeps_old_plan_until_next_streamed_plan_replaces_it() -> None:
+    app = _app_with_plan()
+    app._subscribe()
+    try:
+        future = crewai_event_bus.emit(
+            None,
+            PlanReplanTriggeredEvent(
+                agent_role="Agent",
+                step_number=2,
+                replan_reason="Need updated sources",
+                replan_count=1,
+                completed_steps_preserved=1,
+            ),
+        )
+        if future:
+            future.result(timeout=5)
+    finally:
+        app._unsubscribe()
+
+    assert app._plan is not None
+    assert app._plan_step_status == {1: "pending", 2: "pending", 3: "pending"}
+    assert app._awaiting_replan is True
+
+    app._try_parse_plan(
+        '{"plan":"Updated plan","steps":['
+        '{"step_number":1,"description":"Updated first"},'
+        '{"step_number":2,"description":"Updated second"}]}'
+    )
+
+    assert app._plan == {
+        "plan": "Updated plan",
+        "steps": [
+            {"step_number": 1, "description": "Updated first"},
+            {"step_number": 2, "description": "Updated second"},
+        ],
+    }
+    assert app._plan_step_status == {1: "pending", 2: "pending"}
+    assert app._awaiting_replan is False
+
+
+def test_plan_refinement_updates_descriptions_without_new_statuses() -> None:
+    app = _app_with_plan()
+    app._plan_step_status = {1: "done", 2: "active", 3: "pending"}
+    app._subscribe()
+    try:
+        future = crewai_event_bus.emit(
+            None,
+            PlanRefinementEvent(
+                agent_role="Agent",
+                step_number=2,
+                refined_step_count=1,
+                refinements=["Step 3: Write the final answer from verified facts"],
+            ),
+        )
+        if future:
+            future.result(timeout=5)
+    finally:
+        app._unsubscribe()
+
+    assert app._plan_step_status == {
+        1: "done",
+        2: "done",
+        3: "pending",
+    }
+    assert app._plan["steps"][2]["description"] == (
+        "Write the final answer from verified facts"
+    )
+
+
+def test_step_observation_json_is_hidden_from_streaming_text() -> None:
+    app = _app_with_plan()
+
+    assert (
+        app._strip_step_observation_json(
+            'Visible before {"step_completed_successfully":true,'
+            '"key_information_learned":"Step 2 succeeded."} visible after'
+        )
+        == "Visible before  visible after"
+    )
+
+
+@pytest.mark.asyncio
+async def test_completed_run_keeps_activity_log_keyboard_navigation_active() -> None:
+    app = CrewRunApp()
+
+    async with app.run_test(size=(100, 40)) as pilot:
+        app._log_entries = [_log_entry("search"), _log_entry("scrape")]
+
+        app._on_crew_done("final output")
+        await pilot.pause()
+
+        assert app.focused is app.query_one("#log-panel")
+
+        await pilot.press("down", "enter")
+        await pilot.pause()
+
+        assert app._log_cursor == 1
+        assert app._log_expanded == {1}
+
+        await pilot.press("up")
+        await pilot.pause()
+
+        assert app._log_cursor == 0
+
+
+class _FakeTask:
+    fingerprint = None
+
+    def __init__(self, task_id: str, name: str) -> None:
+        self.id = task_id
+        self.name = name
+        self.description = name
+
+
+def test_async_task_completion_marks_the_right_sidebar_row() -> None:
+    """Overlapping tasks: completing task 1 while task 2 runs must not
+    mark task 2 done, and starting task 2 must not mark task 1 done."""
+    from crewai.events.types.task_events import TaskCompletedEvent, TaskStartedEvent
+    from crewai.tasks.task_output import TaskOutput
+
+    app = CrewRunApp(total_tasks=2, task_names=["first", "second"])
+    app._subscribe()
+    try:
+        task1 = _FakeTask("id-1", "first")
+        task2 = _FakeTask("id-2", "second")
+
+        for task in (task1, task2):
+            future = crewai_event_bus.emit(
+                None, TaskStartedEvent(context=None, task=task)
+            )
+            if future:
+                future.result(timeout=5)
+
+        # Both started: neither prematurely done
+        assert app._task_statuses == {1: "active", 2: "active"}
+
+        future = crewai_event_bus.emit(
+            None,
+            TaskCompletedEvent(
+                output=TaskOutput(description="first", raw="done", agent="a"),
+                task=task1,
+            ),
+        )
+        if future:
+            future.result(timeout=5)
+
+        assert app._task_statuses == {1: "done", 2: "active"}
+    finally:
+        app._unsubscribe()
+
+
+def test_pop_task_state_falls_back_to_current_task() -> None:
+    app = CrewRunApp(total_tasks=2, task_names=["first", "second"])
+    app._current_task_idx = 2
+    app._current_task_desc = "second"
+
+    class _Evt:
+        task = None
+        task_name = "unknown"
+
+    state = app._pop_task_state(_Evt())
+    assert state["idx"] == 2
+    assert state["desc"] == "second"
+
+
+def test_overlapping_task_logs_keep_their_own_state() -> None:
+    """Task 1 completing after task 2 started must log its own description,
+    agent, and output — and must not steal or reset task 2's stream state."""
+    from crewai.events.types.task_events import TaskCompletedEvent, TaskStartedEvent
+    from crewai.tasks.task_output import TaskOutput
+
+    app = CrewRunApp(total_tasks=2, task_names=["first", "second"])
+    app._subscribe()
+    try:
+        task1 = _FakeTask("id-1", "first")
+        task2 = _FakeTask("id-2", "second")
+
+        for task in (task1, task2):
+            future = crewai_event_bus.emit(
+                None, TaskStartedEvent(context=None, task=task)
+            )
+            if future:
+                future.result(timeout=5)
+
+        # Task 2 is current and has streamed state in flight
+        app._task_full_output = "task two streaming output"
+        app._current_task_steps = [{"type": "llm", "summary": "thinking"}]
+
+        future = crewai_event_bus.emit(
+            None,
+            TaskCompletedEvent(
+                output=TaskOutput(
+                    description="first", raw="task one result", agent="a1"
+                ),
+                task=task1,
+            ),
+        )
+        if future:
+            future.result(timeout=5)
+
+        # Task 1's entry carries its own identity and output
+        entry1 = app._task_logs[-1]
+        assert entry1["idx"] == 1
+        assert entry1["desc"] == "first"
+        assert entry1["output"] == "task one result"
+        assert entry1["steps"] == []
+
+        # Task 2's in-flight stream state was not consumed or reset
+        assert app._task_full_output == "task two streaming output"
+        assert app._current_task_steps == [{"type": "llm", "summary": "thinking"}]
+
+        future = crewai_event_bus.emit(
+            None,
+            TaskCompletedEvent(
+                output=TaskOutput(
+                    description="second", raw="task two result", agent="a2"
+                ),
+                task=task2,
+            ),
+        )
+        if future:
+            future.result(timeout=5)
+
+        entry2 = app._task_logs[-1]
+        assert entry2["idx"] == 2
+        assert entry2["desc"] == "second"
+        assert entry2["output"] == "task two streaming output"
+        assert any(step.get("summary") == "thinking" for step in entry2["steps"])
+    finally:
+        app._unsubscribe()

lib/cli/tests/test_run_crew.py

@@ -0,0 +1,144 @@
+"""Tests for crewai_cli.run_crew JSON crew handling."""
+
+import os
+from pathlib import Path
+
+import pytest
+from crewai_core.constants import CREWAI_TRAINED_AGENTS_FILE_ENV
+
+import crewai_cli.run_crew as run_crew_module
+
+
+def test_run_crew_forwards_trained_agents_file_to_json_crews(monkeypatch):
+    """crewai run -f must reach JSON crews, not only classic subprocess crews."""
+    monkeypatch.setattr(run_crew_module, "_has_json_crew", lambda: True)
+    called: dict = {}
+
+    def fake_run_json_crew(trained_agents_file=None):
+        called["trained_agents_file"] = trained_agents_file
+
+    monkeypatch.setattr(run_crew_module, "_run_json_crew", fake_run_json_crew)
+
+    run_crew_module.run_crew(trained_agents_file="some.pkl")
+
+    assert called == {"trained_agents_file": "some.pkl"}
+
+
+def test_run_json_crew_exports_trained_agents_env(monkeypatch, tmp_path: Path):
+    """JSON crews run in-process, so the pickle path must land in the env var."""
+    monkeypatch.chdir(tmp_path)
+    monkeypatch.delenv(CREWAI_TRAINED_AGENTS_FILE_ENV, raising=False)
+
+    try:
+        # No crew.json(c) in tmp_path: the loader fails *after* the env var
+        # export, which is the part under test.
+        with pytest.raises(FileNotFoundError):
+            run_crew_module._run_json_crew(trained_agents_file="some.pkl")
+        assert os.environ[CREWAI_TRAINED_AGENTS_FILE_ENV] == "some.pkl"
+    finally:
+        os.environ.pop(CREWAI_TRAINED_AGENTS_FILE_ENV, None)
+
+
+def test_run_json_crew_leaves_env_untouched_without_flag(monkeypatch, tmp_path: Path):
+    monkeypatch.chdir(tmp_path)
+    monkeypatch.delenv(CREWAI_TRAINED_AGENTS_FILE_ENV, raising=False)
+
+    with pytest.raises(FileNotFoundError):
+        run_crew_module._run_json_crew()
+
+    assert CREWAI_TRAINED_AGENTS_FILE_ENV not in os.environ
+
+
+def test_missing_input_names_accepts_hyphenated_placeholders():
+    """The prompt regex must accept the same names kickoff interpolation does."""
+    from types import SimpleNamespace
+
+    crew = SimpleNamespace(
+        agents=[
+            SimpleNamespace(
+                role="Researcher", goal="Cover {my-topic}", backstory=""
+            )
+        ],
+        tasks=[
+            SimpleNamespace(
+                description="Write about {my-topic} for {target-audience}",
+                expected_output="Post",
+                output_file=None,
+            )
+        ],
+    )
+
+    assert run_crew_module._missing_input_names(crew, {}) == [
+        "my-topic",
+        "target-audience",
+    ]
+
+
+def _patch_tui_run(monkeypatch, status: str):
+    """Stub the TUI pieces of _run_json_crew so only exit handling runs."""
+
+    class FakeApp:
+        def __init__(self, **kwargs):
+            self._status = status
+            self._crew_result = "result" if status == "completed" else None
+            self._want_deploy = False
+
+        def run(self):
+            pass
+
+    from types import SimpleNamespace
+
+    crew = SimpleNamespace(name="Demo", tasks=[], agents=[])
+    monkeypatch.setattr(
+        run_crew_module, "find_crew_json_file", lambda: Path("crew.jsonc")
+    )
+    monkeypatch.setattr(
+        run_crew_module,
+        "_load_json_crew_for_tui",
+        lambda _path: (FakeApp, crew, {}, [], []),
+    )
+    monkeypatch.setattr(
+        run_crew_module, "_prompt_for_missing_inputs", lambda _crew, inputs: inputs
+    )
+    monkeypatch.setattr(run_crew_module, "_print_post_tui_summary", lambda _app: None)
+
+
+def test_run_json_crew_failed_status_exits_nonzero(monkeypatch, tmp_path: Path):
+    monkeypatch.chdir(tmp_path)
+    _patch_tui_run(monkeypatch, status="failed")
+
+    with pytest.raises(SystemExit) as exc_info:
+        run_crew_module._run_json_crew()
+
+    assert exc_info.value.code == 1
+
+
+def test_run_json_crew_completed_status_returns_result(monkeypatch, tmp_path: Path):
+    monkeypatch.chdir(tmp_path)
+    _patch_tui_run(monkeypatch, status="completed")
+
+    assert run_crew_module._run_json_crew() == "result"
+
+
+def test_has_json_crew_defers_to_declared_flow_type(monkeypatch, tmp_path: Path):
+    """A flow project containing a stray crew.jsonc must still run as a flow."""
+    monkeypatch.chdir(tmp_path)
+    (tmp_path / "crew.jsonc").write_text("{}")
+    (tmp_path / "pyproject.toml").write_text('[tool.crewai]\ntype = "flow"\n')
+
+    assert run_crew_module._has_json_crew() is False
+
+
+def test_has_json_crew_true_for_declared_crew_type(monkeypatch, tmp_path: Path):
+    monkeypatch.chdir(tmp_path)
+    (tmp_path / "crew.jsonc").write_text("{}")
+    (tmp_path / "pyproject.toml").write_text('[tool.crewai]\ntype = "crew"\n')
+
+    assert run_crew_module._has_json_crew() is True
+
+
+def test_has_json_crew_true_without_pyproject(monkeypatch, tmp_path: Path):
+    monkeypatch.chdir(tmp_path)
+    (tmp_path / "crew.jsonc").write_text("{}")
+
+    assert run_crew_module._has_json_crew() is True

lib/cli/tests/tools/test_main.py

@@ -157,14 +157,16 @@ def test_install_api_error(mock_get, capsys, tool_command):
     mock_get.assert_called_once_with("error-tool")
 
 
-@patch("crewai_cli.tools.main.git.Repository.fetch")
-@patch("crewai_cli.tools.main.git.Repository.is_synced", return_value=False)
-def test_publish_when_not_in_sync(mock_is_synced, mock_fetch, capsys, tool_command):
+@patch("crewai_cli.tools.main.git.Repository")
+def test_publish_when_not_in_sync(mock_repository, capsys, tool_command):
+    mock_repository.return_value.is_synced.return_value = False
+
     with raises(SystemExit):
         tool_command.publish(is_public=True)
 
     output = capsys.readouterr().out
     assert "Local changes need to be resolved before publishing" in output
+    mock_repository.return_value.is_synced.assert_called_once_with()
 
 
 @patch("crewai_cli.tools.main.get_project_name", return_value="sample-tool")

lib/crewai-files/src/crewai_files/core/sources.py

@@ -4,6 +4,7 @@ from __future__ import annotations
 
 from collections.abc import AsyncIterator, Iterator
 import inspect
+import json
 import mimetypes
 from pathlib import Path
 from typing import Annotated, Any, BinaryIO, Protocol, cast, runtime_checkable
@@ -23,6 +24,9 @@ from typing_extensions import TypeIs
 from crewai_files.core.constants import DEFAULT_MAX_FILE_SIZE_BYTES, MAGIC_BUFFER_SIZE
 
 
+OCTET_STREAM = "application/octet-stream"
+
+
 @runtime_checkable
 class AsyncReadable(Protocol):
     """Protocol for async readable streams."""
@@ -56,13 +60,51 @@ class _AsyncReadableValidator:
 ValidatedAsyncReadable = Annotated[AsyncReadable, _AsyncReadableValidator()]
 
 
-def _fallback_content_type(filename: str | None) -> str:
-    """Get content type from filename extension or return default."""
+def _detect_content_type_from_bytes(data: bytes) -> str | None:
+    if data.startswith(b"\x89PNG\r\n\x1a\n"):
+        return "image/png"
+    if data.startswith(b"\xff\xd8\xff"):
+        return "image/jpeg"
+    if data.startswith(b"%PDF-"):
+        return "application/pdf"
+
+    try:
+        decoded = data.decode("utf-8")
+    except UnicodeDecodeError:
+        return None
+
+    stripped = decoded.lstrip()
+    if stripped.startswith(("{", "[")):
+        try:
+            json.loads(decoded)
+            return "application/json"
+        except json.JSONDecodeError:
+            pass
+
+    if "\x00" not in decoded:
+        return "text/plain"
+
+    return None
+
+
+def _fallback_content_type(filename: str | None, data: bytes | None = None) -> str:
+    """Get content type from filename extension, then content sniffing.
+
+    The extension lookup runs first so specific types like ``text/csv`` or
+    ``application/xml`` are not degraded to generic sniffed types such as
+    ``text/plain``; byte sniffing only covers extensionless/unknown names.
+    """
     if filename:
         mime_type, _ = mimetypes.guess_type(filename)
         if mime_type:
             return mime_type
-    return "application/octet-stream"
+
+    if data:
+        content_type = _detect_content_type_from_bytes(data)
+        if content_type:
+            return content_type
+
+    return OCTET_STREAM
 
 
 def generate_filename(content_type: str) -> str:
@@ -97,9 +139,19 @@ def detect_content_type(data: bytes, filename: str | None = None) -> str:
         import magic
 
         result: str = magic.from_buffer(data[:MAGIC_BUFFER_SIZE], mime=True)
-        return result
+        if result != OCTET_STREAM:
+            return result
+        return _fallback_content_type(filename, data)
     except ImportError:
-        return _fallback_content_type(filename)
+        return _fallback_content_type(filename, data)
+
+
+def _read_magic_header(path: Path) -> bytes | None:
+    try:
+        with path.open("rb") as file:
+            return file.read(MAGIC_BUFFER_SIZE)
+    except OSError:
+        return None
 
 
 def detect_content_type_from_path(path: Path, filename: str | None = None) -> str:
@@ -115,13 +167,16 @@ def detect_content_type_from_path(path: Path, filename: str | None = None) -> st
     Returns:
         The detected MIME type.
     """
+    fallback_filename = filename or path.name
     try:
         import magic
 
         result: str = magic.from_file(str(path), mime=True)
-        return result
+        if result != OCTET_STREAM:
+            return result
+        return _fallback_content_type(fallback_filename, _read_magic_header(path))
     except ImportError:
-        return _fallback_content_type(filename or path.name)
+        return _fallback_content_type(fallback_filename, _read_magic_header(path))
 
 
 class _BinaryIOValidator:

lib/crewai-files/src/crewai_files/resolution/resolver.py

@@ -129,6 +129,20 @@ class FileResolver:
         """
         return constraints is not None and constraints.supports_url_references
 
+    @classmethod
+    def _should_resolve_as_url_reference(
+        cls,
+        file: FileInput,
+        provider: ProviderType,
+        constraints: ProviderConstraints | None,
+    ) -> bool:
+        """Check if the provider can accept the current URL source directly."""
+        if not cls._is_url_source(file) or not cls._supports_url(constraints):
+            return False
+
+        provider_lower = provider.lower()
+        return "bedrock" not in provider_lower and "aws" not in provider_lower
+
     @staticmethod
     def _resolve_as_url(file: FileInput) -> UrlReference:
         """Resolve a URL source as UrlReference.
@@ -159,7 +173,7 @@ class FileResolver:
         """
         constraints = get_constraints_for_provider(provider)
 
-        if self._is_url_source(file) and self._supports_url(constraints):
+        if self._should_resolve_as_url_reference(file, provider, constraints):
             return self._resolve_as_url(file)
 
         context = self._build_file_context(file)
@@ -424,7 +438,7 @@ class FileResolver:
         """
         constraints = get_constraints_for_provider(provider)
 
-        if self._is_url_source(file) and self._supports_url(constraints):
+        if self._should_resolve_as_url_reference(file, provider, constraints):
             return self._resolve_as_url(file)
 
         context = self._build_file_context(file)

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

@@ -758,6 +758,31 @@ class Agent(BaseAgent):
         self._check_execution_error(e, task)
         return await self.aexecute_task(task, context, tools)
 
+    def message(self, content: str, **kwargs: Any) -> str:
+        """Send a single message and get a response.
+
+        Creates a temporary Task + Crew, executes, and returns the raw output.
+        """
+        from crewai.crew import Crew
+        from crewai.task import Task
+        from crewai.types.streaming import CrewStreamingOutput
+
+        task = Task(
+            description=content,
+            expected_output="Respond to the user's message appropriately.",
+            agent=self,
+        )
+        crew = Crew(
+            agents=[self],
+            tasks=[task],
+            verbose=self.verbose,
+            memory=self.memory or False,
+        )
+        result = crew.kickoff()
+        if isinstance(result, CrewStreamingOutput):
+            return result.result.raw
+        return result.raw
+
     def execute_task(
         self,
         task: Task,

lib/crewai/src/crewai/agent/planning_config.py

@@ -1,9 +1,10 @@
 from __future__ import annotations
 
-from typing import Literal
+from typing import Annotated, Literal
 
-from pydantic import BaseModel, Field
+from pydantic import BaseModel, BeforeValidator, Field
 
+from crewai.agents.agent_builder.base_agent import _validate_llm_ref
 from crewai.llms.base_llm import BaseLLM
 
 
@@ -69,7 +70,7 @@ class PlanningConfig(BaseModel):
                 max_attempts=3,
                 max_steps=10,
                 plan_prompt="Create a focused plan for: {description}",
-                llm="gpt-4o-mini",
+                llm="gpt-5.4-mini",
             ),
         )
         ```
@@ -139,7 +140,10 @@ class PlanningConfig(BaseModel):
             "whether to continue or replan. None means no per-step timeout."
         ),
     )
-    llm: str | BaseLLM | None = Field(
+    llm: Annotated[
+        str | BaseLLM | None,
+        BeforeValidator(_validate_llm_ref),
+    ] = Field(
         default=None,
         description="LLM to use for planning. Uses agent's LLM if None.",
     )

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

@@ -81,7 +81,7 @@ class OpenAIAgentAdapter(BaseAgentAdapter):
         Raises:
             ImportError: If OpenAI agent dependencies are not installed.
         """
-        self.llm = kwargs.pop("model", "gpt-4o-mini")
+        self.llm = kwargs.pop("model", "gpt-5.4-mini")
         super().__init__(**kwargs)
         self._tool_adapter = OpenAIAgentToolAdapter(tools=kwargs.get("tools"))
         self._converter_adapter = OpenAIConverterAdapter(agent_adapter=self)

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

@@ -85,9 +85,28 @@ def _validate_llm_ref(value: Any) -> Any:
         import inspect
 
         llm_type = value.get("llm_type")
-        if not llm_type or llm_type not in _LLM_TYPE_REGISTRY:
+        if not llm_type:
+            model = (
+                value.get("model")
+                or value.get("model_name")
+                or value.get("deployment_name")
+            )
+            if not model:
+                raise ValueError(
+                    "LLM config objects must include 'model', 'model_name', "
+                    "or 'deployment_name', or a serialized 'llm_type'. "
+                    f"Got keys: {list(value)}"
+                )
+            from crewai.llm import LLM
+
+            llm_kwargs = {**value, "model": model}
+            llm_kwargs.pop("model_name", None)
+            llm_kwargs.pop("deployment_name", None)
+            return LLM(**llm_kwargs)
+
+        if llm_type not in _LLM_TYPE_REGISTRY:
             raise ValueError(
-                f"Unknown or missing llm_type: {llm_type!r}. "
+                f"Unknown llm_type: {llm_type!r}. "
                 f"Expected one of {list(_LLM_TYPE_REGISTRY)}"
             )
         dotted = _LLM_TYPE_REGISTRY[llm_type]
@@ -618,7 +637,10 @@ class BaseAgent(BaseModel, ABC, metaclass=AgentMeta):
         if self.memory is True:
             from crewai.memory.unified_memory import Memory
 
-            self.memory = Memory()
+            memory_kwargs: dict[str, Any] = {}
+            if self.llm is not None:
+                memory_kwargs["llm"] = self.llm
+            self.memory = Memory(**memory_kwargs)
         elif self.memory is False:
             self.memory = None
         return self

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

@@ -53,6 +53,7 @@ from crewai.types.callback import SerializableCallable
 from crewai.utilities.agent_utils import (
     _llm_stop_words_applied,
     aget_llm_response,
+    build_text_tool_calling_fallback_message,
     convert_tools_to_openai_schema,
     enforce_rpm_limit,
     format_message_for_llm,
@@ -64,6 +65,7 @@ from crewai.utilities.agent_utils import (
     handle_unknown_error,
     has_reached_max_iterations,
     is_context_length_exceeded,
+    is_native_tool_calling_unsupported_error,
     parse_tool_call_args,
     process_llm_response,
     track_delegation_if_needed,
@@ -464,6 +466,20 @@ class CrewAgentExecutor(BaseAgentExecutor):
         self._show_logs(formatted_answer)
         return formatted_answer
 
+    def _append_text_tool_calling_fallback_message(self) -> None:
+        """Add text tool-calling instructions after native tools are rejected."""
+        if not self.tools:
+            return
+        self.messages.append(
+            format_message_for_llm(
+                build_text_tool_calling_fallback_message(
+                    self.tools_description,
+                    self.tools_names,
+                ),
+                role="user",
+            )
+        )
+
     def _invoke_loop_native_tools(self) -> AgentFinish:
         """Execute agent loop using native function calling.
 
@@ -557,6 +573,9 @@ class CrewAgentExecutor(BaseAgentExecutor):
                 return formatted_answer
 
             except Exception as e:
+                if is_native_tool_calling_unsupported_error(e):
+                    self._append_text_tool_calling_fallback_message()
+                    return self._invoke_loop_react()
                 if e.__class__.__module__.startswith("litellm"):
                     raise e
                 if is_context_length_exceeded(e):
@@ -1369,6 +1388,9 @@ class CrewAgentExecutor(BaseAgentExecutor):
                 return formatted_answer
 
             except Exception as e:
+                if is_native_tool_calling_unsupported_error(e):
+                    self._append_text_tool_calling_fallback_message()
+                    return await self._ainvoke_loop_react()
                 if e.__class__.__module__.startswith("litellm"):
                     raise e
                 if is_context_length_exceeded(e):

lib/crewai/src/crewai/agents/step_executor.py

@@ -29,14 +29,17 @@ from crewai.events.types.tool_usage_events import (
     ToolUsageStartedEvent,
 )
 from crewai.utilities.agent_utils import (
+    build_text_tool_calling_fallback_message,
     build_tool_calls_assistant_message,
     check_native_tool_support,
     enforce_rpm_limit,
     execute_single_native_tool_call,
     extract_task_section,
     format_message_for_llm,
+    is_native_tool_calling_unsupported_error,
     is_tool_call_list,
     process_llm_response,
+    render_text_description_and_args,
     setup_native_tools,
 )
 from crewai.utilities.i18n import I18N_DEFAULT
@@ -153,6 +156,7 @@ class StepExecutor:
             if self._use_native_tools:
                 result_text = self._execute_native(
                     messages,
+                    todo,
                     tool_calls_made,
                     max_step_iterations=max_step_iterations,
                     step_timeout=step_timeout,
@@ -161,6 +165,7 @@ class StepExecutor:
             else:
                 result_text = self._execute_text_parsed(
                     messages,
+                    todo,
                     tool_calls_made,
                     max_step_iterations=max_step_iterations,
                     step_timeout=step_timeout,
@@ -176,6 +181,46 @@ class StepExecutor:
                 execution_time=elapsed,
             )
         except Exception as e:
+            if self._use_native_tools and is_native_tool_calling_unsupported_error(e):
+                try:
+                    self._use_native_tools = False
+                    self._openai_tools = []
+                    self._available_functions = {}
+                    # Keep the conversation built so far (including any native
+                    # tool round-trips already appended to ``messages``) and
+                    # append the text-tooling instructions instead of
+                    # restarting the step, so completed tool calls are not
+                    # re-executed against a fresh context.
+                    messages.append(
+                        format_message_for_llm(
+                            build_text_tool_calling_fallback_message(
+                                render_text_description_and_args(self.tools),
+                                ", ".join(
+                                    sanitize_tool_name(t.name) for t in self.tools
+                                ),
+                            ),
+                            role="user",
+                        )
+                    )
+                    result_text = self._execute_text_parsed(
+                        messages,
+                        todo,
+                        tool_calls_made,
+                        max_step_iterations=max_step_iterations,
+                        step_timeout=step_timeout,
+                        start_time=start_time,
+                    )
+                    self._validate_expected_tool_usage(todo, tool_calls_made)
+                    elapsed = time.monotonic() - start_time
+                    return StepResult(
+                        success=True,
+                        result=result_text,
+                        tool_calls_made=tool_calls_made,
+                        execution_time=elapsed,
+                    )
+                except Exception as fallback_error:
+                    e = fallback_error
+
             elapsed = time.monotonic() - start_time
             return StepResult(
                 success=False,
@@ -272,6 +317,7 @@ class StepExecutor:
     def _execute_text_parsed(
         self,
         messages: list[LLMMessage],
+        todo: TodoItem,
         tool_calls_made: list[str],
         max_step_iterations: int = 15,
         step_timeout: int | None = None,
@@ -310,7 +356,7 @@ class StepExecutor:
 
             if isinstance(formatted, AgentAction):
                 tool_calls_made.append(formatted.tool)
-                tool_result = self._execute_text_tool_with_events(formatted)
+                tool_result = self._execute_text_tool_with_events(formatted, todo)
                 last_tool_result = tool_result
                 messages.append({"role": "assistant", "content": answer_str})
                 messages.append(self._build_observation_message(tool_result))
@@ -320,7 +366,9 @@ class StepExecutor:
 
         return last_tool_result
 
-    def _execute_text_tool_with_events(self, formatted: AgentAction) -> str:
+    def _execute_text_tool_with_events(
+        self, formatted: AgentAction, todo: TodoItem
+    ) -> str:
         """Execute text-parsed tool calls with tool usage events."""
         args_dict = self._parse_tool_args(formatted.tool_input)
         agent_key = getattr(self.agent, "key", "unknown") if self.agent else "unknown"
@@ -333,6 +381,8 @@ class StepExecutor:
                 from_agent=self.agent,
                 from_task=self.task,
                 agent_key=agent_key,
+                plan_step_number=todo.step_number,
+                plan_step_description=todo.description,
             ),
         )
 
@@ -368,6 +418,8 @@ class StepExecutor:
                     from_agent=self.agent,
                     from_task=self.task,
                     agent_key=agent_key,
+                    plan_step_number=todo.step_number,
+                    plan_step_description=todo.description,
                     error=e,
                 ),
             )
@@ -382,6 +434,8 @@ class StepExecutor:
                 from_agent=self.agent,
                 from_task=self.task,
                 agent_key=agent_key,
+                plan_step_number=todo.step_number,
+                plan_step_description=todo.description,
                 started_at=started_at,
                 finished_at=datetime.now(),
             ),
@@ -474,6 +528,7 @@ class StepExecutor:
     def _execute_native(
         self,
         messages: list[LLMMessage],
+        todo: TodoItem,
         tool_calls_made: list[str],
         max_step_iterations: int = 15,
         step_timeout: int | None = None,
@@ -513,7 +568,7 @@ class StepExecutor:
 
             if isinstance(answer, list) and answer and is_tool_call_list(answer):
                 result = self._execute_native_tool_calls(
-                    answer, messages, tool_calls_made
+                    answer, messages, todo, tool_calls_made
                 )
                 accumulated_results.append(result)
                 continue
@@ -526,6 +581,7 @@ class StepExecutor:
         self,
         tool_calls: list[Any],
         messages: list[LLMMessage],
+        todo: TodoItem,
         tool_calls_made: list[str],
     ) -> str:
         """Execute a batch of native tool calls and return their results.
@@ -551,6 +607,8 @@ class StepExecutor:
                 event_source=self,
                 printer=PRINTER,
                 verbose=bool(self.agent and self.agent.verbose),
+                plan_step_number=todo.step_number,
+                plan_step_description=todo.description,
             )
 
             if call_result.func_name:

lib/crewai/src/crewai/crew.py

@@ -658,7 +658,14 @@ class Crew(FlowTrackable, BaseModel):
                 from crewai.rag.embeddings.factory import build_embedder
 
                 embedder = build_embedder(cast(dict[str, Any], self.embedder))
-            self._memory = Memory(embedder=embedder, root_scope=crew_root_scope)
+            memory_kwargs: dict[str, Any] = {
+                "embedder": embedder,
+                "root_scope": crew_root_scope,
+            }
+            memory_llm = self._memory_llm()
+            if memory_llm is not None:
+                memory_kwargs["llm"] = memory_llm
+            self._memory = Memory(**memory_kwargs)
         elif self.memory:
             # User passed a Memory / MemoryScope / MemorySlice instance
             # Respect user's configuration — don't auto-set root_scope
@@ -668,6 +675,16 @@ class Crew(FlowTrackable, BaseModel):
 
         return self
 
+    def _memory_llm(self) -> str | BaseLLM | None:
+        """Return the LLM auto-created memory should use for analysis."""
+        if self.chat_llm is not None:
+            return self.chat_llm
+        for agent in self.agents:
+            agent_llm: str | BaseLLM | None = getattr(agent, "llm", None)
+            if agent_llm is not None:
+                return agent_llm
+        return None
+
     @model_validator(mode="after")
     def create_crew_knowledge(self) -> Crew:
         """Create the knowledge for the crew."""

lib/crewai/src/crewai/events/__init__.py

@@ -116,6 +116,11 @@ if TYPE_CHECKING:
         MemorySaveFailedEvent,
         MemorySaveStartedEvent,
     )
+    from crewai.events.types.observation_events import (
+        PlanStepCompletedEvent,
+        PlanStepEvent,
+        PlanStepStartedEvent,
+    )
     from crewai.events.types.reasoning_events import (
         AgentReasoningCompletedEvent,
         AgentReasoningFailedEvent,
@@ -220,6 +225,9 @@ _LAZY_EVENT_MAPPING: dict[str, str] = {
     "MemorySaveCompletedEvent": "crewai.events.types.memory_events",
     "MemorySaveFailedEvent": "crewai.events.types.memory_events",
     "MemorySaveStartedEvent": "crewai.events.types.memory_events",
+    "PlanStepCompletedEvent": "crewai.events.types.observation_events",
+    "PlanStepEvent": "crewai.events.types.observation_events",
+    "PlanStepStartedEvent": "crewai.events.types.observation_events",
     "AgentReasoningCompletedEvent": "crewai.events.types.reasoning_events",
     "AgentReasoningFailedEvent": "crewai.events.types.reasoning_events",
     "AgentReasoningStartedEvent": "crewai.events.types.reasoning_events",
@@ -349,6 +357,9 @@ __all__ = [
     "MethodExecutionFailedEvent",
     "MethodExecutionFinishedEvent",
     "MethodExecutionStartedEvent",
+    "PlanStepCompletedEvent",
+    "PlanStepEvent",
+    "PlanStepStartedEvent",
     "ReasoningEvent",
     "SkillActivatedEvent",
     "SkillDiscoveryCompletedEvent",

lib/crewai/src/crewai/events/event_types.py

@@ -99,6 +99,10 @@ from crewai.events.types.memory_events import (
     MemorySaveFailedEvent,
     MemorySaveStartedEvent,
 )
+from crewai.events.types.observation_events import (
+    PlanStepCompletedEvent,
+    PlanStepStartedEvent,
+)
 from crewai.events.types.reasoning_events import (
     AgentReasoningCompletedEvent,
     AgentReasoningFailedEvent,
@@ -191,6 +195,8 @@ EventTypes = (
     | MemoryRetrievalStartedEvent
     | MemoryRetrievalCompletedEvent
     | MemoryRetrievalFailedEvent
+    | PlanStepStartedEvent
+    | PlanStepCompletedEvent
     | MCPConnectionStartedEvent
     | MCPConnectionCompletedEvent
     | MCPConnectionFailedEvent

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

@@ -24,6 +24,7 @@ from crewai.events.listeners.tracing.types import TraceEvent
 from crewai.events.listeners.tracing.utils import (
     get_user_id,
     is_tracing_enabled_in_context,
+    is_tui_mode,
     should_auto_collect_first_time_traces,
 )
 from crewai.plus_api import PlusAPI
@@ -74,6 +75,7 @@ class TraceBatchManager:
         self.defer_session_finalization: bool = False
         self._batch_finalized: bool = False
         self.backend_initialized: bool = False
+        self.trace_url: str | None = None
         self.ephemeral_trace_url: str | None = None
         try:
             self.plus_api = PlusAPI(
@@ -108,7 +110,9 @@ class TraceBatchManager:
 
             self.record_start_time("execution")
 
-            if should_auto_collect_first_time_traces():
+            if should_auto_collect_first_time_traces() or (
+                is_tui_mode() and not is_tracing_enabled_in_context()
+            ):
                 self.trace_batch_id = self.current_batch.batch_id
             else:
                 self._initialize_backend_batch(
@@ -411,6 +415,7 @@ class TraceBatchManager:
                         else f"{base_url}/crewai_plus/ephemeral_trace_batches/{batch_id}?access_code={access_code}"
                     )
 
+                    self.trace_url = return_link
                     if is_ephemeral:
                         self.ephemeral_trace_url = return_link
 
@@ -428,7 +433,10 @@ class TraceBatchManager:
                         title="Trace Batch Finalization",
                         border_style="green",
                     )
-                    if not should_auto_collect_first_time_traces():
+                    if (
+                        not should_auto_collect_first_time_traces()
+                        and not is_tui_mode()
+                    ):
                         console.print(panel)
                     return True
 

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

@@ -18,6 +18,7 @@ from crewai.events.listeners.tracing.trace_batch_manager import TraceBatchManage
 from crewai.events.listeners.tracing.types import TraceEvent
 from crewai.events.listeners.tracing.utils import (
     is_tracing_enabled_in_context,
+    is_tui_mode,
     safe_serialize_to_dict,
     should_auto_collect_first_time_traces,
     should_enable_tracing,
@@ -212,8 +213,8 @@ class TraceCollectionListener(BaseEventListener):
             not should_enable_tracing()
             and not is_tracing_enabled_in_context()
             and not should_auto_collect_first_time_traces()
+            and not is_tui_mode()
         ):
-            self._listeners_setup = True
             return
 
         self._register_flow_event_handlers(crewai_event_bus)
@@ -297,6 +298,12 @@ class TraceCollectionListener(BaseEventListener):
             if self._nested_in_flow_execution():
                 return
             if self.batch_manager.batch_owner_type == "crew":
+                if is_tui_mode():
+                    if self.first_time_handler.is_first_time:
+                        self.first_time_handler.mark_events_collected()
+                    elif is_tracing_enabled_in_context() or should_enable_tracing():
+                        self.batch_manager.finalize_batch()
+                    return
                 if self.first_time_handler.is_first_time:
                     self.first_time_handler.mark_events_collected()
                     self.first_time_handler.handle_execution_completion()
@@ -310,6 +317,12 @@ class TraceCollectionListener(BaseEventListener):
                 return
             if self._nested_in_flow_execution():
                 return
+            if is_tui_mode():
+                if self.first_time_handler.is_first_time:
+                    self.first_time_handler.mark_events_collected()
+                elif is_tracing_enabled_in_context() or should_enable_tracing():
+                    self.batch_manager.finalize_batch()
+                return
             if self.first_time_handler.is_first_time:
                 self.first_time_handler.mark_events_collected()
                 self.first_time_handler.handle_execution_completion()

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

@@ -42,6 +42,7 @@ __all__ = [
     "is_first_execution",
     "is_tracing_enabled",
     "is_tracing_enabled_in_context",
+    "is_tui_mode",
     "mark_first_execution_completed",
     "mark_first_execution_done",
     "on_first_execution_tracing_confirmation",
@@ -50,6 +51,7 @@ __all__ = [
     "safe_serialize_to_dict",
     "set_suppress_tracing_messages",
     "set_tracing_enabled",
+    "set_tui_mode",
     "should_auto_collect_first_time_traces",
     "should_enable_tracing",
     "should_suppress_tracing_messages",
@@ -71,6 +73,16 @@ _suppress_tracing_messages: ContextVar[bool] = ContextVar(
     "_suppress_tracing_messages", default=False
 )
 
+_tui_mode: ContextVar[bool] = ContextVar("_tui_mode", default=False)
+
+
+def set_tui_mode(enabled: bool) -> object:
+    return _tui_mode.set(enabled)
+
+
+def is_tui_mode() -> bool:
+    return _tui_mode.get()
+
 
 def set_suppress_tracing_messages(suppress: bool) -> object:
     """Set whether to suppress tracing-related console messages.

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

@@ -26,6 +26,38 @@ class ObservationEvent(BaseEvent):
         self._set_agent_params(data)
 
 
+class PlanStepEvent(BaseEvent):
+    """Base event for authoritative plan step lifecycle updates."""
+
+    type: str
+    agent_role: str
+    step_number: int
+    step_description: str = ""
+    tool_to_use: str | None = None
+    from_task: Any | None = None
+    from_agent: Any | None = None
+
+    def __init__(self, **data: Any) -> None:
+        super().__init__(**data)
+        self._set_task_params(data)
+        self._set_agent_params(data)
+
+
+class PlanStepStartedEvent(PlanStepEvent):
+    """Emitted when a concrete plan step starts executing."""
+
+    type: Literal["plan_step_started"] = "plan_step_started"
+
+
+class PlanStepCompletedEvent(PlanStepEvent):
+    """Emitted when a concrete plan step reaches a terminal state."""
+
+    type: Literal["plan_step_completed"] = "plan_step_completed"
+    success: bool = True
+    result: str | None = None
+    error: str | None = None
+
+
 class StepObservationStartedEvent(ObservationEvent):
     """Emitted when the Planner begins observing a step's result.
 

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

@@ -21,6 +21,8 @@ class ToolUsageEvent(BaseEvent):
     agent: Any | None = None
     task_name: str | None = None
     task_id: str | None = None
+    plan_step_number: int | None = None
+    plan_step_description: str | None = None
     from_task: Any | None = None
     from_agent: Any | None = None
 

lib/crewai/src/crewai/experimental/agent_executor.py

@@ -46,6 +46,8 @@ from crewai.events.types.observation_events import (
     GoalAchievedEarlyEvent,
     PlanRefinementEvent,
     PlanReplanTriggeredEvent,
+    PlanStepCompletedEvent,
+    PlanStepStartedEvent,
 )
 from crewai.events.types.tool_usage_events import (
     ToolUsageErrorEvent,
@@ -73,6 +75,7 @@ from crewai.tools.base_tool import BaseTool
 from crewai.tools.structured_tool import CrewStructuredTool
 from crewai.utilities.agent_utils import (
     _llm_stop_words_applied,
+    build_text_tool_calling_fallback_message,
     check_native_tool_support,
     enforce_rpm_limit,
     extract_tool_call_info,
@@ -86,6 +89,7 @@ from crewai.utilities.agent_utils import (
     has_reached_max_iterations,
     is_context_length_exceeded,
     is_inside_event_loop,
+    is_native_tool_calling_unsupported_error,
     is_tool_call_list,
     parse_tool_call_args,
     process_llm_response,
@@ -241,6 +245,23 @@ class AgentExecutor(Flow[AgentExecutorState], BaseAgentExecutor):
                 self._tool_name_mapping,
             ) = setup_native_tools(self.original_tools)
 
+    def _downgrade_to_text_tool_calling(self) -> None:
+        """Switch a running execution from native tools to text tool calls."""
+        self.state.use_native_tools = False
+        self.state.pending_tool_calls.clear()
+        self._openai_tools = []
+        self._available_functions = {}
+        if self.tools:
+            self.state.messages.append(
+                format_message_for_llm(
+                    build_text_tool_calling_fallback_message(
+                        self.tools_description,
+                        self.tools_names,
+                    ),
+                    role="user",
+                )
+            )
+
     def _is_tool_call_list(self, response: list[Any]) -> bool:
         """Check if a response is a list of tool calls."""
         return is_tool_call_list(response)
@@ -349,6 +370,84 @@ class AgentExecutor(Flow[AgentExecutorState], BaseAgentExecutor):
 
         self.state.todos = TodoList(items=todos)
 
+    def _emit_plan_step_started(self, todo: TodoItem) -> None:
+        try:
+            crewai_event_bus.emit(
+                self.agent,
+                event=PlanStepStartedEvent(
+                    agent_role=self.agent.role,
+                    step_number=todo.step_number,
+                    step_description=todo.description,
+                    tool_to_use=todo.tool_to_use,
+                    from_task=self.task,
+                    from_agent=self.agent,
+                ),
+            )
+        except Exception:  # noqa: S110
+            pass
+
+    def _emit_plan_step_completed(
+        self,
+        todo: TodoItem,
+        *,
+        success: bool,
+        result: str | None = None,
+        error: str | None = None,
+    ) -> None:
+        try:
+            crewai_event_bus.emit(
+                self.agent,
+                event=PlanStepCompletedEvent(
+                    agent_role=self.agent.role,
+                    step_number=todo.step_number,
+                    step_description=todo.description,
+                    tool_to_use=todo.tool_to_use,
+                    success=success,
+                    result=result,
+                    error=error,
+                    from_task=self.task,
+                    from_agent=self.agent,
+                ),
+            )
+        except Exception:  # noqa: S110
+            pass
+
+    def _mark_todo_running(self, todo: TodoItem) -> None:
+        previous_status = todo.status
+        self.state.todos.mark_running(todo.step_number)
+        if previous_status != "running":
+            self._emit_plan_step_started(todo)
+
+    def _mark_todo_completed(
+        self,
+        step_number: int,
+        result: str | None = None,
+    ) -> None:
+        todo = self.state.todos.get_by_step_number(step_number)
+        previous_status = todo.status if todo else None
+        self.state.todos.mark_completed(step_number, result=result)
+        todo = self.state.todos.get_by_step_number(step_number)
+        if todo and previous_status != "completed":
+            self._emit_plan_step_completed(todo, success=True, result=result)
+
+    def _mark_todo_failed(
+        self,
+        step_number: int,
+        result: str | None = None,
+        error: str | None = None,
+    ) -> None:
+        todo = self.state.todos.get_by_step_number(step_number)
+        previous_status = todo.status if todo else None
+        self.state.todos.mark_failed(step_number, result=result)
+        todo = self.state.todos.get_by_step_number(step_number)
+        if todo and previous_status != "failed":
+            self._emit_plan_step_completed(
+                todo,
+                success=False,
+                result=result,
+                error=error,
+            )
+
     def _ensure_step_executor(self) -> Any:
         """Lazily create the StepExecutor (avoids circular imports)."""
         if self._step_executor is None:
@@ -597,8 +696,10 @@ class AgentExecutor(Flow[AgentExecutorState], BaseAgentExecutor):
             and not observation.step_completed_successfully
             and observation.needs_full_replan
         ):
-            self.state.todos.mark_failed(
-                current_todo.step_number, result=current_todo.result
+            self._mark_todo_failed(
+                current_todo.step_number,
+                result=current_todo.result,
+                error=observation.replan_reason,
             )
             if self.agent.verbose:
                 PRINTER.print(
@@ -614,8 +715,9 @@ class AgentExecutor(Flow[AgentExecutorState], BaseAgentExecutor):
             return "replan_now"
 
         if observation and not observation.step_completed_successfully:
-            self.state.todos.mark_failed(
-                current_todo.step_number, result=current_todo.result
+            self._mark_todo_failed(
+                current_todo.step_number,
+                result=current_todo.result,
             )
             if self.agent.verbose:
                 failed = len(self.state.todos.get_failed_todos())
@@ -629,9 +731,7 @@ class AgentExecutor(Flow[AgentExecutorState], BaseAgentExecutor):
                 )
             return "continue_plan"
 
-        self.state.todos.mark_completed(
-            current_todo.step_number, result=current_todo.result
-        )
+        self._mark_todo_completed(current_todo.step_number, result=current_todo.result)
 
         if self.agent.verbose:
             completed = self.state.todos.completed_count
@@ -661,7 +761,7 @@ class AgentExecutor(Flow[AgentExecutorState], BaseAgentExecutor):
 
         # If observation is missing or step succeeded — continue
         if not observation or observation.step_completed_successfully:
-            self.state.todos.mark_completed(
+            self._mark_todo_completed(
                 current_todo.step_number, result=current_todo.result
             )
             if self.agent.verbose:
@@ -676,8 +776,10 @@ class AgentExecutor(Flow[AgentExecutorState], BaseAgentExecutor):
         # Step failed — only replan if observer explicitly requires it,
         # otherwise mark done and continue (same gate as low-effort).
         if observation.needs_full_replan:
-            self.state.todos.mark_failed(
-                current_todo.step_number, result=current_todo.result
+            self._mark_todo_failed(
+                current_todo.step_number,
+                result=current_todo.result,
+                error=observation.replan_reason,
             )
             if self.agent.verbose:
                 PRINTER.print(
@@ -694,9 +796,7 @@ class AgentExecutor(Flow[AgentExecutorState], BaseAgentExecutor):
 
         # Step failed but observer does not require a full replan — mark as
         # failed (not completed) so get_failed_todos() tracks it correctly.
-        self.state.todos.mark_failed(
-            current_todo.step_number, result=current_todo.result
-        )
+        self._mark_todo_failed(current_todo.step_number, result=current_todo.result)
         if self.agent.verbose:
             failed = len(self.state.todos.get_failed_todos())
             total = len(self.state.todos.items)
@@ -731,12 +831,12 @@ class AgentExecutor(Flow[AgentExecutorState], BaseAgentExecutor):
         observation = self.state.observations.get(current_todo.step_number)
         if not observation:
             # No observation available — default to continue
-            self.state.todos.mark_completed(current_todo.step_number)
+            self._mark_todo_completed(current_todo.step_number)
             return "continue_plan"
 
         # Goal already achieved — early termination
         if observation.goal_already_achieved:
-            self.state.todos.mark_completed(
+            self._mark_todo_completed(
                 current_todo.step_number, result=current_todo.result
             )
             if self.agent.verbose:
@@ -748,8 +848,10 @@ class AgentExecutor(Flow[AgentExecutorState], BaseAgentExecutor):
 
         # Full replan needed
         if observation.needs_full_replan:
-            self.state.todos.mark_failed(
-                current_todo.step_number, result=current_todo.result
+            self._mark_todo_failed(
+                current_todo.step_number,
+                result=current_todo.result,
+                error=observation.replan_reason,
             )
             if self.agent.verbose:
                 PRINTER.print(
@@ -761,9 +863,7 @@ class AgentExecutor(Flow[AgentExecutorState], BaseAgentExecutor):
 
         # Step failed — also trigger replan
         if not observation.step_completed_successfully:
-            self.state.todos.mark_failed(
-                current_todo.step_number, result=current_todo.result
-            )
+            self._mark_todo_failed(current_todo.step_number, result=current_todo.result)
             if self.agent.verbose:
                 PRINTER.print(
                     content="[Decide] Step failed — triggering replan",
@@ -773,7 +873,7 @@ class AgentExecutor(Flow[AgentExecutorState], BaseAgentExecutor):
             return "replan_now"
 
         if observation.remaining_plan_still_valid and observation.suggested_refinements:
-            self.state.todos.mark_completed(
+            self._mark_todo_completed(
                 current_todo.step_number, result=current_todo.result
             )
             if self.agent.verbose:
@@ -783,9 +883,7 @@ class AgentExecutor(Flow[AgentExecutorState], BaseAgentExecutor):
                 )
             return "refine_and_continue"
 
-        self.state.todos.mark_completed(
-            current_todo.step_number, result=current_todo.result
-        )
+        self._mark_todo_completed(current_todo.step_number, result=current_todo.result)
         if self.agent.verbose:
             completed = self.state.todos.completed_count
             total = len(self.state.todos.items)
@@ -961,7 +1059,7 @@ class AgentExecutor(Flow[AgentExecutorState], BaseAgentExecutor):
             return "needs_replan"
 
         if len(ready) == 1:
-            self.state.todos.mark_running(ready[0].step_number)
+            self._mark_todo_running(ready[0])
             return "single_todo_ready"
 
         return "multiple_todos_ready"
@@ -1099,7 +1197,7 @@ class AgentExecutor(Flow[AgentExecutorState], BaseAgentExecutor):
 
         # Mark all ready todos as running
         for todo in ready:
-            self.state.todos.mark_running(todo.step_number)
+            self._mark_todo_running(todo)
 
         # Build context and executor for each todo, then run in parallel
         async def _run_step(todo: TodoItem) -> tuple[TodoItem, object]:
@@ -1127,7 +1225,11 @@ class AgentExecutor(Flow[AgentExecutorState], BaseAgentExecutor):
             if isinstance(item, BaseException):
                 error_msg = f"Error: {item!s}"
                 todo.result = error_msg
-                self.state.todos.mark_failed(todo.step_number, result=error_msg)
+                self._mark_todo_failed(
+                    todo.step_number,
+                    result=error_msg,
+                    error=error_msg,
+                )
                 if self.agent.verbose:
                     PRINTER.print(
                         content=f"Todo {todo.step_number} failed: {error_msg}",
@@ -1197,9 +1299,9 @@ class AgentExecutor(Flow[AgentExecutorState], BaseAgentExecutor):
 
             # Mark based on observation result
             if observation.step_completed_successfully:
-                self.state.todos.mark_completed(todo.step_number, result=todo.result)
+                self._mark_todo_completed(todo.step_number, result=todo.result)
             else:
-                self.state.todos.mark_failed(todo.step_number, result=todo.result)
+                self._mark_todo_failed(todo.step_number, result=todo.result)
 
             if self.agent.verbose:
                 PRINTER.print(
@@ -1349,7 +1451,11 @@ class AgentExecutor(Flow[AgentExecutorState], BaseAgentExecutor):
     def call_llm_native_tools(
         self,
     ) -> Literal[
-        "native_tool_calls", "native_finished", "context_error", "todo_satisfied"
+        "native_tool_calls",
+        "native_finished",
+        "context_error",
+        "todo_satisfied",
+        "continue_reasoning",
     ]:
         """Execute LLM call with native function calling.
 
@@ -1428,6 +1534,9 @@ class AgentExecutor(Flow[AgentExecutorState], BaseAgentExecutor):
             return self._route_finish_with_todos("native_finished")
 
         except Exception as e:
+            if is_native_tool_calling_unsupported_error(e):
+                self._downgrade_to_text_tool_calling()
+                return "continue_reasoning"
             if is_context_length_exceeded(e):
                 self._last_context_error = e
                 return "context_error"
@@ -2085,7 +2194,7 @@ class AgentExecutor(Flow[AgentExecutorState], BaseAgentExecutor):
             step_number: The step number to mark.
             result: The result of the todo.
         """
-        self.state.todos.mark_completed(step_number, result=result)
+        self._mark_todo_completed(step_number, result=result)
 
         if self.agent.verbose:
             completed = self.state.todos.completed_count

lib/crewai/src/crewai/flow/async_feedback/__init__.py

@@ -20,7 +20,7 @@ Example:
         @human_feedback(
             message="Review this:",
             emit=["approved", "rejected"],
-            llm="gpt-4o-mini",
+            llm="gpt-5.4-mini",
             provider=SlackProvider(),
         )
         def review(self):

lib/crewai/src/crewai/flow/async_feedback/types.py

@@ -47,7 +47,7 @@ class PendingFeedbackContext:
             method_output={"title": "Draft", "body": "..."},
             message="Please review and approve or reject:",
             emit=["approved", "rejected"],
-            llm="gpt-4o-mini",
+            llm="gpt-5.4-mini",
         )
         ```
     """

lib/crewai/src/crewai/flow/dsl/_human_feedback.py

@@ -23,7 +23,7 @@ __all__ = ["HumanFeedbackResult", "human_feedback"]
 def human_feedback(
     message: str,
     emit: Sequence[str] | None = None,
-    llm: str | BaseLLM | None = "gpt-4o-mini",
+    llm: str | BaseLLM | None = "gpt-5.4-mini",
     default_outcome: str | None = None,
     metadata: dict[str, Any] | None = None,
     provider: HumanFeedbackProvider | None = None,

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

@@ -11,9 +11,17 @@ from __future__ import annotations
 
 import json
 import logging
+import re
 from typing import Any, Literal as TypingLiteral
 
-from pydantic import BaseModel, ConfigDict, Field, field_serializer, model_validator
+from pydantic import (
+    BaseModel,
+    ConfigDict,
+    Field,
+    RootModel,
+    field_serializer,
+    model_validator,
+)
 import yaml
 
 from crewai.flow.conversational_definition import (
@@ -25,6 +33,7 @@ from crewai.flow.conversational_definition import (
 logger = logging.getLogger(__name__)
 
 FlowDefinitionCondition = str | dict[str, Any]
+_STEP_NAME_PATTERN = re.compile(r"^[A-Za-z_][A-Za-z0-9_]*$")
 
 __all__ = [
     "FlowActionDefinition",
@@ -35,6 +44,8 @@ __all__ = [
     "FlowDefinition",
     "FlowDefinitionCondition",
     "FlowDefinitionDiagnostic",
+    "FlowEachActionDefinition",
+    "FlowEachInnerActionDefinition",
     "FlowExpressionActionDefinition",
     "FlowHumanFeedbackDefinition",
     "FlowMethodDefinition",
@@ -148,10 +159,11 @@ class FlowHumanFeedbackDefinition(BaseModel):
 class FlowCodeActionDefinition(BaseModel):
     """A Flow method action that executes importable Python code."""
 
-    model_config = ConfigDict(extra="forbid")
+    model_config = ConfigDict(populate_by_name=True, extra="forbid")
 
     call: TypingLiteral["code"] = "code"
     ref: str
+    with_: dict[str, Any] | None = Field(default=None, alias="with")
 
 
 class FlowToolActionDefinition(BaseModel):
@@ -173,14 +185,66 @@ class FlowExpressionActionDefinition(BaseModel):
     expr: str
 
 
-FlowActionDefinition = (
+FlowInnerActionDefinition = (
     FlowCodeActionDefinition | FlowToolActionDefinition | FlowExpressionActionDefinition
 )
 
 
+class FlowEachInnerActionDefinition(RootModel[dict[str, FlowInnerActionDefinition]]):
+    """One named action inside an ``each`` composite action."""
+
+    @model_validator(mode="after")
+    def _validate_action_mapping(self) -> FlowEachInnerActionDefinition:
+        if len(self.root) != 1:
+            raise ValueError("each.do entries must be one-key mappings")
+        _validate_step_name(self.name, field="each.do action names")
+        return self
+
+    @property
+    def name(self) -> str:
+        return next(iter(self.root))
+
+    @property
+    def action(self) -> FlowInnerActionDefinition:
+        return next(iter(self.root.values()))
+
+
+class FlowEachActionDefinition(BaseModel):
+    """A composite action that runs a sequential mini-pipeline for each item."""
+
+    model_config = ConfigDict(populate_by_name=True, extra="forbid")
+
+    call: TypingLiteral["each"]
+    in_: str = Field(alias="in")
+    do: list[FlowEachInnerActionDefinition]
+
+    @model_validator(mode="after")
+    def _validate_inner_action_list(self) -> FlowEachActionDefinition:
+        if not self.do:
+            raise ValueError("each.do must contain at least one action")
+
+        seen: set[str] = set()
+        for inner_action in self.do:
+            name = inner_action.name
+            if name in seen:
+                raise ValueError(f"each.do action names must be unique: {name!r}")
+            seen.add(name)
+
+        return self
+
+
+FlowActionDefinition = (
+    FlowCodeActionDefinition
+    | FlowToolActionDefinition
+    | FlowExpressionActionDefinition
+    | FlowEachActionDefinition
+)
+
+
 class FlowMethodDefinition(BaseModel):
     """Static definition of one Flow method and its execution roles."""
 
+    description: str | None = None
     do: FlowActionDefinition
     start: bool | FlowDefinitionCondition | None = None
     listen: FlowDefinitionCondition | None = None
@@ -227,6 +291,12 @@ class FlowDefinition(BaseModel):
     methods: dict[str, FlowMethodDefinition] = Field(default_factory=dict)
     diagnostics: list[FlowDefinitionDiagnostic] = Field(default_factory=list)
 
+    @model_validator(mode="after")
+    def _validate_method_names(self) -> FlowDefinition:
+        for method_name in self.methods:
+            _validate_step_name(method_name, field="Flow method names")
+        return self
+
     def to_dict(self, *, exclude_none: bool = True) -> dict[str, Any]:
         """Serialize the definition to a JSON/YAML-ready dictionary."""
         return self.model_dump(by_alias=True, exclude_none=exclude_none, mode="json")
@@ -369,6 +439,11 @@ def _deserialize_diagnostics(value: Any) -> list[FlowDefinitionDiagnostic]:
     return [FlowDefinitionDiagnostic.model_validate(item) for item in value or []]
 
 
+def _validate_step_name(name: str, *, field: str) -> None:
+    if not isinstance(name, str) or not _STEP_NAME_PATTERN.fullmatch(name):
+        raise ValueError(f"{field} must match {_STEP_NAME_PATTERN.pattern}")
+
+
 def _merge_diagnostics(
     *diagnostic_groups: list[FlowDefinitionDiagnostic],
 ) -> list[FlowDefinitionDiagnostic]:

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

@@ -20,7 +20,7 @@ Example (synchronous, default):
         @human_feedback(
             message="Please review this content:",
             emit=["approved", "rejected"],
-            llm="gpt-4o-mini",
+            llm="gpt-5.4-mini",
         )
         def generate_content(self):
             return {"title": "Article", "body": "Content..."}
@@ -48,7 +48,7 @@ Example (asynchronous with custom provider):
         @human_feedback(
             message="Review this:",
             emit=["approved", "rejected"],
-            llm="gpt-4o-mini",
+            llm="gpt-5.4-mini",
             provider=SlackProvider(),
         )
         def generate_content(self):
@@ -173,7 +173,7 @@ class HumanFeedbackConfig:
 
     message: str
     emit: Sequence[str] | None = None
-    llm: str | BaseLLM | None = "gpt-4o-mini"
+    llm: str | BaseLLM | None = "gpt-5.4-mini"
     default_outcome: str | None = None
     metadata: dict[str, Any] | None = None
     provider: HumanFeedbackProvider | None = None
@@ -212,7 +212,7 @@ def _validate_human_feedback_options(
         if not llm:
             raise ValueError(
                 "llm is required when emit is specified. "
-                "Provide an LLM model string (e.g., 'gpt-4o-mini') or a BaseLLM instance. "
+                "Provide an LLM model string (e.g., 'gpt-5.4-mini') or a BaseLLM instance. "
                 "See the CrewAI Human-in-the-Loop (HITL) documentation for more information: "
                 "https://docs.crewai.com/en/learn/human-feedback-in-flows"
             )
@@ -235,12 +235,12 @@ def _resolve_llm_instance(llm: Any) -> Any:
     from crewai.llm import LLM
 
     if llm is None:
-        return LLM(model="gpt-4o-mini")
+        return LLM(model="gpt-5.4-mini")
     if isinstance(llm, str):
         return LLM(model=llm)
     if isinstance(llm, dict):
         deserialized = _deserialize_llm_from_context(llm)
-        return deserialized if deserialized is not None else LLM(model="gpt-4o-mini")
+        return deserialized if deserialized is not None else LLM(model="gpt-5.4-mini")
     return llm  # already a BaseLLM instance
 
 
@@ -362,7 +362,7 @@ def _distill_and_store_lessons(
 def human_feedback(
     message: str,
     emit: Sequence[str] | None = None,
-    llm: str | BaseLLM | None = "gpt-4o-mini",
+    llm: str | BaseLLM | None = "gpt-5.4-mini",
     default_outcome: str | None = None,
     metadata: dict[str, Any] | None = None,
     provider: HumanFeedbackProvider | None = None,

lib/crewai/src/crewai/flow/runtime/__init__.py

@@ -121,11 +121,8 @@ from crewai.flow.human_feedback import (
 )
 from crewai.flow.input_provider import InputProvider
 from crewai.flow.persistence.base import FlowPersistence
-from crewai.flow.runtime._resolvers import (
-    resolve_action,
-    resolve_instance_ref,
-    resolve_ref,
-)
+from crewai.flow.runtime._actions import build_action
+from crewai.flow.runtime._refs import resolve_instance_ref, resolve_ref
 from crewai.flow.types import (
     FlowExecutionData,
     FlowMethodName,
@@ -1092,9 +1089,9 @@ class Flow(BaseModel, Generic[T], metaclass=FlowMeta):
         self._methods.update(methods)
 
     def _action_bound_methods(self) -> dict[FlowMethodName, Callable[..., Any]]:
-        def resolve(name: str, definition: FlowMethodDefinition) -> Callable[..., Any]:
+        def build(name: str, definition: FlowMethodDefinition) -> Callable[..., Any]:
             try:
-                return resolve_action(self, definition.do)
+                return build_action(self, definition.do)
             except Exception as e:
                 unresolved.append(f"{name}: {e}")
                 return lambda *args, **kwargs: None
@@ -1102,9 +1099,7 @@ class Flow(BaseModel, Generic[T], metaclass=FlowMeta):
         methods: dict[FlowMethodName, Callable[..., Any]] = {}
         unresolved: list[str] = []
         for method_name, method_definition in self._definition.methods.items():
-            methods[FlowMethodName(method_name)] = resolve(
-                method_name, method_definition
-            )
+            methods[FlowMethodName(method_name)] = build(method_name, method_definition)
         if unresolved:
             raise ValueError(
                 f"Cannot build flow {self._definition.name!r} from its definition; "

lib/crewai/src/crewai/flow/runtime/_actions.py

@@ -0,0 +1,200 @@
+"""Build FlowDefinition actions into live runtime callables."""
+
+from __future__ import annotations
+
+from collections.abc import Callable
+import inspect
+from typing import TYPE_CHECKING, Any, cast
+
+from crewai.flow.flow_definition import (
+    FlowActionDefinition,
+    FlowCodeActionDefinition,
+    FlowEachActionDefinition,
+    FlowEachInnerActionDefinition,
+    FlowExpressionActionDefinition,
+    FlowToolActionDefinition,
+)
+from crewai.flow.runtime._expressions import evaluate_expression, render_with_block
+from crewai.flow.runtime._refs import InvalidRefError, resolve_ref
+
+
+if TYPE_CHECKING:
+    from crewai.flow.runtime import Flow
+
+
+__all__ = ["build_action"]
+
+LocalContext = dict[str, Any]
+_LOCAL_CONTEXT_KWARG = "__flow_definition_local_context"
+
+
+class CodeAction:
+    definition_type = FlowCodeActionDefinition
+
+    def __init__(self, flow: Flow[Any], definition: FlowCodeActionDefinition) -> None:
+        self.flow = flow
+        self.definition = definition
+        self.handler = self._resolve_handler()
+        self.signature = inspect.signature(self.handler)
+
+    def run(self, *args: Any, **kwargs: Any) -> Any:
+        local_context = _pop_local_context(kwargs)
+        if self.definition.with_ is None:
+            return self.handler(*args, **kwargs)
+        return self.handler(
+            **render_with_block(
+                self.flow, self.definition.with_, local_context=local_context
+            )
+        )
+
+    def _resolve_handler(self) -> Callable[..., Any]:
+        ref = self.definition.ref
+        target = resolve_ref(ref, field="do")
+        if not callable(target):
+            raise InvalidRefError(f"invalid do ref {ref!r}; object is not callable")
+        handler = cast(Callable[..., Any], target)
+        if getattr(handler, "__self__", None) is None:
+            handler = handler.__get__(self.flow, type(self.flow))
+        return handler
+
+
+class ToolAction:
+    definition_type = FlowToolActionDefinition
+
+    def __init__(self, flow: Flow[Any], definition: FlowToolActionDefinition) -> None:
+        self.flow = flow
+        self.definition = definition
+        self.tool = self._build_tool()
+        self.kwargs = definition.with_ or {}
+
+    def run(self, *_args: Any, **kwargs: Any) -> Any:
+        local_context = _pop_local_context(kwargs)
+        return self.tool.run(
+            **render_with_block(self.flow, self.kwargs, local_context=local_context)
+        )
+
+    def _build_tool(self) -> Any:
+        target = resolve_ref(self.definition.ref, field="do")
+        from crewai.tools import BaseTool
+
+        if not (inspect.isclass(target) and issubclass(target, BaseTool)):
+            raise InvalidRefError(
+                f"invalid tool ref {self.definition.ref!r}; expected a BaseTool class"
+            )
+
+        try:
+            tool_cls = cast(Callable[[], BaseTool], target)
+            return tool_cls()
+        except Exception as e:
+            raise InvalidRefError(
+                f"cannot instantiate tool ref {self.definition.ref!r} "
+                f"without arguments: {e}"
+            ) from e
+
+
+class ExpressionAction:
+    definition_type = FlowExpressionActionDefinition
+
+    def __init__(
+        self, flow: Flow[Any], definition: FlowExpressionActionDefinition
+    ) -> None:
+        self.flow = flow
+        self.definition = definition
+
+    def run(self, *_args: Any, **kwargs: Any) -> Any:
+        local_context = _pop_local_context(kwargs)
+        return evaluate_expression(
+            self.flow, self.definition.expr, local_context=local_context
+        )
+
+
+class EachAction:
+    definition_type = FlowEachActionDefinition
+
+    def __init__(self, flow: Flow[Any], definition: FlowEachActionDefinition) -> None:
+        self.flow = flow
+        self.definition = definition
+        self.inner_actions = [
+            (inner_action.name, self._build_inner_action(inner_action))
+            for inner_action in definition.do
+        ]
+
+    async def run(self, *_args: Any, **_kwargs: Any) -> list[Any]:
+        items = evaluate_expression(self.flow, self.definition.in_)
+        if not isinstance(items, list):
+            raise ValueError("each.in must evaluate to an array")
+
+        results: list[Any] = []
+
+        for item in items:
+            local_outputs: dict[str, Any] = {}
+            last_output: Any = None
+            for name, run_inner_action in self.inner_actions:
+                last_output = await run_inner_action(
+                    {"item": item, "outputs": local_outputs}
+                )
+                local_outputs[name] = last_output
+            results.append(last_output)
+
+        return results
+
+    def _build_inner_action(
+        self, inner_action: FlowEachInnerActionDefinition
+    ) -> Callable[[LocalContext], Any]:
+        run_action = build_action(self.flow, inner_action.action)
+
+        async def run_inner_action(local_context: LocalContext) -> Any:
+            result = run_action(**{_LOCAL_CONTEXT_KWARG: local_context})
+            if inspect.isawaitable(result):
+                result = await result
+            return result
+
+        return run_inner_action
+
+
+_ACTION_TYPES: tuple[type[Any], ...] = (
+    EachAction,
+    CodeAction,
+    ToolAction,
+    ExpressionAction,
+)
+
+
+def build_action(
+    flow: Flow[Any], definition: FlowActionDefinition
+) -> Callable[..., Any]:
+    """Turn one `do:` action into the callable the flow runs for that node."""
+    for action_type in _ACTION_TYPES:
+        if isinstance(definition, action_type.definition_type):
+            return _as_flow_method(action_type(flow, definition))
+    raise ValueError(f"unknown call type {getattr(definition, 'call', None)!r}")
+
+
+def _as_flow_method(action: Any) -> Callable[..., Any]:
+    run: Callable[..., Any]
+    if inspect.iscoroutinefunction(action.run):
+
+        async def run_async(*args: Any, **kwargs: Any) -> Any:
+            return await action.run(*args, **kwargs)
+
+        run = run_async
+    else:
+
+        def run_sync(*args: Any, **kwargs: Any) -> Any:
+            return action.run(*args, **kwargs)
+
+        run = run_sync
+
+    signature = getattr(action, "signature", None)
+    if signature is not None:
+        object.__setattr__(run, "__signature__", signature)
+    return run
+
+
+def _pop_local_context(kwargs: dict[str, Any]) -> LocalContext | None:
+    local_context = kwargs.pop(_LOCAL_CONTEXT_KWARG, None)
+    if local_context is None:
+        return None
+    if not isinstance(local_context, dict):
+        raise TypeError("flow definition local context must be a mapping")
+    return cast(LocalContext, local_context)

lib/crewai/src/crewai/flow/runtime/_expressions.py

@@ -2,7 +2,6 @@
 
 from __future__ import annotations
 
-import copy
 import dataclasses
 from itertools import pairwise
 import json
@@ -25,25 +24,44 @@ class FlowExpressionError(ValueError):
     """A FlowDefinition expression failed to parse or evaluate."""
 
 
-def render_with_block(flow: Flow[Any], value: Any) -> Any:
+def render_with_block(
+    flow: Flow[Any], value: Any, local_context: dict[str, Any] | None = None
+) -> Any:
     """Render CEL expressions inside a FlowDefinition ``with:`` payload."""
-    context = _expression_context(flow)
+    context = _expression_context(flow, local_context=local_context)
     return _render_value(value, context)
 
 
-def evaluate_expression(flow: Flow[Any], expression: str) -> Any:
+def evaluate_expression(
+    flow: Flow[Any], expression: str, local_context: dict[str, Any] | None = None
+) -> Any:
     """Evaluate a FlowDefinition CEL expression against runtime context."""
     expression = expression.strip()
     if not expression:
         raise FlowExpressionError("empty CEL expression")
-    return _eval_cel(expression, _expression_context(flow))
+    return _eval_cel(expression, _expression_context(flow, local_context=local_context))
 
 
-def _expression_context(flow: Flow[Any]) -> dict[str, Any]:
-    return {
+def _expression_context(
+    flow: Flow[Any], local_context: dict[str, Any] | None = None
+) -> dict[str, Any]:
+    outputs = _outputs_by_name(flow._method_outputs)
+    context = {
         "state": flow._copy_and_serialize_state(),
-        "outputs": _outputs_by_name(flow._method_outputs),
+        "outputs": outputs,
     }
+    if local_context:
+        local_values = {
+            key: _to_json_safe(value) for key, value in local_context.items()
+        }
+        local_outputs = local_values.pop("outputs", None)
+        local_values.pop("state", None)
+        context.update(local_values)
+        if local_outputs is not None:
+            if not isinstance(local_outputs, dict):
+                raise TypeError("flow definition local outputs must be a mapping")
+            context["outputs"] = {**outputs, **local_outputs}
+    return context
 
 
 def _outputs_by_name(method_outputs: list[Any]) -> dict[str, Any]:
@@ -54,15 +72,24 @@ def _outputs_by_name(method_outputs: list[Any]) -> dict[str, Any]:
         if isinstance(entry, dict) and "output" in entry:
             method = str(entry.get("method", ""))
             output = entry["output"]
-        output = copy.deepcopy(output)
-        if isinstance(output, BaseModel):
-            output = output.model_dump(mode="json")
-        elif dataclasses.is_dataclass(output) and not isinstance(output, type):
-            output = dataclasses.asdict(output)
-        outputs[method] = output
+        outputs[method] = _to_json_safe(output)
     return outputs
 
 
+def _to_json_safe(value: Any) -> Any:
+    if isinstance(value, BaseModel):
+        return value.model_dump(mode="json")
+    if dataclasses.is_dataclass(value) and not isinstance(value, type):
+        return dataclasses.asdict(value)
+    if isinstance(value, dict):
+        return {key: _to_json_safe(item) for key, item in value.items()}
+    if isinstance(value, list):
+        return [_to_json_safe(item) for item in value]
+    if isinstance(value, tuple):
+        return [_to_json_safe(item) for item in value]
+    return value
+
+
 def _render_value(value: Any, context: dict[str, Any]) -> Any:
     if isinstance(value, str):
         return _render_string(value, context)

lib/crewai/src/crewai/flow/runtime/_refs.py

@@ -0,0 +1,38 @@
+"""Resolution of ``module:qualname`` refs into live Python objects."""
+
+from __future__ import annotations
+
+import importlib
+import inspect
+from operator import attrgetter
+from typing import Any
+
+
+class InvalidRefError(ValueError):
+    """A definition ref that cannot be resolved to a live object."""
+
+
+def resolve_ref(ref: str, *, field: str) -> Any:
+    """Import the object a definition's `module:qualname` ref points to."""
+    module_name, _, qualname = ref.partition(":")
+    if "<" in ref or not module_name or not qualname:
+        raise InvalidRefError(
+            f"invalid {field} ref {ref!r}; expected 'module:qualname'"
+        )
+    try:
+        return attrgetter(qualname)(importlib.import_module(module_name))
+    except (ImportError, AttributeError) as e:
+        raise InvalidRefError(f"unresolvable {field} ref {ref!r}") from e
+
+
+def resolve_instance_ref(ref: str, *, field: str) -> Any:
+    """Resolve a ref, auto-instantiating a no-arg class into an instance."""
+    target = resolve_ref(ref, field=field)
+    if not inspect.isclass(target):
+        return target
+    try:
+        return target()
+    except Exception as e:
+        raise InvalidRefError(
+            f"cannot instantiate {field} ref {ref!r} without arguments: {e}"
+        ) from e

lib/crewai/src/crewai/flow/runtime/_resolvers.py

@@ -1,116 +0,0 @@
-"""Resolution of FlowDefinition refs (``module:qualname``) into live objects.
-
-Every ref-shaped value in a definition — ``do`` actions, ``state.ref``,
-``config.input_provider``, ``human_feedback.provider`` — resolves through
-:func:`resolve_ref`. Failures are loud and name the field and the ref.
-"""
-
-from __future__ import annotations
-
-from collections.abc import Callable
-import importlib
-import inspect
-from operator import attrgetter
-from typing import TYPE_CHECKING, Any, cast
-
-from crewai.flow.flow_definition import (
-    FlowActionDefinition,
-    FlowCodeActionDefinition,
-    FlowExpressionActionDefinition,
-    FlowToolActionDefinition,
-)
-from crewai.flow.runtime._expressions import evaluate_expression, render_with_block
-
-
-if TYPE_CHECKING:
-    from crewai.flow.runtime import Flow
-
-
-class InvalidRefError(ValueError):
-    """A definition ref that cannot be resolved to a live object."""
-
-
-def resolve_ref(ref: str, *, field: str) -> Any:
-    """Import the object a definition's `module:qualname` ref points to."""
-    module_name, _, qualname = ref.partition(":")
-    if "<" in ref or not module_name or not qualname:
-        raise InvalidRefError(
-            f"invalid {field} ref {ref!r}; expected 'module:qualname'"
-        )
-    try:
-        return attrgetter(qualname)(importlib.import_module(module_name))
-    except (ImportError, AttributeError) as e:
-        raise InvalidRefError(f"unresolvable {field} ref {ref!r}") from e
-
-
-def resolve_instance_ref(ref: str, *, field: str) -> Any:
-    """Resolve a ref, auto-instantiating a no-arg class into an instance."""
-    target = resolve_ref(ref, field=field)
-    if not inspect.isclass(target):
-        return target
-    try:
-        return target()
-    except Exception as e:
-        raise InvalidRefError(
-            f"cannot instantiate {field} ref {ref!r} without arguments: {e}"
-        ) from e
-
-
-def _resolve_code_action(
-    flow: Flow[Any], action: FlowCodeActionDefinition
-) -> Callable[..., Any]:
-    ref = action.ref
-    target = resolve_ref(ref, field="do")
-    if not callable(target):
-        raise InvalidRefError(f"invalid do ref {ref!r}; object is not callable")
-    handler = cast(Callable[..., Any], target)
-    if getattr(handler, "__self__", None) is None:
-        handler = handler.__get__(flow, type(flow))
-    return handler
-
-
-def _resolve_tool_action(
-    flow: Flow[Any], action: FlowToolActionDefinition
-) -> Callable[..., Any]:
-    target = resolve_ref(action.ref, field="do")
-    from crewai.tools import BaseTool
-
-    if not (inspect.isclass(target) and issubclass(target, BaseTool)):
-        raise InvalidRefError(
-            f"invalid tool ref {action.ref!r}; expected a BaseTool class"
-        )
-
-    try:
-        tool_cls = cast(Callable[[], BaseTool], target)
-        tool = tool_cls()
-    except Exception as e:
-        raise InvalidRefError(
-            f"cannot instantiate tool ref {action.ref!r} without arguments: {e}"
-        ) from e
-
-    tool_kwargs = action.with_ or {}
-
-    def run_tool(*_args: Any, **_kwargs: Any) -> Any:
-        return tool.run(**render_with_block(flow, tool_kwargs))
-
-    return run_tool
-
-
-def _resolve_expression_action(
-    flow: Flow[Any], action: FlowExpressionActionDefinition
-) -> Callable[..., Any]:
-    def run_expression(*_args: Any, **_kwargs: Any) -> Any:
-        return evaluate_expression(flow, action.expr)
-
-    return run_expression
-
-
-def resolve_action(flow: Flow[Any], action: FlowActionDefinition) -> Callable[..., Any]:
-    """Turn one `do:` action into the callable the flow runs for that node."""
-    if action.call == "code":
-        return _resolve_code_action(flow, action)
-    if action.call == "tool":
-        return _resolve_tool_action(flow, action)
-    if action.call == "expression":
-        return _resolve_expression_action(flow, action)
-    raise ValueError(f"unknown call type {action.call!r}")

lib/crewai/src/crewai/lite_agent.py

@@ -390,7 +390,10 @@ class LiteAgent(FlowTrackable, BaseModel):
         if self.memory is True:
             from crewai.memory.unified_memory import Memory
 
-            object.__setattr__(self, "_memory", Memory())
+            memory_kwargs: dict[str, Any] = {}
+            if self.llm is not None:
+                memory_kwargs["llm"] = self.llm
+            object.__setattr__(self, "_memory", Memory(**memory_kwargs))
         elif self.memory is not None and self.memory is not False:
             object.__setattr__(self, "_memory", self.memory)
         else:

lib/crewai/src/crewai/llm.py

@@ -68,7 +68,17 @@ if TYPE_CHECKING:
     from crewai.tools.base_tool import BaseTool
     from crewai.utilities.types import LLMMessage
 
-try:
+load_dotenv()
+logger = logging.getLogger(__name__)
+
+# litellm is lazy-loaded to avoid its module-level dotenv.load_dotenv()
+# from polluting env vars (e.g. MODEL= overriding embedder model_name).
+# The TYPE_CHECKING imports give mypy the real types; at runtime the names
+# stay None until _ensure_litellm() rebinds them.
+_litellm_loaded = False
+LITELLM_AVAILABLE = False
+
+if TYPE_CHECKING:
     import litellm
     from litellm.litellm_core_utils.get_supported_openai_params import (
         get_supported_openai_params,
@@ -85,28 +95,70 @@ try:
         StreamingChoices as LiteLLMStreamingChoices,
     )
     from litellm.utils import supports_response_schema
-
-    LITELLM_AVAILABLE = True
-except ImportError:
-    LITELLM_AVAILABLE = False
-    litellm = None  # type: ignore[assignment]
-    Choices = None  # type: ignore[assignment, misc]
-    LiteLLMDelta = None  # type: ignore[assignment, misc]
-    Message = None  # type: ignore[assignment, misc]
-    ModelResponseBase = None  # type: ignore[assignment, misc]
-    ModelResponseStream = None  # type: ignore[assignment, misc]
-    LiteLLMStreamingChoices = None  # type: ignore[assignment, misc]
-    get_supported_openai_params = None  # type: ignore[assignment]
-    ChatCompletionDeltaToolCall = None  # type: ignore[assignment, misc]
-    Function = None  # type: ignore[assignment, misc]
-    ModelResponse = None  # type: ignore[assignment, misc]
-    supports_response_schema = None  # type: ignore[assignment]
+else:
+    litellm = None
+    Choices = None
+    LiteLLMDelta = None
+    Message = None
+    ModelResponseBase = None
+    ModelResponseStream = None
+    LiteLLMStreamingChoices = None
+    get_supported_openai_params = None
+    ChatCompletionDeltaToolCall = None
+    Function = None
+    ModelResponse = None
+    supports_response_schema = None
 
 
-load_dotenv()
-logger = logging.getLogger(__name__)
-if LITELLM_AVAILABLE:
-    litellm.suppress_debug_info = True
+def _ensure_litellm() -> bool:
+    """Lazy-load litellm on first use. Returns True if available."""
+    global _litellm_loaded, LITELLM_AVAILABLE
+    global litellm, Choices, LiteLLMDelta, Message, ModelResponseBase
+    global ModelResponseStream, LiteLLMStreamingChoices, get_supported_openai_params
+    global ChatCompletionDeltaToolCall, Function
+    global ModelResponse, supports_response_schema
+
+    if _litellm_loaded:
+        return LITELLM_AVAILABLE
+    _litellm_loaded = True
+
+    try:
+        import litellm as _litellm
+        from litellm.litellm_core_utils.get_supported_openai_params import (
+            get_supported_openai_params as _get_supported_openai_params,
+        )
+        from litellm.types.utils import (
+            ChatCompletionDeltaToolCall as _ChatCompletionDeltaToolCall,
+            Choices as _Choices,
+            Delta as _LiteLLMDelta,
+            Function as _Function,
+            Message as _Message,
+            ModelResponse as _ModelResponse,
+            ModelResponseBase as _ModelResponseBase,
+            ModelResponseStream as _ModelResponseStream,
+            StreamingChoices as _LiteLLMStreamingChoices,
+        )
+        from litellm.utils import supports_response_schema as _supports_response_schema
+
+        litellm = _litellm
+        Choices = _Choices  # type: ignore[misc]
+        LiteLLMDelta = _LiteLLMDelta  # type: ignore[misc]
+        Message = _Message  # type: ignore[misc]
+        ModelResponseBase = _ModelResponseBase  # type: ignore[misc]
+        ModelResponseStream = _ModelResponseStream  # type: ignore[misc]
+        LiteLLMStreamingChoices = _LiteLLMStreamingChoices  # type: ignore[misc]
+        get_supported_openai_params = _get_supported_openai_params
+        ChatCompletionDeltaToolCall = _ChatCompletionDeltaToolCall  # type: ignore[misc]
+        Function = _Function  # type: ignore[misc]
+        ModelResponse = _ModelResponse  # type: ignore[misc]
+        supports_response_schema = _supports_response_schema
+
+        _litellm.suppress_debug_info = True
+        LITELLM_AVAILABLE = True
+    except ImportError:
+        LITELLM_AVAILABLE = False
+
+    return LITELLM_AVAILABLE
 
 
 MIN_CONTEXT: Final[int] = 1024
@@ -117,6 +169,7 @@ LLM_CONTEXT_WINDOW_SIZES: Final[dict[str, int]] = {
     "gpt-4": 8192,
     "gpt-4o": 128000,
     "gpt-4o-mini": 200000,
+    "gpt-5.4-mini": 200000,
     "gpt-4-turbo": 128000,
     "gpt-4.1": 1047576,  # Based on official docs
     "gpt-4.1-mini-2025-04-14": 1047576,
@@ -411,7 +464,8 @@ class LLM(BaseLLM):
             except Exception as e:
                 raise ImportError(f"Error importing native provider: {e}") from e
 
-        if not LITELLM_AVAILABLE:
+        # FALLBACK to LiteLLM — lazy-load on first use
+        if not _ensure_litellm():
             native_list = ", ".join(SUPPORTED_NATIVE_PROVIDERS)
             error_msg = (
                 f"Unable to initialize LLM with model '{model}'. "
@@ -632,7 +686,7 @@ class LLM(BaseLLM):
     @model_validator(mode="after")
     def _init_litellm(self) -> LLM:
         self.is_litellm = True
-        if LITELLM_AVAILABLE:
+        if _ensure_litellm():
             litellm.drop_params = True
             self.set_callbacks(self.callbacks or [])
             self.set_env_callbacks()
@@ -2290,7 +2344,8 @@ class LLM(BaseLLM):
         Note: This validation only applies to the litellm fallback path.
         Native providers have their own validation.
         """
-        if not LITELLM_AVAILABLE or supports_response_schema is None:
+        if not _ensure_litellm() or supports_response_schema is None:
+            # When litellm is not available, skip validation
             # (this path should only be reached for litellm fallback models)
             return
 
@@ -2310,7 +2365,7 @@ class LLM(BaseLLM):
         Note: This method is only used by the litellm fallback path.
         Native providers override this method with their own implementation.
         """
-        if not LITELLM_AVAILABLE:
+        if not _ensure_litellm():
             # When litellm is not available, assume function calling is supported
             # (all modern models support it)
             return True
@@ -2334,7 +2389,7 @@ class LLM(BaseLLM):
         if "gpt-5" in model_lower:
             return False
 
-        if not LITELLM_AVAILABLE or get_supported_openai_params is None:
+        if not _ensure_litellm() or get_supported_openai_params is None:
             # When litellm is not available, assume stop words are supported
             return True
 
@@ -2382,7 +2437,8 @@ class LLM(BaseLLM):
         Note: This only affects the litellm fallback path. Native providers
         don't use litellm callbacks - they emit events via base_llm.py.
         """
-        if not LITELLM_AVAILABLE:
+        if not _ensure_litellm():
+            # When litellm is not available, callbacks are still stored
             # but not registered with litellm globals
             return
 
@@ -2420,7 +2476,8 @@ class LLM(BaseLLM):
         This will set `litellm.success_callback` to ["langfuse", "langsmith"] and
         `litellm.failure_callback` to ["langfuse"].
         """
-        if not LITELLM_AVAILABLE:
+        if not _ensure_litellm():
+            # When litellm is not available, env callbacks have no effect
             return
 
         with suppress_warnings():

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

@@ -1300,6 +1300,7 @@ class AzureCompletion(BaseLLM):
             "gpt-4": 8192,
             "gpt-4o": 128000,
             "gpt-4o-mini": 200000,
+            "gpt-5.4-mini": 200000,
             "gpt-4-turbo": 128000,
             "gpt-35-turbo": 16385,
             "gpt-3.5-turbo": 16385,

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

@@ -2406,6 +2406,7 @@ class OpenAICompletion(BaseLLM):
             "gpt-4": 8192,
             "gpt-4o": 128000,
             "gpt-4o-mini": 200000,
+            "gpt-5.4-mini": 200000,
             "gpt-4-turbo": 128000,
             "gpt-4.1": 1047576,
             "gpt-4.1-mini-2025-04-14": 1047576,

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

@@ -8,6 +8,39 @@ from typing import Any, Protocol, runtime_checkable
 from crewai.memory.types import MemoryRecord, ScopeInfo
 
 
+class EmbeddingDimensionMismatchError(ValueError):
+    """Raised when an embedding's dimensionality doesn't match the existing store.
+
+    The most common cause is upgrading CrewAI across the default-embedder
+    change (text-embedding-3-small, 1536 dims → text-embedding-3-large,
+    3072 dims) while keeping a local memory store created before the upgrade.
+
+    Deliberately not a ``RuntimeError``: background-save plumbing treats
+    ``RuntimeError`` as interpreter/executor shutdown and silently drops the
+    save, which would swallow this actionable migration error.
+    """
+
+    def __init__(self, stored_dim: int, new_dim: int) -> None:
+        self.stored_dim = stored_dim
+        self.new_dim = new_dim
+        super().__init__(
+            f"Embedding dimension mismatch: this memory store contains "
+            f"{stored_dim}-dimensional vectors, but the current embedder produced "
+            f"a {new_dim}-dimensional vector.\n\n"
+            "This usually means the store was created with a different embedding "
+            "model. CrewAI's default embedder changed from "
+            "text-embedding-3-small (1536 dims) to text-embedding-3-large "
+            "(3072 dims), so memory stores created before the upgrade are "
+            "incompatible with the new default.\n\n"
+            "To fix, do one of the following:\n"
+            "  - Reset local memory so it is rebuilt with the new embedder:\n"
+            "      crewai reset-memories --memory   (or crew.reset_memories())\n"
+            "  - Keep existing memories by pinning the previous embedder:\n"
+            '      embedder={"provider": "openai", '
+            '"config": {"model": "text-embedding-3-small"}}'
+        )
+
+
 @runtime_checkable
 class StorageBackend(Protocol):
     """Protocol for pluggable memory storage backends."""

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

@@ -15,15 +15,16 @@ from typing import Any
 from crewai_core.lock_store import lock as store_lock
 import lancedb  # type: ignore[import-untyped]
 
+from crewai.memory.storage.backend import EmbeddingDimensionMismatchError
 from crewai.memory.types import MemoryRecord, ScopeInfo
 
 
 _logger = logging.getLogger(__name__)
 
-# Default embedding vector dimensionality (matches OpenAI text-embedding-3-small).
+# Default embedding vector dimensionality (matches OpenAI text-embedding-3-large).
 # Used when creating new tables and for zero-vector placeholder scans.
 # Callers can override via the ``vector_dim`` constructor parameter.
-DEFAULT_VECTOR_DIM = 1536
+DEFAULT_VECTOR_DIM = 3072
 
 # Safety cap on the number of rows returned by a single scan query.
 # Prevents unbounded memory use when scanning large tables for scope info,
@@ -288,13 +289,19 @@ class LanceDBStorage:
     def save(self, records: list[MemoryRecord]) -> None:
         if not records:
             return
-        # Auto-detect dimension from the first real embedding.
+        # Auto-detect dimension from the first real embedding and validate
+        # the whole batch against it — a silent mismatch would otherwise be
+        # zero-filled below and corrupt search results.
         dim = None
         for r in records:
             if r.embedding and len(r.embedding) > 0:
-                dim = len(r.embedding)
-                break
+                if dim is None:
+                    dim = len(r.embedding)
+                elif len(r.embedding) != dim:
+                    raise EmbeddingDimensionMismatchError(dim, len(r.embedding))
         is_new_table = self._table is None
+        if not is_new_table and dim and self._vector_dim and dim != self._vector_dim:
+            raise EmbeddingDimensionMismatchError(self._vector_dim, dim)
         with store_lock(self._lock_name):
             self._ensure_table(vector_dim=dim)
             rows = [self._record_to_row(rec) for rec in records]
@@ -311,6 +318,15 @@ class LanceDBStorage:
 
     def update(self, record: MemoryRecord) -> None:
         """Update a record by ID. Preserves created_at, updates last_accessed."""
+        if (
+            self._table is not None
+            and record.embedding
+            and self._vector_dim
+            and len(record.embedding) != self._vector_dim
+        ):
+            raise EmbeddingDimensionMismatchError(
+                self._vector_dim, len(record.embedding)
+            )
         with store_lock(self._lock_name):
             self._ensure_table()
             safe_id = str(record.id).replace("'", "''")
@@ -363,6 +379,10 @@ class LanceDBStorage:
     ) -> list[tuple[MemoryRecord, float]]:
         if self._table is None:
             return []
+        if self._vector_dim and len(query_embedding) != self._vector_dim:
+            raise EmbeddingDimensionMismatchError(
+                self._vector_dim, len(query_embedding)
+            )
         query = self._table.search(query_embedding)
         if scope_prefix is not None and scope_prefix.strip("/"):
             prefix = scope_prefix.rstrip("/")

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

@@ -36,6 +36,7 @@ from qdrant_edge import (
     UpdateOperation,
 )
 
+from crewai.memory.storage.backend import EmbeddingDimensionMismatchError
 from crewai.memory.types import MemoryRecord, ScopeInfo
 
 
@@ -43,7 +44,7 @@ _logger = logging.getLogger(__name__)
 
 VECTOR_NAME: Final[str] = "memory"
 
-DEFAULT_VECTOR_DIM: Final[int] = 1536
+DEFAULT_VECTOR_DIM: Final[int] = 3072
 
 _SCROLL_BATCH: Final[int] = 256
 
@@ -183,6 +184,10 @@ class QdrantEdgeStorage:
         except Exception:
             _logger.debug("Index creation failed (may already exist)", exc_info=True)
 
+    def _has_existing_data(self) -> bool:
+        """True when either shard already holds persisted records."""
+        return self._local_has_data or self._central_path.exists()
+
     def _record_to_point(self, record: MemoryRecord) -> Point:
         """Convert a MemoryRecord to a Qdrant Point."""
         return Point(
@@ -277,11 +282,19 @@ class QdrantEdgeStorage:
         if not records:
             return
 
+        # Validate the batch is internally consistent before touching the
+        # store-level dimension.
+        batch_dim = 0
+        for r in records:
+            if r.embedding and len(r.embedding) > 0:
+                if batch_dim == 0:
+                    batch_dim = len(r.embedding)
+                elif len(r.embedding) != batch_dim:
+                    raise EmbeddingDimensionMismatchError(batch_dim, len(r.embedding))
         if self._vector_dim == 0:
-            for r in records:
-                if r.embedding and len(r.embedding) > 0:
-                    self._vector_dim = len(r.embedding)
-                    break
+            self._vector_dim = batch_dim
+        elif batch_dim and batch_dim != self._vector_dim and self._has_existing_data():
+            raise EmbeddingDimensionMismatchError(self._vector_dim, batch_dim)
         if self._config is None and self._vector_dim > 0:
             self._config = self._build_config(self._vector_dim)
         if self._config is None:
@@ -308,6 +321,14 @@ class QdrantEdgeStorage:
         min_score: float = 0.0,
     ) -> list[tuple[MemoryRecord, float]]:
         """Search both central and local shards, merge results."""
+        if (
+            self._vector_dim
+            and len(query_embedding) != self._vector_dim
+            and self._has_existing_data()
+        ):
+            raise EmbeddingDimensionMismatchError(
+                self._vector_dim, len(query_embedding)
+            )
         filt = self._build_scope_filter(scope_prefix)
         fetch_limit = limit * 3 if (categories or metadata_filter) else limit
         all_scored: list[tuple[dict[str, Any], float, bool]] = []
@@ -466,6 +487,16 @@ class QdrantEdgeStorage:
 
     def update(self, record: MemoryRecord) -> None:
         """Update a record by upserting with the same point ID."""
+        if (
+            self._config is not None
+            and record.embedding
+            and self._vector_dim
+            and len(record.embedding) != self._vector_dim
+            and self._has_existing_data()
+        ):
+            raise EmbeddingDimensionMismatchError(
+                self._vector_dim, len(record.embedding)
+            )
         if self._config is None:
             if record.embedding and len(record.embedding) > 0:
                 self._vector_dim = len(record.embedding)

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

@@ -66,7 +66,7 @@ class Memory(BaseModel):
     memory_kind: Literal["memory"] = "memory"
 
     llm: Annotated[BaseLLM | str, PlainValidator(_passthrough)] = Field(
-        default="gpt-4o-mini",
+        default="gpt-5.4-mini",
         description="LLM for analysis (model name or BaseLLM instance).",
     )
     storage: Annotated[StorageBackend | str, PlainValidator(_passthrough)] = Field(
@@ -239,7 +239,7 @@ class Memory(BaseModel):
                 raise RuntimeError(
                     f"Memory requires an LLM for analysis but initialization failed: {e}\n\n"
                     "To fix this, do one of the following:\n"
-                    "  - Set OPENAI_API_KEY for the default model (gpt-4o-mini)\n"
+                    "  - Set OPENAI_API_KEY for the default model (gpt-5.4-mini)\n"
                     '  - Pass a different model: Memory(llm="anthropic/claude-3-haiku-20240307")\n'
                     '  - Pass any LLM instance: Memory(llm=LLM(model="your-model"))\n'
                     "  - To skip LLM analysis, pass all fields explicitly to remember()\n"
@@ -261,7 +261,7 @@ class Memory(BaseModel):
                 raise RuntimeError(
                     f"Memory requires an embedder for vector search but initialization failed: {e}\n\n"
                     "To fix this, do one of the following:\n"
-                    "  - Set OPENAI_API_KEY for the default embedder (text-embedding-3-small)\n"
+                    "  - Set OPENAI_API_KEY for the default embedder (text-embedding-3-large)\n"
                     '  - Pass a different embedder: Memory(embedder={{"provider": "google", "config": {{...}}}})\n'
                     "  - Pass a callable: Memory(embedder=my_embedding_function)\n\n"
                     f"Docs: {self._MEMORY_DOCS_URL}"
@@ -322,12 +322,16 @@ class Memory(BaseModel):
         """Block until all pending background saves have completed.
 
         Called automatically by ``recall()`` and should be called by the
-        crew at shutdown to ensure no saves are lost.
+        crew at shutdown to ensure no saves are lost. Background save failures
+        are already reported through ``MemorySaveFailedEvent`` and should not
+        fail the task, crew, or flow that produced the output.
         """
         with self._pending_lock:
             pending = list(self._pending_saves)
         for future in pending:
-            future.result()  # blocks until done; re-raises exceptions
+            if future.cancelled():
+                continue
+            future.exception()  # blocks until done without re-raising failures
 
     def close(self) -> None:
         """Drain pending saves, flush storage, and shut down the background thread pool."""
@@ -605,12 +609,16 @@ class Memory(BaseModel):
                 root_scope,
             )
             elapsed_ms = (time.perf_counter() - start) * 1000
-        except RuntimeError:
+        except RuntimeError as e:
             # The encoding pipeline uses asyncio.run() -> to_thread() internally.
             # If the process is shutting down, the default executor is closed and
             # to_thread raises "cannot schedule new futures after shutdown".
             # Silently abandon the save -- the process is exiting anyway.
-            return []
+            # Any other RuntimeError must propagate so the save future's
+            # done-callback reports it via MemorySaveFailedEvent.
+            if "cannot schedule new futures" in str(e):
+                return []
+            raise
 
         try:
             crewai_event_bus.emit(

lib/crewai/src/crewai/project/__init__.py

@@ -14,6 +14,8 @@ from crewai.project.annotations import (
     tool,
 )
 from crewai.project.crew_base import CrewBase
+from crewai.project.crew_loader import load_crew, load_crew_and_kickoff
+from crewai.project.json_loader import load_agent, strip_jsonc_comments
 
 
 __all__ = [
@@ -25,8 +27,12 @@ __all__ = [
     "callback",
     "crew",
     "llm",
+    "load_agent",
+    "load_crew",
+    "load_crew_and_kickoff",
     "output_json",
     "output_pydantic",
+    "strip_jsonc_comments",
     "task",
     "tool",
 ]

lib/crewai/src/crewai/project/crew_loader.py

@@ -0,0 +1,101 @@
+"""Load crew definitions from JSON/JSONC files and produce Crew instances."""
+
+from __future__ import annotations
+
+from pathlib import Path
+from typing import Any
+
+from pydantic import ValidationError
+
+from crewai.project.json_loader import (
+    JSONProjectError,
+    JSONProjectValidationError,
+    _crew_kwargs_from_definition,
+    _task_kwargs_from_definition,
+    load_json_crew_project,
+)
+
+
+def load_crew(
+    source: Path | str,
+    agents_dir: Path | None = None,
+) -> tuple[Any, dict[str, Any]]:
+    """Load a ``Crew`` from a JSON/JSONC definition file.
+
+    The definition file describes the crew's agents, tasks, process type, and
+    default inputs.  Agent definitions are resolved from individual
+    ``<name>.jsonc`` / ``<name>.json`` files inside an ``agents/`` directory.
+    """
+    from crewai import Agent, Crew, Task
+
+    crew_path = Path(source)
+    project = load_json_crew_project(crew_path, agents_dir=agents_dir)
+
+    agents_map: dict[str, Any] = {}
+    for name in project.agent_names:
+        agent_def = project.agents[name]
+        try:
+            agents_map[name] = Agent(**agent_def.kwargs)
+        except ValidationError as exc:
+            raise JSONProjectError(
+                f"{agent_def.path}: validation failed: {exc}"
+            ) from exc
+        except Exception as exc:
+            raise JSONProjectError(
+                f"{agent_def.path}: failed to load agent: {exc}"
+            ) from exc
+
+    tasks_list: list[Task] = []
+    task_name_map: dict[str, Task] = {}
+
+    for index, task_defn in enumerate(project.task_definitions):
+        source_label = f"{crew_path}: tasks[{index}]"
+        task_kwargs = _task_kwargs_from_definition(
+            task_defn,
+            agents_map=agents_map,
+            task_name_map=task_name_map,
+            source=source_label,
+            project_root=crew_path.parent,
+        )
+        try:
+            task = Task(**task_kwargs)
+        except ValidationError as exc:
+            raise JSONProjectError(f"{source_label}: validation failed: {exc}") from exc
+
+        tasks_list.append(task)
+        task_name = task_defn.get("name")
+        if isinstance(task_name, str) and task_name:
+            task_name_map[task_name] = task
+
+    crew_kwargs = _crew_kwargs_from_definition(
+        project.definition,
+        agents=list(agents_map.values()),
+        tasks=tasks_list,
+        agents_map=agents_map,
+        source=crew_path,
+    )
+
+    try:
+        crew = Crew(**crew_kwargs)
+    except ValidationError as exc:
+        raise JSONProjectError(f"{crew_path}: validation failed: {exc}") from exc
+    except JSONProjectValidationError:
+        raise
+    except Exception as exc:
+        raise JSONProjectError(f"{crew_path}: failed to load crew: {exc}") from exc
+
+    return crew, project.definition.get("inputs", {})
+
+
+def load_crew_and_kickoff(
+    crew_path: Path | str,
+    input_overrides: dict[str, Any] | None = None,
+) -> Any:
+    """Convenience function: load a crew and immediately kick it off."""
+    crew, default_inputs = load_crew(crew_path)
+
+    merged_inputs = {**default_inputs}
+    if input_overrides:
+        merged_inputs.update(input_overrides)
+
+    return crew.kickoff(inputs=merged_inputs)

lib/crewai/src/crewai/project/json_loader.py

@@ -0,0 +1,837 @@
+"""Loader utilities for JSON/JSONC agent, crew, task, and tool definitions."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+import json
+import logging
+from pathlib import Path
+import re
+from typing import Any
+
+from pydantic import ValidationError
+
+
+logger = logging.getLogger(__name__)
+
+
+class JSONProjectError(ValueError):
+    """User-facing error raised while loading JSON-first crew projects."""
+
+
+class JSONProjectValidationError(JSONProjectError):
+    """Aggregates validation errors found without executing a JSON project."""
+
+    def __init__(self, errors: list[str]) -> None:
+        self.errors = errors
+        super().__init__("\n".join(errors))
+
+
+_AGENT_RUNTIME_FIELDS = {
+    "id",
+    "crew",
+    "cache_handler",
+    "tools_handler",
+    "tools_results",
+    "knowledge",
+    "knowledge_storage",
+    "adapted_agent",
+    "agent_knowledge_context",
+    "crew_knowledge_context",
+    "knowledge_search_query",
+    "execution_context",
+    "checkpoint_kickoff_event_id",
+}
+
+_TASK_RUNTIME_FIELDS = {
+    "id",
+    "used_tools",
+    "tools_errors",
+    "delegations",
+    "output",
+    "processed_by_agents",
+    "retry_count",
+    "start_time",
+    "end_time",
+    "checkpoint_original_description",
+    "checkpoint_original_expected_output",
+}
+
+_CREW_RUNTIME_FIELDS = {
+    "id",
+    "usage_metrics",
+    "task_execution_output_json_files",
+    "execution_logs",
+    "token_usage",
+    "execution_context",
+    "checkpoint_inputs",
+    "checkpoint_train",
+    "checkpoint_kickoff_event_id",
+}
+
+
+JSON_PROJECT_EXTENSIONS = (".jsonc", ".json")
+
+
+@dataclass(frozen=True)
+class JSONAgentDefinition:
+    """Parsed JSON agent definition and constructor kwargs."""
+
+    name: str
+    path: Path
+    definition: dict[str, Any]
+    kwargs: dict[str, Any]
+
+
+@dataclass(frozen=True)
+class JSONCrewProject:
+    """Parsed JSON crew project used by runtime loading and validation."""
+
+    crew_path: Path
+    agents_dir: Path
+    definition: dict[str, Any]
+    agent_names: list[str]
+    agents: dict[str, JSONAgentDefinition]
+    task_definitions: list[dict[str, Any]]
+
+
+def find_json_project_file(directory: str | Path, stem: str) -> Path | None:
+    """Return ``stem.jsonc`` or ``stem.json``, preferring JSONC."""
+    root = Path(directory)
+    for ext in JSON_PROJECT_EXTENSIONS:
+        candidate = root / f"{stem}{ext}"
+        if candidate.exists():
+            return candidate
+    return None
+
+
+def find_crew_json_file(project_root: str | Path = ".") -> Path | None:
+    """Find the JSON crew definition in a project root."""
+    return find_json_project_file(project_root, "crew")
+
+
+def strip_jsonc_comments(text: str) -> str:
+    """Strip JSONC comments and trailing commas while preserving string values."""
+    without_comments = _strip_jsonc_comments(text)
+    return _strip_trailing_commas(without_comments)
+
+
+def parse_jsonc(text: str, source: str | Path = "<string>") -> Any:
+    """Parse JSON/JSONC text into Python data with path-aware error messages."""
+    source_label = str(source)
+    try:
+        return json.loads(strip_jsonc_comments(text))
+    except json.JSONDecodeError as exc:
+        raise JSONProjectError(
+            f"{source_label}: invalid JSON at line {exc.lineno}, "
+            f"column {exc.colno}: {exc.msg}"
+        ) from exc
+
+
+def load_jsonc_file(source: str | Path) -> Any:
+    """Load a JSON or JSONC file."""
+    path = Path(source)
+    return parse_jsonc(path.read_text(encoding="utf-8"), source=path)
+
+
+def load_agent(source: str | Path) -> Any:
+    """Load an existing ``Agent`` from a ``.json`` / ``.jsonc`` definition file."""
+    from crewai import Agent
+
+    path = Path(source)
+    defn = _expect_object(load_jsonc_file(path), path)
+    root = path.parent.parent if path.parent.name == "agents" else Path.cwd()
+    agent_kwargs = _agent_kwargs_from_definition(defn, path, project_root=root)
+
+    try:
+        return Agent(**agent_kwargs)
+    except ValidationError as exc:
+        raise JSONProjectError(_format_validation_error(path, exc)) from exc
+    except Exception as exc:
+        raise JSONProjectError(f"{path}: failed to load agent: {exc}") from exc
+
+
+def validate_crew_project(
+    source: str | Path,
+    agents_dir: Path | None = None,
+) -> JSONCrewProject:
+    """Validate JSON crew structure without kicking off the crew."""
+    return load_json_crew_project(source, agents_dir=agents_dir, collect_errors=True)
+
+
+def load_json_crew_project(
+    source: str | Path,
+    agents_dir: Path | None = None,
+    *,
+    collect_errors: bool = False,
+) -> JSONCrewProject:
+    """Parse and structurally validate a JSON crew project.
+
+    When ``collect_errors`` is true, all discoverable structural errors are
+    returned as a single ``JSONProjectValidationError`` for deploy validation.
+    Runtime loading keeps the previous fail-fast behavior where possible.
+    """
+    crew_path = Path(source)
+    if agents_dir is None:
+        agents_dir = crew_path.parent / "agents"
+
+    errors: list[str] = []
+
+    def fail(message: str, exc_type: type[Exception] = JSONProjectError) -> None:
+        if collect_errors:
+            errors.append(message)
+            return
+        raise exc_type(message)
+
+    def fail_many(messages: list[str]) -> None:
+        if not messages:
+            return
+        if collect_errors:
+            errors.extend(messages)
+            return
+        raise JSONProjectValidationError(messages)
+
+    try:
+        defn = _expect_object(load_jsonc_file(crew_path), crew_path)
+    except Exception as exc:
+        if collect_errors:
+            raise JSONProjectValidationError([str(exc)]) from exc
+        raise
+
+    fail_many(
+        _field_errors(
+            defn,
+            _crew_allowed_fields(),
+            _CREW_RUNTIME_FIELDS,
+            crew_path,
+            {"inputs"},
+        )
+    )
+
+    agent_names = defn.get("agents", [])
+    if not isinstance(agent_names, list) or not agent_names:
+        fail(f"{crew_path}: 'agents' must be a non-empty list")
+        agent_names = []
+
+    agents_dir = Path(agents_dir)
+    agent_definitions: dict[str, JSONAgentDefinition] = {}
+    for agent_name in agent_names:
+        if not isinstance(agent_name, str) or not agent_name:
+            fail(f"{crew_path}: each agent reference must be a non-empty string")
+            continue
+        agent_file = find_json_project_file(agents_dir, agent_name)
+        if agent_file is None:
+            message = (
+                f"Agent definition for '{agent_name}' not found in {agents_dir} "
+                f"(tried {agent_name}.jsonc and {agent_name}.json)"
+            )
+            if collect_errors:
+                errors.append(
+                    f"{crew_path}: agent '{agent_name}' not found in {agents_dir} "
+                    f"(tried {agent_name}.jsonc and {agent_name}.json)"
+                )
+            else:
+                raise FileNotFoundError(message)
+            continue
+        try:
+            agent_defn = _expect_object(load_jsonc_file(agent_file), agent_file)
+            agent_kwargs = _agent_kwargs_from_definition(
+                agent_defn,
+                agent_file,
+                # Validation must never execute project code (custom tools).
+                resolve_tools=not collect_errors,
+                project_root=crew_path.parent,
+            )
+        except Exception as exc:
+            if collect_errors:
+                errors.append(str(exc))
+                continue
+            raise
+        agent_definitions[agent_name] = JSONAgentDefinition(
+            name=agent_name,
+            path=agent_file,
+            definition=agent_defn,
+            kwargs=agent_kwargs,
+        )
+
+    task_defs = defn.get("tasks", [])
+    if not isinstance(task_defs, list) or not task_defs:
+        fail(f"{crew_path}: 'tasks' must be a non-empty list")
+        task_defs = []
+
+    known_tasks: set[str] = set()
+    known_agents = {name for name in agent_names if isinstance(name, str)}
+    for index, task_defn in enumerate(task_defs):
+        task_path = f"{crew_path}: tasks[{index}]"
+        if not isinstance(task_defn, dict):
+            fail(f"{task_path} must be an object")
+            continue
+        fail_many(
+            _field_errors(
+                task_defn,
+                _task_allowed_fields(),
+                _TASK_RUNTIME_FIELDS,
+                task_path,
+            )
+        )
+        missing_required = [
+            f"{task_path} missing required field '{required}'"
+            for required in ("description", "expected_output")
+            if required not in task_defn
+        ]
+        fail_many(missing_required)
+
+        agent_ref = task_defn.get("agent")
+        if agent_ref is not None and agent_ref not in known_agents:
+            fail(
+                f"{task_path} references agent '{agent_ref}' which is not in the crew agents list"
+            )
+
+        fail_many(
+            _tool_definition_errors(task_defn.get("tools"), task_path, crew_path.parent)
+        )
+
+        context_names = task_defn.get("context")
+        if context_names is not None:
+            if not isinstance(context_names, list):
+                fail(f"{task_path} field 'context' must be a list of task names")
+            else:
+                fail_many(
+                    [
+                        f"{task_path} has context reference '{ctx_name}' but that task "
+                        "has not been defined yet"
+                        for ctx_name in context_names
+                        if ctx_name not in known_tasks
+                    ]
+                )
+
+        task_name = task_defn.get("name")
+        if isinstance(task_name, str) and task_name:
+            known_tasks.add(task_name)
+
+    if errors:
+        raise JSONProjectValidationError(errors)
+
+    return JSONCrewProject(
+        crew_path=crew_path,
+        agents_dir=agents_dir,
+        definition=defn,
+        agent_names=list(agent_names),
+        agents=agent_definitions,
+        task_definitions=task_defs,
+    )
+
+
+def _strip_jsonc_comments(text: str) -> str:
+    result: list[str] = []
+    i = 0
+    in_string = False
+    escape = False
+
+    while i < len(text):
+        char = text[i]
+
+        if in_string:
+            result.append(char)
+            if escape:
+                escape = False
+            elif char == "\\":
+                escape = True
+            elif char == '"':
+                in_string = False
+            i += 1
+            continue
+
+        if char == '"':
+            in_string = True
+            result.append(char)
+            i += 1
+            continue
+
+        next_char = text[i + 1] if i + 1 < len(text) else ""
+        if char == "/" and next_char == "/":
+            i += 2
+            while i < len(text) and text[i] not in "\r\n":
+                i += 1
+            continue
+
+        if char == "/" and next_char == "*":
+            i += 2
+            closed = False
+            while i < len(text) - 1:
+                if text[i] == "\n":
+                    result.append("\n")
+                if text[i] == "*" and text[i + 1] == "/":
+                    i += 2
+                    closed = True
+                    break
+                i += 1
+            if not closed:
+                raise JSONProjectError("unterminated block comment in JSONC input")
+            continue
+
+        result.append(char)
+        i += 1
+
+    return "".join(result)
+
+
+def _strip_trailing_commas(text: str) -> str:
+    result: list[str] = []
+    i = 0
+    in_string = False
+    escape = False
+
+    while i < len(text):
+        char = text[i]
+
+        if in_string:
+            result.append(char)
+            if escape:
+                escape = False
+            elif char == "\\":
+                escape = True
+            elif char == '"':
+                in_string = False
+            i += 1
+            continue
+
+        if char == '"':
+            in_string = True
+            result.append(char)
+            i += 1
+            continue
+
+        if char == ",":
+            j = i + 1
+            while j < len(text) and text[j].isspace():
+                j += 1
+            if j < len(text) and text[j] in "}]":
+                i += 1
+                continue
+
+        result.append(char)
+        i += 1
+
+    return "".join(result)
+
+
+def _expect_object(value: Any, source: str | Path) -> dict[str, Any]:
+    if not isinstance(value, dict):
+        raise JSONProjectError(f"{source}: expected a JSON object")
+    return value
+
+
+def _agent_kwargs_from_definition(
+    defn: dict[str, Any],
+    path: Path | str,
+    *,
+    resolve_tools: bool = True,
+    project_root: Path | None = None,
+) -> dict[str, Any]:
+    errors = _field_errors(
+        defn,
+        _agent_allowed_fields(),
+        _AGENT_RUNTIME_FIELDS,
+        path,
+        {"settings"},
+    )
+    for required in ("role", "goal", "backstory"):
+        if required not in defn:
+            errors.append(f"{path}: missing required field '{required}'")
+
+    settings = defn.get("settings", {})
+    if settings is None:
+        settings = {}
+    if not isinstance(settings, dict):
+        errors.append(f"{path}: 'settings' must be an object when provided")
+        settings = {}
+    else:
+        errors.extend(
+            _field_errors(
+                settings,
+                _agent_allowed_fields(),
+                _AGENT_RUNTIME_FIELDS,
+                f"{path}: settings",
+            )
+        )
+
+    if errors:
+        raise JSONProjectValidationError(errors)
+
+    agent_kwargs = {
+        key: value for key, value in defn.items() if key in _agent_allowed_fields()
+    }
+    agent_kwargs.update(settings)
+    if resolve_tools:
+        _resolve_tool_fields(agent_kwargs, project_root=project_root)
+    else:
+        # Validation/deploy mode: check tool declarations structurally without
+        # importing or instantiating anything — custom:<name> tools execute
+        # project Python on resolution, which must not happen here.
+        tool_errors = _tool_definition_errors(
+            agent_kwargs.get("tools"), path, project_root
+        )
+        if tool_errors:
+            raise JSONProjectValidationError(tool_errors)
+    return agent_kwargs
+
+
+def _task_kwargs_from_definition(
+    task_defn: dict[str, Any],
+    agents_map: dict[str, Any],
+    task_name_map: dict[str, Any],
+    source: str,
+    project_root: Path | None = None,
+) -> dict[str, Any]:
+    errors = _field_errors(
+        task_defn,
+        _task_allowed_fields(),
+        _TASK_RUNTIME_FIELDS,
+        source,
+    )
+    if errors:
+        raise JSONProjectValidationError(errors)
+
+    task_kwargs = {
+        key: value for key, value in task_defn.items() if key in _task_allowed_fields()
+    }
+
+    agent_ref = task_kwargs.get("agent")
+    if agent_ref is not None and isinstance(agent_ref, str):
+        if agent_ref not in agents_map:
+            raise JSONProjectError(
+                f"{source} references agent '{agent_ref}' which is not in the crew agents list"
+            )
+        task_kwargs["agent"] = agents_map[agent_ref]
+
+    context_names = task_kwargs.get("context")
+    if context_names:
+        context_tasks: list[Any] = []
+        for ctx_name in context_names:
+            if ctx_name not in task_name_map:
+                raise JSONProjectError(
+                    f"{source} has context reference '{ctx_name}' but that task "
+                    "has not been defined yet"
+                )
+            context_tasks.append(task_name_map[ctx_name])
+        task_kwargs["context"] = context_tasks
+
+    _resolve_tool_fields(task_kwargs, project_root=project_root)
+    return task_kwargs
+
+
+def _crew_kwargs_from_definition(
+    defn: dict[str, Any],
+    agents: list[Any],
+    tasks: list[Any],
+    agents_map: dict[str, Any],
+    source: Path | str,
+) -> dict[str, Any]:
+    errors = _field_errors(
+        defn,
+        _crew_allowed_fields(),
+        _CREW_RUNTIME_FIELDS,
+        source,
+        {"inputs"},
+    )
+    if errors:
+        raise JSONProjectValidationError(errors)
+
+    crew_kwargs = {
+        key: value for key, value in defn.items() if key in _crew_allowed_fields()
+    }
+    crew_kwargs["agents"] = agents
+    crew_kwargs["tasks"] = tasks
+
+    manager_agent = crew_kwargs.get("manager_agent")
+    if isinstance(manager_agent, str):
+        if manager_agent not in agents_map:
+            raise JSONProjectError(
+                f"{source}: manager_agent '{manager_agent}' is not in the crew agents list"
+            )
+        crew_kwargs["manager_agent"] = agents_map[manager_agent]
+
+    return crew_kwargs
+
+
+def _resolve_tool_fields(
+    kwargs: dict[str, Any], project_root: Path | None = None
+) -> None:
+    tools = kwargs.get("tools")
+    if tools is not None:
+        kwargs["tools"] = _resolve_tools(tools, project_root=project_root)
+
+
+def _field_errors(
+    data: dict[str, Any],
+    allowed_fields: set[str],
+    runtime_fields: set[str],
+    source: str | Path,
+    extra_allowed: set[str] | None = None,
+) -> list[str]:
+    extra_allowed = extra_allowed or set()
+    keys = set(data)
+    runtime = sorted(keys & runtime_fields)
+    unknown = sorted(keys - allowed_fields - runtime_fields - extra_allowed)
+
+    errors: list[str] = []
+    if runtime:
+        errors.append(
+            f"{source}: runtime-only field(s) are not supported in JSON config: "
+            + ", ".join(runtime)
+        )
+    if unknown:
+        errors.append(f"{source}: unsupported field(s): " + ", ".join(unknown))
+    return errors
+
+
+def _agent_allowed_fields() -> set[str]:
+    from crewai import Agent
+
+    return set(Agent.model_fields) - _AGENT_RUNTIME_FIELDS
+
+
+def _task_allowed_fields() -> set[str]:
+    from crewai import Task
+
+    return set(Task.model_fields) - _TASK_RUNTIME_FIELDS
+
+
+def _crew_allowed_fields() -> set[str]:
+    from crewai import Crew
+
+    return set(Crew.model_fields) - _CREW_RUNTIME_FIELDS
+
+
+def _format_validation_error(path: str | Path, exc: ValidationError) -> str:
+    return f"{path}: validation failed: {exc}"
+
+
+def _resolve_tools(tool_defs: list[Any], project_root: Path | None = None) -> list[Any]:
+    """Resolve tool specs into tool instances or serialized BaseTool dicts.
+
+    Strings keep the existing shorthand behavior. Dicts are passed through so
+    ``BaseTool``'s Pydantic validator can hydrate serialized ``tool_type`` data.
+    """
+    if not isinstance(tool_defs, list):
+        raise JSONProjectError("'tools' must be a list")
+
+    tools: list[Any] = []
+    for tool_def in tool_defs:
+        if isinstance(tool_def, dict):
+            tools.append(tool_def)
+            continue
+        if not isinstance(tool_def, str):
+            raise JSONProjectError(
+                f"Tool definitions must be strings or objects, got {type(tool_def).__name__}"
+            )
+        if not tool_def:
+            continue
+        if tool_def.startswith("custom:"):
+            tools.append(_resolve_custom_tool(tool_def[7:], project_root=project_root))
+            continue
+        try:
+            tool_cls = _find_tool_class(tool_def)
+        except Exception as e:
+            raise JSONProjectError(f"Failed to resolve tool '{tool_def}': {e}") from e
+        if tool_cls is None:
+            raise JSONProjectError(
+                f"Unknown tool '{tool_def}'. Tool names must match a class from "
+                f"the 'crewai_tools' package (e.g. 'SerperDevTool') or use the "
+                f"'custom:<name>' prefix for a tool defined in tools/<name>.py."
+            )
+        try:
+            tools.append(tool_cls())
+        except Exception as e:
+            raise JSONProjectError(
+                f"Failed to initialize tool '{tool_def}': {e}"
+            ) from e
+    return tools
+
+
+_tool_class_cache: dict[str, type | None] = {}
+
+
+def _find_tool_class(name: str) -> type | None:
+    """Look up a tool class by name from the ``crewai_tools`` package."""
+    if name in _tool_class_cache:
+        return _tool_class_cache[name]
+
+    candidates = [name]
+    if not name.endswith("Tool"):
+        candidates.append(name + "Tool")
+    snake_pascal = "".join(word.capitalize() for word in name.split("_")) + "Tool"
+    if snake_pascal not in candidates:
+        candidates.append(snake_pascal)
+
+    for class_name in candidates:
+        cls = _try_import_tool(class_name)
+        if cls is not None:
+            _tool_class_cache[name] = cls
+            return cls
+
+    _tool_class_cache[name] = None
+    return None
+
+
+def _try_import_tool(class_name: str) -> type | None:
+    """Attempt to import a single tool class without loading all of crewai_tools."""
+    import re as _re
+
+    base = (
+        class_name.removesuffix("Tool") if class_name.endswith("Tool") else class_name
+    )
+    snake = _re.sub(r"(?<=[a-z0-9])(?=[A-Z])", "_", base).lower()
+    tool_snake = snake + "_tool" if not snake.endswith("_tool") else snake
+
+    module_paths = [
+        f"crewai_tools.tools.{tool_snake}.{tool_snake}",
+        f"crewai_tools.tools.{tool_snake}",
+    ]
+
+    for mod_path in module_paths:
+        cls = _import_tool_class(mod_path, class_name)
+        if cls is not None:
+            return cls
+
+    try:
+        import crewai_tools
+
+        return getattr(crewai_tools, class_name, None)
+    except ImportError:
+        return None
+
+
+def _import_tool_class(mod_path: str, class_name: str) -> type | None:
+    try:
+        import importlib
+
+        mod = importlib.import_module(mod_path)
+    except (ImportError, ModuleNotFoundError):
+        return None
+    return getattr(mod, class_name, None)
+
+
+_CUSTOM_TOOL_NAME_RE = re.compile(r"[A-Za-z_][A-Za-z0-9_]*")
+
+
+def _custom_tool_file(tool_name: str, project_root: Path | None) -> Path:
+    """Return the validated path of a custom tool inside ``tools/``.
+
+    Rejects names that aren't plain identifiers and (belt-and-suspenders)
+    any resolved path that escapes the project's ``tools/`` directory, so
+    ``custom:../evil`` or absolute-path style names cannot execute code
+    outside the project.
+    """
+    if not _CUSTOM_TOOL_NAME_RE.fullmatch(tool_name):
+        raise JSONProjectError(
+            f"Invalid custom tool name 'custom:{tool_name}': names must match "
+            f"[A-Za-z_][A-Za-z0-9_]* and resolve to tools/<name>.py inside "
+            f"the project."
+        )
+    tools_dir = ((project_root or Path.cwd()) / "tools").resolve()
+    tool_file = (tools_dir / f"{tool_name}.py").resolve()
+    try:
+        tool_file.relative_to(tools_dir)
+    except ValueError:
+        raise JSONProjectError(
+            f"Custom tool 'custom:{tool_name}' resolves outside the project's "
+            f"tools/ directory."
+        ) from None
+    return tool_file
+
+
+def _tool_definition_errors(
+    tool_defs: Any, source: Path | str, project_root: Path | None
+) -> list[str]:
+    """Structurally validate tool declarations WITHOUT importing anything.
+
+    Used by validation/deploy paths where executing project code (which
+    ``custom:`` resolution does) would be unsafe. Library tool names are not
+    resolved here either — that requires importing crewai_tools modules and
+    would falsely fail when optional dependencies are absent in the
+    validation environment.
+    """
+    if tool_defs is None:
+        return []
+    if not isinstance(tool_defs, list):
+        return [f"{source}: 'tools' must be a list"]
+    errors: list[str] = []
+    for tool_def in tool_defs:
+        if isinstance(tool_def, dict):
+            continue
+        if not isinstance(tool_def, str):
+            errors.append(
+                f"{source}: tool definitions must be strings or objects, "
+                f"got {type(tool_def).__name__}"
+            )
+            continue
+        if not tool_def.startswith("custom:"):
+            continue
+        try:
+            tool_file = _custom_tool_file(tool_def[7:], project_root)
+        except JSONProjectError as exc:
+            errors.append(f"{source}: {exc}")
+            continue
+        if not tool_file.exists():
+            errors.append(
+                f"{source}: custom tool '{tool_def}' not found: expected "
+                f"{tool_file}. Create the file with a BaseTool subclass, or "
+                f"remove the tool from your crew JSON."
+            )
+    return errors
+
+
+def _resolve_custom_tool(tool_name: str, project_root: Path | None = None) -> Any:
+    """Resolve a custom tool from the project's ``tools/`` directory.
+
+    Note: ``custom:<name>`` tools execute ``tools/<name>.py`` as local Python
+    code at load time — JSON configs referencing them are no longer pure data.
+    Only run JSON crew projects from sources you trust. Validation paths must
+    use ``_tool_definition_errors`` instead, which never executes anything.
+    """
+    tool_file = _custom_tool_file(tool_name, project_root)
+    if not tool_file.exists():
+        raise JSONProjectError(
+            f"Custom tool 'custom:{tool_name}' not found: expected {tool_file}. "
+            f"Create the file with a BaseTool subclass, or remove the tool from "
+            f"your crew JSON."
+        )
+    try:
+        import importlib.util
+
+        spec = importlib.util.spec_from_file_location(
+            f"custom_tools.{tool_name}", tool_file
+        )
+        if spec is None or spec.loader is None:
+            raise JSONProjectError(
+                f"Could not load custom tool 'custom:{tool_name}' from {tool_file}"
+            )
+        logger.debug("Executing custom tool module: %s", tool_file)
+        module = importlib.util.module_from_spec(spec)
+        spec.loader.exec_module(module)
+
+        from crewai.tools.base_tool import BaseTool
+
+        for attr_name in dir(module):
+            attr = getattr(module, attr_name)
+            if (
+                isinstance(attr, type)
+                and issubclass(attr, BaseTool)
+                and attr is not BaseTool
+            ):
+                # Concrete subclasses supply name/description defaults that
+                # BaseTool's signature requires.
+                tool_cls: type[Any] = attr
+                return tool_cls()
+        raise JSONProjectError(
+            f"No BaseTool subclass found in {tool_file}. Custom tools must "
+            f"define a class inheriting from crewai.tools.BaseTool."
+        )
+    except JSONProjectError:
+        raise
+    except Exception as e:
+        raise JSONProjectError(
+            f"Failed to load custom tool 'custom:{tool_name}' from {tool_file}: {e}"
+        ) from e

lib/crewai/src/crewai/rag/embeddings/providers/microsoft/azure.py

@@ -5,7 +5,7 @@ from typing import Any
 from chromadb.utils.embedding_functions.openai_embedding_function import (
     OpenAIEmbeddingFunction,
 )
-from pydantic import AliasChoices, Field
+from pydantic import AliasChoices, Field, model_validator
 
 from crewai.rag.core.base_embeddings_provider import BaseEmbeddingsProvider
 
@@ -13,6 +13,14 @@ from crewai.rag.core.base_embeddings_provider import BaseEmbeddingsProvider
 class AzureProvider(BaseEmbeddingsProvider[OpenAIEmbeddingFunction]):
     """Azure OpenAI embeddings provider."""
 
+    @model_validator(mode="before")
+    @classmethod
+    def _normalize_model_alias(cls, data: Any) -> Any:
+        if isinstance(data, dict) and "model" in data and "model_name" not in data:
+            data = data.copy()
+            data["model_name"] = data["model"]
+        return data
+
     embedding_callable: type[OpenAIEmbeddingFunction] = Field(
         default=OpenAIEmbeddingFunction,
         description="Azure OpenAI embedding function class",
@@ -43,13 +51,11 @@ class AzureProvider(BaseEmbeddingsProvider[OpenAIEmbeddingFunction]):
         ),
     )
     model_name: str = Field(
-        default="text-embedding-ada-002",
+        default="text-embedding-3-large",
         description="Model name to use for embeddings",
         validation_alias=AliasChoices(
             "EMBEDDINGS_OPENAI_MODEL_NAME",
-            "OPENAI_MODEL_NAME",
             "AZURE_OPENAI_MODEL_NAME",
-            "model",
         ),
     )
     default_headers: dict[str, Any] | None = Field(

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

@@ -12,7 +12,7 @@ class AzureProviderConfig(TypedDict, total=False):
     api_base: str
     api_type: Annotated[str, "azure"]
     api_version: str
-    model_name: Annotated[str, "text-embedding-ada-002"]
+    model_name: Annotated[str, "text-embedding-3-large"]
     default_headers: dict[str, Any]
     dimensions: int
     deployment_id: Required[str]

lib/crewai/src/crewai/rag/embeddings/providers/openai/openai_provider.py

@@ -5,7 +5,7 @@ from typing import Any
 from chromadb.utils.embedding_functions.openai_embedding_function import (
     OpenAIEmbeddingFunction,
 )
-from pydantic import AliasChoices, Field
+from pydantic import AliasChoices, Field, model_validator
 
 from crewai.rag.core.base_embeddings_provider import BaseEmbeddingsProvider
 
@@ -13,6 +13,14 @@ from crewai.rag.core.base_embeddings_provider import BaseEmbeddingsProvider
 class OpenAIProvider(BaseEmbeddingsProvider[OpenAIEmbeddingFunction]):
     """OpenAI embeddings provider."""
 
+    @model_validator(mode="before")
+    @classmethod
+    def _normalize_model_alias(cls, data: Any) -> Any:
+        if isinstance(data, dict) and "model" in data and "model_name" not in data:
+            data = data.copy()
+            data["model_name"] = data["model"]
+        return data
+
     embedding_callable: type[OpenAIEmbeddingFunction] = Field(
         default=OpenAIEmbeddingFunction,
         description="OpenAI embedding function class",
@@ -23,12 +31,11 @@ class OpenAIProvider(BaseEmbeddingsProvider[OpenAIEmbeddingFunction]):
         validation_alias=AliasChoices("EMBEDDINGS_OPENAI_API_KEY", "OPENAI_API_KEY"),
     )
     model_name: str = Field(
-        default="text-embedding-ada-002",
+        default="text-embedding-3-large",
         description="Model name to use for embeddings",
         validation_alias=AliasChoices(
             "EMBEDDINGS_OPENAI_MODEL_NAME",
-            "OPENAI_MODEL_NAME",
-            "model",
+            "model_name",
         ),
     )
     api_base: str | None = Field(

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

@@ -9,7 +9,7 @@ class OpenAIProviderConfig(TypedDict, total=False):
     """Configuration for OpenAI provider."""
 
     api_key: str
-    model_name: Annotated[str, "text-embedding-ada-002"]
+    model_name: Annotated[str, "text-embedding-3-large"]
     api_base: str
     api_type: str
     api_version: str

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

@@ -931,7 +931,7 @@ class Telemetry:
             value: The attribute value.
         """
 
-        if span is None:
+        if span is None or value is None:
             return
 
         def _operation() -> None:
@@ -982,6 +982,11 @@ class Telemetry:
         def _operation() -> None:
             tracer = trace.get_tracer("crewai.telemetry")
             span = tracer.start_span("Flow Execution")
+            self._add_attribute(
+                span,
+                "crewai_version",
+                version("crewai"),
+            )
             self._add_attribute(span, "flow_name", flow_name)
             self._add_attribute(span, "node_names", json.dumps(node_names))
             close_span(span)

lib/crewai/src/crewai/utilities/agent_utils.py

@@ -65,6 +65,15 @@ class SummaryContent(TypedDict):
 console = Console()
 
 _MULTIPLE_NEWLINES: Final[re.Pattern[str]] = re.compile(r"\n+")
+_NATIVE_TOOL_UNSUPPORTED_PATTERNS: Final[tuple[str, ...]] = (
+    "does not support tools",
+    "doesn't support tools",
+    "tools are not supported",
+    "tool calling is not supported",
+    "tool calls are not supported",
+    "function calling is not supported",
+    "does not support function calling",
+)
 
 
 def is_inside_event_loop() -> bool:
@@ -1273,6 +1282,28 @@ def check_native_tool_support(llm: Any, original_tools: list[BaseTool] | None) -
     )
 
 
+def is_native_tool_calling_unsupported_error(error: BaseException) -> bool:
+    """Return whether an error means native tool calling is unavailable."""
+    message = str(error).lower()
+    return any(pattern in message for pattern in _NATIVE_TOOL_UNSUPPORTED_PATTERNS)
+
+
+def build_text_tool_calling_fallback_message(
+    tools_description: str,
+    tools_names: str,
+) -> str:
+    """Build instructions for downgrading native tools to text tool calls."""
+    text_tooling_prompt = I18N_DEFAULT.slice("tools").format(
+        tools=tools_description,
+        tool_names=tools_names,
+    )
+    return (
+        "Native tool calling is unavailable for this model/provider. "
+        "Continue using CrewAI text tool calling instead.\n"
+        f"{text_tooling_prompt}"
+    )
+
+
 def setup_native_tools(
     original_tools: list[BaseTool],
 ) -> tuple[
@@ -1365,6 +1396,8 @@ def execute_single_native_tool_call(
     event_source: Any,
     printer: Printer | None = None,
     verbose: bool = False,
+    plan_step_number: int | None = None,
+    plan_step_description: str | None = None,
 ) -> NativeToolCallResult:
     """Execute a single native tool call with full lifecycle management.
 
@@ -1446,6 +1479,8 @@ def execute_single_native_tool_call(
             from_agent=agent,
             from_task=task,
             agent_key=agent_key,
+            plan_step_number=plan_step_number,
+            plan_step_description=plan_step_description,
         ),
     )
 
@@ -1509,6 +1544,8 @@ def execute_single_native_tool_call(
                         from_agent=agent,
                         from_task=task,
                         agent_key=agent_key,
+                        plan_step_number=plan_step_number,
+                        plan_step_description=plan_step_description,
                         error=e,
                     ),
                 )
@@ -1542,6 +1579,8 @@ def execute_single_native_tool_call(
                 from_agent=agent,
                 from_task=task,
                 agent_key=agent_key,
+                plan_step_number=plan_step_number,
+                plan_step_description=plan_step_description,
                 started_at=started_at,
                 finished_at=datetime.now(),
             ),

lib/crewai/src/crewai/utilities/llm_utils.py

@@ -11,12 +11,13 @@ logger = logging.getLogger(__name__)
 
 
 def create_llm(
-    llm_value: str | LLM | Any | None = None,
+    llm_value: str | dict[str, Any] | LLM | Any | None = None,
 ) -> LLM | BaseLLM | None:
     """Creates or returns an LLM instance based on the given llm_value.
 
     Args:
-        llm_value: LLM instance, model name string, None, or an object with LLM attributes.
+        llm_value: LLM instance, model name string, config dict, None, or an
+            object with LLM attributes.
 
     Returns:
         A BaseLLM instance if successful, or None if something fails.
@@ -32,6 +33,26 @@ def create_llm(
             logger.error(f"Error instantiating LLM from string: {e}")
             raise e
 
+    if isinstance(llm_value, dict):
+        try:
+            model = (
+                llm_value.get("model")
+                or llm_value.get("model_name")
+                or llm_value.get("deployment_name")
+            )
+            if not model:
+                raise ValueError(
+                    "LLM config dictionaries must include 'model', "
+                    "'model_name', or 'deployment_name'"
+                )
+            llm_params = {**llm_value, "model": model}
+            llm_params.pop("model_name", None)
+            llm_params.pop("deployment_name", None)
+            return LLM(**llm_params)
+        except Exception as e:
+            logger.error(f"Error instantiating LLM from dict: {e}")
+            raise e
+
     if llm_value is None:
         return _llm_via_environment_or_fallback()
 

lib/crewai/src/crewai/utilities/planning_handler.py

@@ -52,7 +52,7 @@ class CrewPlanner:
             planning_agent_llm: Optional LLM model for the planning agent. Defaults to None.
         """
         self.tasks = tasks
-        self.planning_agent_llm = planning_agent_llm or "gpt-4o-mini"
+        self.planning_agent_llm = planning_agent_llm or "gpt-5.4-mini"
 
     def _handle_crew_planning(self) -> PlannerTaskPydanticOutput:
         """Handles the Crew planning by creating detailed step-by-step plans for each task.

lib/crewai/src/crewai/utilities/token_counter_callback.py

@@ -13,15 +13,6 @@ from crewai.agents.agent_builder.utilities.base_token_process import TokenProces
 from crewai.utilities.logger_utils import suppress_warnings
 
 
-try:
-    from litellm.integrations.custom_logger import CustomLogger as LiteLLMCustomLogger
-
-    LITELLM_AVAILABLE = True
-except ImportError:
-    LiteLLMCustomLogger = None  # type: ignore[misc, assignment]
-    LITELLM_AVAILABLE = False
-
-
 class TokenCalcHandler(BaseModel):
     """Handler for calculating and tracking token usage in LLM calls.
 

lib/crewai/tests/agents/test_agent.py

@@ -28,6 +28,19 @@ from crewai.tools import tool
 from crewai.utilities import RPMController
 
 
+def test_agent_memory_true_uses_agent_llm_model():
+    agent = Agent(
+        role="test role",
+        goal="test goal",
+        backstory="test backstory",
+        llm="ollama/llama3",
+        memory=True,
+    )
+
+    assert agent.memory is not None
+    assert agent.memory.llm == "ollama/llama3"
+
+
 def test_agent_llm_creation_with_env_vars():
     original_api_key = os.environ.get("OPENAI_API_KEY")
     original_api_base = os.environ.get("OPENAI_API_BASE")

lib/crewai/tests/agents/test_agent_executor.py

@@ -59,6 +59,10 @@ from crewai.experimental.agent_executor import (
 )
 from crewai.agents.parser import AgentAction, AgentFinish
 from crewai.events.event_bus import crewai_event_bus
+from crewai.events.types.observation_events import (
+    PlanStepCompletedEvent,
+    PlanStepStartedEvent,
+)
 from crewai.events.types.tool_usage_events import (
     ToolUsageFinishedEvent,
     ToolUsageStartedEvent,
@@ -318,6 +322,41 @@ class TestAgentExecutor:
         assert result == "native_finished"
         assert get_llm_response_mock.call_args.kwargs["response_model"] is None
 
+    def test_call_llm_native_tools_falls_back_when_provider_rejects_tools(
+        self, mock_dependencies
+    ):
+        """Provider-level unsupported tools errors should downgrade to ReAct."""
+        executor = _build_executor(
+            **mock_dependencies,
+            original_tools=[Mock()],
+            callbacks=[],
+        )
+        executor._openai_tools = [{"type": "function", "function": {"name": "lookup"}}]
+        executor.state.use_native_tools = True
+        executor.state.pending_tool_calls = [Mock()]
+        executor.state.messages = [{"role": "user", "content": "Use a tool"}]
+        executor.tools = [Mock()]
+        executor.tools_names = "lookup"
+        executor.tools_description = "lookup: search for information"
+
+        with patch(
+            "crewai.experimental.agent_executor.get_llm_response",
+            side_effect=RuntimeError(
+                "Error code: 400 - registry.ollama.ai/library/mariner:latest "
+                "does not support tools"
+            ),
+        ):
+            result = executor.call_llm_native_tools()
+
+        assert result == "continue_reasoning"
+        assert executor.state.use_native_tools is False
+        assert executor.state.pending_tool_calls == []
+        assert executor.state.messages[-1]["role"] == "user"
+        assert "Native tool calling is unavailable" in executor.state.messages[-1][
+            "content"
+        ]
+        assert "Action Input" in executor.state.messages[-1]["content"]
+
     def test_finalize_success(self, mock_dependencies):
         """Test finalize with valid AgentFinish."""
         with patch.object(AgentExecutor, "_show_logs") as mock_show_logs:
@@ -545,6 +584,7 @@ class TestStepExecutorCriticalFixes:
 
         tool = Mock()
         tool.name = "count_words"
+        tool.description = "count_words: Counts words in text"
         task = Mock()
         task.name = "test-task"
         task.description = "test task description"
@@ -610,13 +650,126 @@ class TestStepExecutorCriticalFixes:
             "crewai.agents.step_executor.execute_tool_and_check_finality",
             return_value=ToolResult(result="2", result_as_answer=False),
         ):
-            output = step_executor._execute_text_tool_with_events(action)
+            todo = TodoItem(step_number=2, description="Count words")
+            output = step_executor._execute_text_tool_with_events(action, todo)
 
         crewai_event_bus.flush()
 
         assert output == "2"
         assert len(started_events) >= 1
         assert len(finished_events) >= 1
+        assert started_events[-1].plan_step_number == 2
+        assert started_events[-1].plan_step_description == "Count words"
+        assert finished_events[-1].plan_step_number == 2
+        assert finished_events[-1].plan_step_description == "Count words"
+
+    def test_step_executor_falls_back_when_native_tools_are_rejected(
+        self, step_executor
+    ):
+        """Plan steps should retry through text tool calls when native tools fail."""
+        step_executor._use_native_tools = True
+        step_executor._openai_tools = [{"type": "function", "function": {"name": "count_words"}}]
+        step_executor._available_functions = {"count_words": Mock()}
+        todo = TodoItem(step_number=1, description="Count words")
+        context = StepExecutionContext(task_description="task", task_goal="goal")
+
+        with (
+            patch.object(
+                step_executor,
+                "_execute_native",
+                side_effect=RuntimeError(
+                    "registry.ollama.ai/library/mariner:latest does not support tools"
+                ),
+            ),
+            patch.object(
+                step_executor,
+                "_execute_text_parsed",
+                return_value="Counted words",
+            ) as text_parsed,
+        ):
+            result = step_executor.execute(todo, context)
+
+        assert result.success is True
+        assert result.result == "Counted words"
+        assert step_executor._use_native_tools is False
+        fallback_messages = text_parsed.call_args.args[0]
+        # The original conversation is preserved (system + user) and the
+        # text-tooling instructions are appended instead of rebuilding.
+        assert fallback_messages[0]["role"] == "system"
+        assert fallback_messages[-1]["role"] == "user"
+        assert "Action Input" in fallback_messages[-1]["content"]
+
+    def test_plan_step_lifecycle_events_are_emitted_from_todo_transitions(
+        self, mock_dependencies
+    ):
+        """Todo transitions should publish authoritative plan step events."""
+        from crewai.utilities.planning_types import TodoList
+
+        executor = _build_executor(**mock_dependencies)
+        todo = TodoItem(
+            step_number=1,
+            description="Search the official release",
+            tool_to_use="search",
+        )
+        executor.state.todos = TodoList(items=[todo])
+
+        started_events: list[PlanStepStartedEvent] = []
+        completed_events: list[PlanStepCompletedEvent] = []
+
+        with crewai_event_bus.scoped_handlers():
+
+            @crewai_event_bus.on(PlanStepStartedEvent)
+            def _on_started(_source, event):
+                started_events.append(event)
+
+            @crewai_event_bus.on(PlanStepCompletedEvent)
+            def _on_completed(_source, event):
+                completed_events.append(event)
+
+            executor._mark_todo_running(todo)
+            executor._mark_todo_completed(1, result="Found release")
+            crewai_event_bus.flush()
+
+        assert todo.status == "completed"
+        assert len(started_events) == 1
+        assert started_events[0].step_number == 1
+        assert started_events[0].step_description == "Search the official release"
+        assert started_events[0].tool_to_use == "search"
+        assert len(completed_events) == 1
+        assert completed_events[0].success is True
+        assert completed_events[0].step_number == 1
+        assert completed_events[0].result == "Found release"
+
+    def test_failed_todo_transition_emits_failed_plan_step_event(
+        self, mock_dependencies
+    ):
+        """Failed todo transitions should publish failed plan step events."""
+        from crewai.utilities.planning_types import TodoList
+
+        executor = _build_executor(**mock_dependencies)
+        todo = TodoItem(step_number=1, description="Search release")
+        executor.state.todos = TodoList(items=[todo])
+        completed_events: list[PlanStepCompletedEvent] = []
+
+        with crewai_event_bus.scoped_handlers():
+
+            @crewai_event_bus.on(PlanStepCompletedEvent)
+            def _on_completed(_source, event):
+                completed_events.append(event)
+
+            executor._mark_todo_failed(
+                1,
+                result="Error: no result",
+                error="No result",
+            )
+            crewai_event_bus.flush()
+
+        assert todo.status == "failed"
+        assert len(completed_events) == 1
+        assert completed_events[0].success is False
+        assert completed_events[0].step_number == 1
+        assert completed_events[0].result == "Error: no result"
+        assert completed_events[0].error == "No result"
 
     @patch("crewai.experimental.agent_executor.handle_output_parser_exception")
     def test_recover_from_parser_error(
@@ -1649,6 +1802,12 @@ class TestReasoningEffort:
         executor.handle_step_observed_medium = (
             AgentExecutor.handle_step_observed_medium.__get__(executor)
         )
+        executor._mark_todo_completed = (
+            AgentExecutor._mark_todo_completed.__get__(executor)
+        )
+        executor._mark_todo_failed = (
+            AgentExecutor._mark_todo_failed.__get__(executor)
+        )
 
         success_todo = TodoItem(
             step_number=1,
@@ -1715,6 +1874,9 @@ class TestReasoningEffort:
         executor.handle_step_observed_low = (
             AgentExecutor.handle_step_observed_low.__get__(executor)
         )
+        executor._mark_todo_completed = (
+            AgentExecutor._mark_todo_completed.__get__(executor)
+        )
 
         todo = TodoItem(
             step_number=1,
@@ -1748,6 +1910,12 @@ class TestReasoningEffort:
         executor.handle_step_observed_low = (
             AgentExecutor.handle_step_observed_low.__get__(executor)
         )
+        executor._mark_todo_completed = (
+            AgentExecutor._mark_todo_completed.__get__(executor)
+        )
+        executor._mark_todo_failed = (
+            AgentExecutor._mark_todo_failed.__get__(executor)
+        )
 
         todo = TodoItem(
             step_number=1,
@@ -2065,13 +2233,13 @@ class TestTodoStatusTracking:
         from crewai.experimental.agent_executor import AgentExecutor
 
         source = inspect.getsource(AgentExecutor.handle_step_observed_medium)
-        assert "mark_failed" in source, (
-            "handle_step_observed_medium should use mark_failed for failed steps"
+        assert "_mark_todo_failed" in source, (
+            "handle_step_observed_medium should use _mark_todo_failed for failed steps"
         )
         failed_no_replan_idx = source.index("failed but no replan")
         after_comment = source[failed_no_replan_idx:]
-        assert "mark_completed" not in after_comment, (
-            "mark_completed should not be called on failed steps"
+        assert "_mark_todo_completed" not in after_comment, (
+            "_mark_todo_completed should not be called on failed steps"
         )
 
     def test_failed_step_appears_in_get_failed_todos(self):

lib/crewai/tests/agents/test_lite_agent.py

@@ -1096,6 +1096,7 @@ def test_lite_agent_memory_true_resolves_to_default_memory():
     )
     assert agent._memory is not None
     assert isinstance(agent._memory, Memory)
+    assert agent._memory.llm is agent.llm
 
 
 @pytest.mark.filterwarnings("ignore:LiteAgent is deprecated")

lib/crewai/tests/agents/test_native_tool_calling.py

@@ -17,6 +17,7 @@ import pytest
 from pydantic import BaseModel, Field
 
 from crewai import Agent, Crew, Task
+from crewai.agents.parser import AgentFinish
 from crewai.events import crewai_event_bus
 from crewai.hooks import register_after_tool_call_hook, register_before_tool_call_hook
 from crewai.hooks.tool_hooks import ToolCallHookContext
@@ -1196,6 +1197,50 @@ class TestNativeToolCallingJsonParseError:
 
         assert result["result"] == "ran: print(1)"
 
+    def test_native_tool_loop_falls_back_when_provider_rejects_tools(self) -> None:
+        """Unsupported native tools errors should continue through ReAct."""
+
+        class SearchTool(BaseTool):
+            name: str = "search"
+            description: str = "Search for information"
+
+            def _run(self, query: str) -> str:
+                return f"result for {query}"
+
+        executor = self._make_executor([SearchTool()])
+        executor.llm = Mock()
+        executor.messages = [{"role": "user", "content": "Search for CrewAI"}]
+        executor.callbacks = []
+        executor.iterations = 0
+        executor.max_iter = 3
+        executor.request_within_rpm_limit = None
+        executor.respect_context_window = False
+
+        fallback_finish = AgentFinish(
+            thought="done",
+            output="final",
+            text="Final Answer: final",
+        )
+        with (
+            patch(
+                "crewai.agents.crew_agent_executor.get_llm_response",
+                side_effect=RuntimeError(
+                    "registry.ollama.ai/library/mariner:latest does not support tools"
+                ),
+            ),
+            patch.object(
+                executor,
+                "_invoke_loop_react",
+                return_value=fallback_finish,
+            ) as react_loop,
+        ):
+            result = executor._invoke_loop_native_tools()
+
+        assert result is fallback_finish
+        react_loop.assert_called_once()
+        assert "Native tool calling is unavailable" in executor.messages[-1]["content"]
+        assert "Action Input" in executor.messages[-1]["content"]
+
     def test_dict_args_bypass_json_parsing(self) -> None:
         """When func_args is already a dict, no JSON parsing occurs."""
 

lib/crewai/tests/cli/test_run_crew.py

@@ -1,12 +1,22 @@
 """Tests for the ``crewai run`` command and its subprocess plumbing."""
 
+from pathlib import Path
+import sys
+from types import ModuleType
+from types import SimpleNamespace
 from unittest import mock
 
 from click.testing import CliRunner
 import pytest
 
 from crewai_cli.cli import run
-from crewai_cli.run_crew import CrewType, execute_command
+from crewai_cli.run_crew import (
+    CrewType,
+    _load_json_crew_for_tui,
+    _missing_input_names,
+    _prompt_for_missing_inputs,
+    execute_command,
+)
 
 
 @pytest.fixture
@@ -14,15 +24,17 @@ def runner() -> CliRunner:
     return CliRunner()
 
 
-@mock.patch("crewai_cli.cli.run_crew")
+@mock.patch("crewai_cli.run_crew.run_crew")
 def test_run_passes_filename_to_run_crew(run_crew_mock: mock.Mock, runner: CliRunner) -> None:
     result = runner.invoke(run, ["-f", "my_custom_trained.pkl"])
 
-    run_crew_mock.assert_called_once_with(trained_agents_file="my_custom_trained.pkl")
+    run_crew_mock.assert_called_once_with(
+        trained_agents_file="my_custom_trained.pkl",
+    )
     assert result.exit_code == 0
 
 
-@mock.patch("crewai_cli.cli.run_crew")
+@mock.patch("crewai_cli.run_crew.run_crew")
 def test_run_without_filename_passes_none(run_crew_mock: mock.Mock, runner: CliRunner) -> None:
     result = runner.invoke(run)
 
@@ -56,4 +68,101 @@ def test_execute_command_omits_env_var_when_filename_absent(
     execute_command(CrewType.STANDARD)
 
     _, kwargs = subprocess_run.call_args
-    assert "CREWAI_TRAINED_AGENTS_FILE" not in kwargs["env"]
+    assert "CREWAI_TRAINED_AGENTS_FILE" not in kwargs["env"]
+
+
+def test_missing_input_names_scans_agent_and_task_placeholders() -> None:
+    crew = SimpleNamespace(
+        agents=[
+            SimpleNamespace(
+                role="Researcher for {topic}",
+                goal="Write for {audience}",
+                backstory="Ignore escaped {{not_an_input}}",
+            )
+        ],
+        tasks=[
+            SimpleNamespace(
+                description="Research {topic}",
+                expected_output="A post for {channel}",
+                output_file="{slug}.md",
+            )
+        ],
+    )
+
+    assert _missing_input_names(crew, {"topic": "AI"}) == [
+        "audience",
+        "channel",
+        "slug",
+    ]
+
+
+def test_prompt_for_missing_inputs_merges_runtime_values(monkeypatch) -> None:
+    crew = SimpleNamespace(
+        agents=[SimpleNamespace(role="Researcher", goal="Cover {topic}", backstory="")],
+        tasks=[
+            SimpleNamespace(
+                description="Write for {audience}",
+                expected_output="Post",
+                output_file=None,
+            )
+        ],
+    )
+    values = {"audience": "developers"}
+
+    def prompt(label: str, **_kwargs: object) -> str:
+        if "audience" in str(label):
+            return values["audience"]
+        raise AssertionError(f"Unexpected prompt: {label}")
+
+    monkeypatch.setattr("crewai_cli.run_crew.click.prompt", prompt)
+
+    assert _prompt_for_missing_inputs(crew, {"topic": "AI"}) == {
+        "topic": "AI",
+        "audience": "developers",
+    }
+
+
+def test_load_json_crew_for_tui_prepares_metadata_before_prompt(monkeypatch) -> None:
+    class FakeApp:
+        pass
+
+    fake_tui_module = ModuleType("crewai_cli.crew_run_tui")
+    fake_tui_module.CrewRunApp = FakeApp
+    monkeypatch.setitem(sys.modules, "crewai_cli.crew_run_tui", fake_tui_module)
+
+    crew = SimpleNamespace(
+        name="Demo Crew",
+        tasks=[
+            SimpleNamespace(name="research_task", description="Research"),
+            SimpleNamespace(name="", description="Write summary for developers"),
+        ],
+        agents=[
+            SimpleNamespace(role="Researcher", name="researcher"),
+            SimpleNamespace(role="", name="writer"),
+        ],
+    )
+    prepared: list[object] = []
+
+    monkeypatch.setattr(
+        "crewai_cli.run_crew._json_loading_status",
+        lambda _message: mock.MagicMock(),
+    )
+    monkeypatch.setattr(
+        "crewai_cli.run_crew._load_json_crew",
+        lambda _path: (crew, {"topic": "AI"}),
+    )
+    monkeypatch.setattr(
+        "crewai_cli.run_crew._prepare_json_crew_for_tui",
+        lambda loaded_crew: prepared.append(loaded_crew),
+    )
+
+    app_cls, loaded_crew, default_inputs, task_names, agent_names = (
+        _load_json_crew_for_tui(Path("crew.jsonc"))
+    )
+
+    assert app_cls is FakeApp
+    assert loaded_crew is crew
+    assert default_inputs == {"topic": "AI"}
+    assert task_names == ["research_task", "Write summary for developers"]
+    assert agent_names == ["Researcher", "writer"]
+    assert prepared == [crew]

lib/crewai/tests/knowledge/test_knowledge.py

@@ -531,6 +531,7 @@ def test_docling_source(mock_vector_db):
 
 
 @pytest.mark.vcr
+@pytest.mark.timeout(180)
 def test_multiple_docling_sources() -> None:
     urls: list[Path | str] = [
         "https://lilianweng.github.io/posts/2024-11-28-reward-hacking/",

lib/crewai/tests/llms/bedrock/test_bedrock.py

@@ -383,6 +383,7 @@ def test_bedrock_completion_with_tools():
             assert len(call_kwargs['tools']) > 0
 
 
+@pytest.mark.timeout(180)
 def test_bedrock_raises_error_when_model_not_found(bedrock_mocks):
     """Test that BedrockCompletion raises appropriate error when model not found"""
     from botocore.exceptions import ClientError

lib/crewai/tests/memory/test_dimension_mismatch.py

@@ -0,0 +1,159 @@
+"""Embedding dimension mismatch must fail loudly with migration guidance.
+
+The default embedder changed from text-embedding-3-small (1536 dims) to
+text-embedding-3-large (3072 dims); stores created before the upgrade must
+not silently zero-fill vectors or return empty search results.
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+import pytest
+
+from crewai.memory.storage.backend import EmbeddingDimensionMismatchError
+from crewai.memory.types import MemoryRecord
+
+
+@pytest.fixture
+def lancedb_path(tmp_path: Path) -> Path:
+    return tmp_path / "mem"
+
+
+def _record(dim: int, content: str = "test") -> MemoryRecord:
+    return MemoryRecord(content=content, scope="/foo", embedding=[0.1] * dim)
+
+
+def test_lancedb_save_mismatch_raises(lancedb_path: Path) -> None:
+    from crewai.memory.storage.lancedb_storage import LanceDBStorage
+
+    storage = LanceDBStorage(path=str(lancedb_path), vector_dim=4)
+    storage.save([_record(4)])
+
+    with pytest.raises(EmbeddingDimensionMismatchError) as exc_info:
+        storage.save([_record(8, "new embedder output")])
+
+    message = str(exc_info.value)
+    assert "4-dimensional" in message
+    assert "8-dimensional" in message
+    assert "crewai reset-memories --memory" in message
+    assert "text-embedding-3-small" in message
+
+
+def test_lancedb_mixed_batch_mismatch_raises(lancedb_path: Path) -> None:
+    """A single save() batch with inconsistent dimensions must be rejected."""
+    from crewai.memory.storage.lancedb_storage import LanceDBStorage
+
+    storage = LanceDBStorage(path=str(lancedb_path), vector_dim=4)
+    storage.save([_record(4)])
+
+    with pytest.raises(EmbeddingDimensionMismatchError):
+        storage.save([_record(4), _record(8, "stray dimension")])
+
+
+def test_lancedb_mixed_batch_on_fresh_store_raises(lancedb_path: Path) -> None:
+    from crewai.memory.storage.lancedb_storage import LanceDBStorage
+
+    storage = LanceDBStorage(path=str(lancedb_path))
+    with pytest.raises(EmbeddingDimensionMismatchError):
+        storage.save([_record(4), _record(8)])
+
+
+def test_lancedb_search_mismatch_raises(lancedb_path: Path) -> None:
+    from crewai.memory.storage.lancedb_storage import LanceDBStorage
+
+    storage = LanceDBStorage(path=str(lancedb_path), vector_dim=4)
+    storage.save([_record(4)])
+
+    with pytest.raises(EmbeddingDimensionMismatchError):
+        storage.search([0.1] * 8)
+
+
+def test_lancedb_update_mismatch_raises(lancedb_path: Path) -> None:
+    from crewai.memory.storage.lancedb_storage import LanceDBStorage
+
+    storage = LanceDBStorage(path=str(lancedb_path), vector_dim=4)
+    record = _record(4)
+    storage.save([record])
+
+    stale = MemoryRecord(
+        id=record.id, content="updated", scope="/foo", embedding=[0.1] * 8
+    )
+    with pytest.raises(EmbeddingDimensionMismatchError):
+        storage.update(stale)
+
+
+def test_lancedb_reopened_store_detects_mismatch(lancedb_path: Path) -> None:
+    """The upgrade scenario: an old store reopened with a new embedder."""
+    from crewai.memory.storage.lancedb_storage import LanceDBStorage
+
+    old = LanceDBStorage(path=str(lancedb_path), vector_dim=4)
+    old.save([_record(4)])
+
+    reopened = LanceDBStorage(path=str(lancedb_path))
+    with pytest.raises(EmbeddingDimensionMismatchError):
+        reopened.save([_record(8)])
+    with pytest.raises(EmbeddingDimensionMismatchError):
+        reopened.search([0.1] * 8)
+
+
+def test_lancedb_matching_dim_still_works(lancedb_path: Path) -> None:
+    from crewai.memory.storage.lancedb_storage import LanceDBStorage
+
+    storage = LanceDBStorage(path=str(lancedb_path), vector_dim=4)
+    storage.save([_record(4)])
+    storage.save([_record(4, "second")])
+
+    assert len(storage.search([0.1] * 4, limit=5)) == 2
+
+
+def test_error_is_not_a_runtime_error() -> None:
+    """Background-save plumbing treats RuntimeError as executor shutdown and
+    silently drops the save — the mismatch must not be classified that way."""
+    err = EmbeddingDimensionMismatchError(1536, 3072)
+    assert not isinstance(err, RuntimeError)
+    assert isinstance(err, ValueError)
+
+
+def test_background_save_propagates_dimension_mismatch(tmp_path: Path) -> None:
+    from unittest.mock import MagicMock
+
+    from crewai.memory.unified_memory import Memory
+
+    mem = Memory(
+        storage=str(tmp_path / "db"),
+        llm=MagicMock(),
+        embedder=lambda texts: [[0.1] * 4 for _ in texts],
+    )
+
+    def raise_mismatch(*_args: object, **_kwargs: object) -> None:
+        raise EmbeddingDimensionMismatchError(1536, 3072)
+
+    mem._encode_batch = raise_mismatch  # type: ignore[method-assign]
+
+    with pytest.raises(EmbeddingDimensionMismatchError):
+        mem._background_encode_batch(["content"], None, None, None, None, None, False, None)
+
+
+def test_background_save_still_swallows_shutdown_runtime_error(tmp_path: Path) -> None:
+    from unittest.mock import MagicMock
+
+    from crewai.memory.unified_memory import Memory
+
+    mem = Memory(
+        storage=str(tmp_path / "db"),
+        llm=MagicMock(),
+        embedder=lambda texts: [[0.1] * 4 for _ in texts],
+    )
+
+    def raise_shutdown(*_args: object, **_kwargs: object) -> None:
+        raise RuntimeError("cannot schedule new futures after shutdown")
+
+    mem._encode_batch = raise_shutdown  # type: ignore[method-assign]
+
+    assert (
+        mem._background_encode_batch(
+            ["content"], None, None, None, None, None, False, None
+        )
+        == []
+    )

lib/crewai/tests/memory/test_memory_root_scope.py

@@ -409,6 +409,36 @@ class TestCrewAutoScoping:
         assert crew._memory is not None
         assert hasattr(crew._memory, "root_scope")
         assert crew._memory.root_scope == "/crew/research-crew"
+        assert crew._memory.llm is agent.llm
+
+    def test_crew_memory_true_prefers_chat_llm(self) -> None:
+        """Auto-created crew memory uses chat_llm when configured."""
+        from crewai.agent import Agent
+        from crewai.crew import Crew
+        from crewai.task import Task
+
+        agent = Agent(
+            role="Researcher",
+            goal="Research",
+            backstory="Expert researcher",
+            llm="openai/gpt-4o-mini",
+        )
+        task = Task(
+            description="Do research",
+            expected_output="Report",
+            agent=agent,
+        )
+
+        crew = Crew(
+            name="Research Crew",
+            agents=[agent],
+            tasks=[task],
+            chat_llm="ollama/llama3",
+            memory=True,
+        )
+
+        assert crew._memory is not None
+        assert crew._memory.llm == "ollama/llama3"
 
     def test_crew_memory_instance_preserves_no_root_scope(
         self, tmp_path: Path, mock_embedder: MagicMock

lib/crewai/tests/memory/test_unified_memory.py

@@ -4,6 +4,7 @@ from __future__ import annotations
 
 from datetime import datetime, timedelta
 from pathlib import Path
+import threading
 from unittest.mock import MagicMock
 
 import pytest
@@ -489,8 +490,8 @@ def test_composite_score_reranks_results(
     """Same semantic score: high-importance recent memory ranks first."""
     from crewai.memory.unified_memory import Memory
 
-    # Use same dim as default LanceDB (1536) so storage does not overwrite embedding
-    emb = [0.1] * 1536
+    # Use same dim as default LanceDB (3072) so storage does not overwrite embedding
+    emb = [0.1] * 3072
     mem = Memory(
         storage=str(tmp_path / "rerank_db"),
         llm=MagicMock(),
@@ -974,6 +975,42 @@ def test_recall_drains_pending_writes(tmp_path: Path, mock_embedder: MagicMock)
     assert "Python" in matches[0].record.content
 
 
+def test_drain_writes_reports_background_save_failure_without_raising(
+    tmp_path: Path, mock_embedder: MagicMock
+) -> None:
+    """Background memory failures should be reported without failing cleanup."""
+    from crewai.events.event_bus import crewai_event_bus
+    from crewai.events.types.memory_events import MemorySaveFailedEvent
+    from crewai.memory.unified_memory import Memory
+
+    failure_seen = threading.Event()
+    failures: list[MemorySaveFailedEvent] = []
+    mem = Memory(
+        storage=str(tmp_path / "db"),
+        llm=MagicMock(),
+        embedder=mock_embedder,
+    )
+
+    def fail_save() -> None:
+        raise ValueError("invalid model ID")
+
+    with crewai_event_bus.scoped_handlers():
+
+        @crewai_event_bus.on(MemorySaveFailedEvent)
+        def on_memory_save_failed(_source, event):
+            failures.append(event)
+            failure_seen.set()
+
+        mem._submit_save(fail_save)
+        mem.drain_writes()
+
+        assert failure_seen.wait(timeout=2)
+
+    assert failures
+    assert failures[0].value == "background save"
+    assert failures[0].error == "invalid model ID"
+
+
 def test_close_drains_and_shuts_down(tmp_path: Path, mock_embedder: MagicMock) -> None:
     """close() should drain pending saves and shut down the pool."""
     from crewai.memory.unified_memory import Memory

lib/crewai/tests/project/test_crew_loader.py

@@ -0,0 +1,424 @@
+"""Tests for crewai.project.crew_loader."""
+
+from __future__ import annotations
+
+import json
+from pathlib import Path
+
+import pytest
+
+from crewai.llms.base_llm import BaseLLM
+from crewai.project.json_loader import JSONProjectError, JSONProjectValidationError
+from crewai.project.crew_loader import load_crew
+
+
+def _write_agent(agents_dir: Path, name: str, **overrides) -> Path:
+    defn = {
+        "role": f"{name} role",
+        "goal": f"{name} goal",
+        "backstory": f"{name} backstory",
+    }
+    defn.update(overrides)
+    f = agents_dir / f"{name}.jsonc"
+    f.write_text(json.dumps(defn))
+    return f
+
+
+def _write_crew(project_dir: Path, crew_def: dict) -> Path:
+    f = project_dir / "crew.jsonc"
+    f.write_text(json.dumps(crew_def))
+    return f
+
+
+class TestLoadCrew:
+    def test_minimal_crew(self, tmp_path: Path):
+        agents_dir = tmp_path / "agents"
+        agents_dir.mkdir()
+        _write_agent(agents_dir, "researcher")
+
+        crew_def = {
+            "name": "test_crew",
+            "agents": ["researcher"],
+            "tasks": [
+                {
+                    "name": "research",
+                    "description": "Do research",
+                    "expected_output": "Research findings",
+                    "agent": "researcher",
+                }
+            ],
+        }
+        crew_file = _write_crew(tmp_path, crew_def)
+
+        crew, inputs = load_crew(crew_file)
+        assert crew.name == "test_crew"
+        assert len(crew.agents) == 1
+        assert len(crew.tasks) == 1
+        assert crew.tasks[0].description == "Do research"
+        assert inputs == {}
+
+    def test_crew_with_default_inputs(self, tmp_path: Path):
+        agents_dir = tmp_path / "agents"
+        agents_dir.mkdir()
+        _write_agent(agents_dir, "researcher")
+
+        crew_def = {
+            "name": "test_crew",
+            "agents": ["researcher"],
+            "tasks": [
+                {
+                    "name": "research",
+                    "description": "Research {topic}",
+                    "expected_output": "Findings about {topic}",
+                    "agent": "researcher",
+                }
+            ],
+            "inputs": {"topic": "AI"},
+        }
+        crew_file = _write_crew(tmp_path, crew_def)
+
+        crew, inputs = load_crew(crew_file)
+        assert inputs == {"topic": "AI"}
+
+    def test_crew_with_multiple_agents(self, tmp_path: Path):
+        agents_dir = tmp_path / "agents"
+        agents_dir.mkdir()
+        _write_agent(agents_dir, "researcher")
+        _write_agent(agents_dir, "writer")
+
+        crew_def = {
+            "name": "multi_crew",
+            "agents": ["researcher", "writer"],
+            "tasks": [
+                {
+                    "name": "research",
+                    "description": "Do research",
+                    "expected_output": "Findings",
+                    "agent": "researcher",
+                },
+                {
+                    "name": "write",
+                    "description": "Write report",
+                    "expected_output": "Report",
+                    "agent": "writer",
+                    "context": ["research"],
+                },
+            ],
+        }
+        crew_file = _write_crew(tmp_path, crew_def)
+
+        crew, _ = load_crew(crew_file)
+        assert len(crew.agents) == 2
+        assert len(crew.tasks) == 2
+        # Second task should have context referencing first task
+        assert crew.tasks[1].context is not None
+        assert len(crew.tasks[1].context) == 1
+
+    def test_crew_hierarchical_process(self, tmp_path: Path):
+        agents_dir = tmp_path / "agents"
+        agents_dir.mkdir()
+        _write_agent(agents_dir, "worker")
+
+        crew_def = {
+            "name": "hier_crew",
+            "agents": ["worker"],
+            "tasks": [
+                {
+                    "name": "work",
+                    "description": "Do work",
+                    "expected_output": "Work done",
+                    "agent": "worker",
+                }
+            ],
+            "process": "hierarchical",
+            "manager_llm": "openai/gpt-4o",
+        }
+        crew_file = _write_crew(tmp_path, crew_def)
+
+        crew, _ = load_crew(crew_file)
+        from crewai import Process
+        assert crew.process == Process.hierarchical
+
+    def test_crew_accepts_llm_config_objects(self, tmp_path: Path):
+        agents_dir = tmp_path / "agents"
+        agents_dir.mkdir()
+        _write_agent(agents_dir, "worker", llm="ollama/llama3")
+
+        crew_def = {
+            "name": "llm_config_crew",
+            "agents": ["worker"],
+            "tasks": [
+                {
+                    "name": "work",
+                    "description": "Do work",
+                    "expected_output": "Work done",
+                    "agent": "worker",
+                }
+            ],
+            "process": "hierarchical",
+            "manager_llm": {
+                "model": "llama3",
+                "provider": "ollama",
+                "base_url": "http://localhost:11434",
+            },
+            "planning_llm": {
+                "model": "deepseek-chat",
+                "provider": "deepseek",
+                "api_key": "test-key",
+            },
+            "chat_llm": {
+                "model": "openrouter/anthropic/claude-3-opus",
+                "api_key": "test-key",
+            },
+        }
+        crew_file = _write_crew(tmp_path, crew_def)
+
+        crew, _ = load_crew(crew_file)
+
+        assert isinstance(crew.manager_llm, BaseLLM)
+        assert crew.manager_llm.model == "llama3"
+        assert crew.manager_llm.provider == "ollama"
+        assert crew.manager_llm.base_url == "http://localhost:11434/v1"
+        assert isinstance(crew.planning_llm, BaseLLM)
+        assert crew.planning_llm.model == "deepseek-chat"
+        assert crew.planning_llm.provider == "deepseek"
+        assert isinstance(crew.chat_llm, BaseLLM)
+        assert crew.chat_llm.model == "anthropic/claude-3-opus"
+        assert crew.chat_llm.provider == "openrouter"
+
+    def test_crew_accepts_public_crew_config_fields(self, tmp_path: Path):
+        agents_dir = tmp_path / "agents"
+        agents_dir.mkdir()
+        _write_agent(agents_dir, "worker")
+
+        crew_def = {
+            "name": "config_crew",
+            "agents": ["worker"],
+            "tasks": [
+                {
+                    "name": "work",
+                    "description": "Do work",
+                    "expected_output": "Work done",
+                    "agent": "worker",
+                }
+            ],
+            "cache": False,
+            "max_rpm": 12,
+            "planning": True,
+            "planning_llm": "openai/gpt-4o-mini",
+            "share_crew": False,
+            "output_log_file": "crew.log",
+            "tracing": False,
+        }
+        crew_file = _write_crew(tmp_path, crew_def)
+
+        crew, _ = load_crew(crew_file)
+        assert crew.cache is False
+        assert crew.max_rpm == 12
+        assert crew.planning is True
+        assert crew.planning_llm == "openai/gpt-4o-mini"
+        assert crew.output_log_file == "crew.log"
+        assert crew.tracing is False
+
+    def test_crew_with_output_file(self, tmp_path: Path):
+        agents_dir = tmp_path / "agents"
+        agents_dir.mkdir()
+        _write_agent(agents_dir, "writer")
+
+        crew_def = {
+            "name": "output_crew",
+            "agents": ["writer"],
+            "tasks": [
+                {
+                    "name": "write",
+                    "description": "Write something",
+                    "expected_output": "Written content",
+                    "agent": "writer",
+                    "output_file": "output.md",
+                }
+            ],
+        }
+        crew_file = _write_crew(tmp_path, crew_def)
+
+        crew, _ = load_crew(crew_file)
+        assert crew.tasks[0].output_file == "output.md"
+
+    def test_task_accepts_public_task_config_fields(self, tmp_path: Path):
+        agents_dir = tmp_path / "agents"
+        agents_dir.mkdir()
+        _write_agent(agents_dir, "writer")
+
+        schema = {
+            "title": "ReportOutput",
+            "type": "object",
+            "properties": {
+                "summary": {"type": "string"},
+            },
+            "required": ["summary"],
+        }
+        crew_def = {
+            "name": "task_config_crew",
+            "agents": ["writer"],
+            "tasks": [
+                {
+                    "name": "write",
+                    "description": "Write something",
+                    "expected_output": "Written content",
+                    "agent": "writer",
+                    "output_json": schema,
+                    "response_model": schema,
+                    "create_directory": False,
+                    "human_input": True,
+                    "markdown": True,
+                    "guardrail": "Return a summary field.",
+                    "guardrail_max_retries": 1,
+                    "allow_crewai_trigger_context": False,
+                }
+            ],
+        }
+        crew_file = _write_crew(tmp_path, crew_def)
+
+        crew, _ = load_crew(crew_file)
+        task = crew.tasks[0]
+        assert task.output_json is not None
+        assert "summary" in task.output_json.model_fields
+        assert task.response_model is not None
+        assert task.create_directory is False
+        assert task.human_input is True
+        assert task.markdown is True
+        assert task.guardrail == "Return a summary field."
+        assert task.allow_crewai_trigger_context is False
+
+    def test_missing_agent_file_raises(self, tmp_path: Path):
+        agents_dir = tmp_path / "agents"
+        agents_dir.mkdir()
+
+        crew_def = {
+            "name": "broken_crew",
+            "agents": ["nonexistent"],
+            "tasks": [],
+        }
+        crew_file = _write_crew(tmp_path, crew_def)
+
+        with pytest.raises(FileNotFoundError, match="nonexistent"):
+            load_crew(crew_file)
+
+    def test_task_references_unknown_agent_raises(self, tmp_path: Path):
+        agents_dir = tmp_path / "agents"
+        agents_dir.mkdir()
+        _write_agent(agents_dir, "researcher")
+
+        crew_def = {
+            "name": "bad_ref_crew",
+            "agents": ["researcher"],
+            "tasks": [
+                {
+                    "name": "task1",
+                    "description": "Do something",
+                    "expected_output": "Something",
+                    "agent": "unknown_agent",
+                }
+            ],
+        }
+        crew_file = _write_crew(tmp_path, crew_def)
+
+        with pytest.raises(JSONProjectError, match="unknown_agent"):
+            load_crew(crew_file)
+
+    def test_task_context_order_dependency(self, tmp_path: Path):
+        agents_dir = tmp_path / "agents"
+        agents_dir.mkdir()
+        _write_agent(agents_dir, "worker")
+
+        crew_def = {
+            "name": "order_crew",
+            "agents": ["worker"],
+            "tasks": [
+                {
+                    "name": "task2",
+                    "description": "Second task",
+                    "expected_output": "Output",
+                    "agent": "worker",
+                    "context": ["task1"],
+                },
+                {
+                    "name": "task1",
+                    "description": "First task",
+                    "expected_output": "Output",
+                    "agent": "worker",
+                },
+            ],
+        }
+        crew_file = _write_crew(tmp_path, crew_def)
+
+        with pytest.raises(JSONProjectError, match="task1"):
+            load_crew(crew_file)
+
+    def test_runtime_fields_are_rejected(self, tmp_path: Path):
+        agents_dir = tmp_path / "agents"
+        agents_dir.mkdir()
+        _write_agent(agents_dir, "worker")
+
+        crew_def = {
+            "name": "bad_runtime_crew",
+            "id": "00000000-0000-4000-8000-000000000000",
+            "agents": ["worker"],
+            "tasks": [
+                {
+                    "name": "work",
+                    "description": "Work",
+                    "expected_output": "Done",
+                    "agent": "worker",
+                }
+            ],
+        }
+        crew_file = _write_crew(tmp_path, crew_def)
+
+        with pytest.raises(JSONProjectValidationError, match="runtime-only"):
+            load_crew(crew_file)
+
+    def test_custom_agents_dir(self, tmp_path: Path):
+        custom_dir = tmp_path / "my_agents"
+        custom_dir.mkdir()
+        _write_agent(custom_dir, "analyst")
+
+        crew_def = {
+            "name": "custom_dir_crew",
+            "agents": ["analyst"],
+            "tasks": [
+                {
+                    "name": "analyze",
+                    "description": "Analyze data",
+                    "expected_output": "Analysis",
+                    "agent": "analyst",
+                }
+            ],
+        }
+        crew_file = _write_crew(tmp_path, crew_def)
+
+        crew, _ = load_crew(crew_file, agents_dir=custom_dir)
+        assert len(crew.agents) == 1
+
+    def test_crew_verbose_and_memory_flags(self, tmp_path: Path):
+        agents_dir = tmp_path / "agents"
+        agents_dir.mkdir()
+        _write_agent(agents_dir, "worker")
+
+        crew_def = {
+            "name": "flags_crew",
+            "agents": ["worker"],
+            "tasks": [
+                {
+                    "name": "work",
+                    "description": "Work",
+                    "expected_output": "Done",
+                    "agent": "worker",
+                }
+            ],
+            "verbose": True,
+            "memory": True,
+        }
+        crew_file = _write_crew(tmp_path, crew_def)
+
+        crew, _ = load_crew(crew_file)
+        assert crew.verbose is True

lib/crewai/tests/project/test_json_loader.py

@@ -0,0 +1,465 @@
+"""Tests for crewai.project.json_loader."""
+
+from __future__ import annotations
+
+import json
+from pathlib import Path
+import sys
+
+import pytest
+
+from crewai.llms.base_llm import BaseLLM
+from crewai.project.json_loader import (
+    JSONProjectValidationError,
+    find_json_project_file,
+    load_agent,
+    strip_jsonc_comments,
+)
+
+
+class TestStripJsoncComments:
+    def test_strips_single_line_comments(self):
+        text = '{\n  "key": "value" // this is a comment\n}'
+        result = strip_jsonc_comments(text)
+        data = json.loads(result)
+        assert data["key"] == "value"
+
+    def test_strips_block_comments(self):
+        text = '{\n  /* block comment */\n  "key": "value"\n}'
+        result = strip_jsonc_comments(text)
+        data = json.loads(result)
+        assert data["key"] == "value"
+
+    def test_preserves_urls_with_double_slash(self):
+        text = '{\n  "url": "https://example.com"\n}'
+        result = strip_jsonc_comments(text)
+        data = json.loads(result)
+        assert data["url"] == "https://example.com"
+
+    def test_preserves_comment_markers_inside_strings(self):
+        text = """{
+  "url": "https://example.com/a//b",
+  "pattern": "keep /* this */ text",
+  "text": "value // not a comment",
+}"""
+        result = strip_jsonc_comments(text)
+        data = json.loads(result)
+        assert data["url"] == "https://example.com/a//b"
+        assert data["pattern"] == "keep /* this */ text"
+        assert data["text"] == "value // not a comment"
+
+    def test_removes_trailing_commas(self):
+        text = '{\n  "a": 1,\n  "b": 2,\n}'
+        result = strip_jsonc_comments(text)
+        data = json.loads(result)
+        assert data == {"a": 1, "b": 2}
+
+    def test_removes_trailing_commas_in_arrays(self):
+        text = '{"arr": [1, 2, 3,]}'
+        result = strip_jsonc_comments(text)
+        data = json.loads(result)
+        assert data["arr"] == [1, 2, 3]
+
+    def test_plain_json_unchanged(self):
+        text = '{"key": "value"}'
+        result = strip_jsonc_comments(text)
+        assert json.loads(result) == {"key": "value"}
+
+
+def test_find_json_project_file_prefers_jsonc(tmp_path: Path):
+    (tmp_path / "agent.json").write_text("{}")
+    jsonc_path = tmp_path / "agent.jsonc"
+    jsonc_path.write_text("{}")
+
+    assert find_json_project_file(tmp_path, "agent") == jsonc_path
+
+
+class TestLoadAgent:
+    def test_load_minimal_agent(self, tmp_path: Path):
+        agent_def = {
+            "role": "Researcher",
+            "goal": "Find information",
+            "backstory": "Expert researcher.",
+        }
+        agent_file = tmp_path / "agent.json"
+        agent_file.write_text(json.dumps(agent_def))
+
+        agent = load_agent(agent_file)
+        assert agent.role == "Researcher"
+        assert agent.goal == "Find information"
+        assert agent.backstory == "Expert researcher."
+
+    def test_load_agent_with_llm(self, tmp_path: Path):
+        agent_def = {
+            "role": "Coder",
+            "goal": "Write code",
+            "backstory": "Expert coder.",
+            "llm": "openai/gpt-4o",
+        }
+        agent_file = tmp_path / "agent.json"
+        agent_file.write_text(json.dumps(agent_def))
+
+        agent = load_agent(agent_file)
+        assert agent.role == "Coder"
+
+    def test_load_agent_with_llm_config_object(self, tmp_path: Path):
+        agent_def = {
+            "role": "Coder",
+            "goal": "Write code",
+            "backstory": "Expert coder.",
+            "llm": {
+                "model": "llama3",
+                "provider": "ollama",
+                "temperature": 0.2,
+                "base_url": "http://localhost:11434",
+            },
+        }
+        agent_file = tmp_path / "agent.json"
+        agent_file.write_text(json.dumps(agent_def))
+
+        agent = load_agent(agent_file)
+
+        assert isinstance(agent.llm, BaseLLM)
+        assert agent.llm.model == "llama3"
+        assert agent.llm.provider == "ollama"
+        assert agent.llm.temperature == 0.2
+        assert agent.llm.base_url == "http://localhost:11434/v1"
+
+    def test_load_agent_with_planning_config_llm_object(self, tmp_path: Path):
+        agent_def = {
+            "role": "Planner",
+            "goal": "Plan work",
+            "backstory": "Expert planner.",
+            "llm": "ollama/llama3",
+            "planning_config": {
+                "reasoning_effort": "high",
+                "llm": {
+                    "model": "deepseek-chat",
+                    "provider": "deepseek",
+                    "api_key": "test-key",
+                },
+            },
+        }
+        agent_file = tmp_path / "agent.json"
+        agent_file.write_text(json.dumps(agent_def))
+
+        agent = load_agent(agent_file)
+
+        assert agent.planning_config is not None
+        assert isinstance(agent.planning_config.llm, BaseLLM)
+        assert agent.planning_config.llm.model == "deepseek-chat"
+        assert agent.planning_config.llm.provider == "deepseek"
+        assert agent.planning_config.llm.api_key == "test-key"
+
+    def test_load_agent_with_settings_block(self, tmp_path: Path):
+        agent_def = {
+            "role": "Analyst",
+            "goal": "Analyze data",
+            "backstory": "Data expert.",
+            "settings": {
+                "verbose": True,
+                "allow_delegation": True,
+                "max_iter": 10,
+                "cache": False,
+            },
+        }
+        agent_file = tmp_path / "agent.json"
+        agent_file.write_text(json.dumps(agent_def))
+
+        agent = load_agent(agent_file)
+        assert agent.role == "Analyst"
+        assert agent.verbose is True
+        assert agent.allow_delegation is True
+        assert agent.max_iter == 10
+        assert agent.cache is False
+
+    def test_load_agent_with_top_level_settings(self, tmp_path: Path):
+        agent_def = {
+            "role": "Analyst",
+            "goal": "Analyze data",
+            "backstory": "Data expert.",
+            "verbose": True,
+            "max_iter": 15,
+        }
+        agent_file = tmp_path / "agent.json"
+        agent_file.write_text(json.dumps(agent_def))
+
+        agent = load_agent(agent_file)
+        assert agent.verbose is True
+        assert agent.max_iter == 15
+
+    def test_load_agent_accepts_public_agent_config_fields(self, tmp_path: Path):
+        agent_def = {
+            "role": "Analyst",
+            "goal": "Analyze data",
+            "backstory": "Data expert.",
+            "max_execution_time": 30,
+            "use_system_prompt": False,
+            "system_template": "system: {{ .System }}",
+            "prompt_template": "prompt: {{ .Prompt }}",
+            "response_template": "response: {{ .Response }}",
+            "inject_date": True,
+            "date_format": "%Y",
+            "guardrail": "Only return concise answers.",
+            "guardrail_max_retries": 1,
+            "security_config": {"fingerprint": "agent-seed"},
+        }
+        agent_file = tmp_path / "agent.json"
+        agent_file.write_text(json.dumps(agent_def))
+
+        agent = load_agent(agent_file)
+        assert agent.max_execution_time == 30
+        assert agent.use_system_prompt is False
+        assert agent.system_template == "system: {{ .System }}"
+        assert agent.inject_date is True
+        assert agent.guardrail == "Only return concise answers."
+
+    def test_load_agent_accepts_serialized_tool_dict(
+        self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+    ):
+        module = tmp_path / "test_tools.py"
+        module.write_text(
+            "from crewai.tools.base_tool import BaseTool\n"
+            "class EchoTool(BaseTool):\n"
+            "    name: str = 'echo'\n"
+            "    description: str = 'Echo input'\n"
+            "    def _run(self, value: str = '') -> str:\n"
+            "        return value\n"
+        )
+        monkeypatch.syspath_prepend(str(tmp_path))
+        sys.modules.pop("test_tools", None)
+
+        agent_def = {
+            "role": "Tool User",
+            "goal": "Use tools",
+            "backstory": "Tool expert.",
+            "tools": [
+                {
+                    "tool_type": "test_tools.EchoTool",
+                    "name": "echo",
+                    "description": "Echo input",
+                }
+            ],
+        }
+        agent_file = tmp_path / "agent.json"
+        agent_file.write_text(json.dumps(agent_def))
+
+        agent = load_agent(agent_file)
+        assert len(agent.tools or []) == 1
+        assert agent.tools[0].name == "echo"
+
+    def test_load_agent_rejects_runtime_fields(self, tmp_path: Path):
+        agent_def = {
+            "id": "00000000-0000-4000-8000-000000000000",
+            "role": "Analyst",
+            "goal": "Analyze data",
+            "backstory": "Data expert.",
+        }
+        agent_file = tmp_path / "agent.json"
+        agent_file.write_text(json.dumps(agent_def))
+
+        with pytest.raises(JSONProjectValidationError, match="runtime-only"):
+            load_agent(agent_file)
+
+    def test_settings_block_takes_precedence(self, tmp_path: Path):
+        agent_def = {
+            "role": "Analyst",
+            "goal": "Analyze data",
+            "backstory": "Data expert.",
+            "verbose": False,
+            "settings": {
+                "verbose": True,
+            },
+        }
+        agent_file = tmp_path / "agent.json"
+        agent_file.write_text(json.dumps(agent_def))
+
+        agent = load_agent(agent_file)
+        assert agent.verbose is True
+
+    def test_load_agent_from_jsonc(self, tmp_path: Path):
+        jsonc_content = """{
+  // This is a JSONC file with comments
+  "role": "Writer",
+  "goal": "Write articles",
+  "backstory": "Expert writer.",
+  /* multi-line
+     comment */
+}"""
+        agent_file = tmp_path / "agent.jsonc"
+        agent_file.write_text(jsonc_content)
+
+        agent = load_agent(agent_file)
+        assert agent.role == "Writer"
+
+    def test_load_agent_missing_required_fields(self, tmp_path: Path):
+        agent_def = {"role": "Incomplete"}
+        agent_file = tmp_path / "agent.json"
+        agent_file.write_text(json.dumps(agent_def))
+
+        with pytest.raises(Exception):
+            load_agent(agent_file)
+
+    def test_load_agent_file_not_found(self):
+        with pytest.raises(FileNotFoundError):
+            load_agent(Path("/nonexistent/agent.json"))
+
+
+class TestResolveTools:
+    def test_unknown_tool_raises_with_guidance(self):
+        from crewai.project.json_loader import JSONProjectError, _resolve_tools
+
+        with pytest.raises(JSONProjectError, match="Unknown tool 'NotARealToolXYZ'"):
+            _resolve_tools(["NotARealToolXYZ"])
+
+    def test_missing_custom_tool_raises(self, tmp_path, monkeypatch):
+        from crewai.project.json_loader import JSONProjectError, _resolve_tools
+
+        monkeypatch.chdir(tmp_path)
+        with pytest.raises(JSONProjectError, match="custom:missing"):
+            _resolve_tools(["custom:missing"])
+
+    def test_custom_tool_without_basetool_subclass_raises(self, tmp_path, monkeypatch):
+        from crewai.project.json_loader import JSONProjectError, _resolve_tools
+
+        monkeypatch.chdir(tmp_path)
+        tools_dir = tmp_path / "tools"
+        tools_dir.mkdir()
+        (tools_dir / "empty.py").write_text("x = 1\n")
+
+        with pytest.raises(JSONProjectError, match="No BaseTool subclass"):
+            _resolve_tools(["custom:empty"])
+
+    def test_custom_tool_resolves(self, tmp_path, monkeypatch):
+        from crewai.project.json_loader import _resolve_tools
+
+        monkeypatch.chdir(tmp_path)
+        tools_dir = tmp_path / "tools"
+        tools_dir.mkdir()
+        (tools_dir / "echo.py").write_text(
+            "from crewai.tools.base_tool import BaseTool\n"
+            "\n"
+            "class EchoTool(BaseTool):\n"
+            "    name: str = 'echo'\n"
+            "    description: str = 'echo input'\n"
+            "\n"
+            "    def _run(self, text: str) -> str:\n"
+            "        return text\n"
+        )
+
+        tools = _resolve_tools(["custom:echo"])
+
+        assert len(tools) == 1
+        assert tools[0].name == "echo"
+
+    def test_serialized_tool_dicts_pass_through(self):
+        from crewai.project.json_loader import _resolve_tools
+
+        spec = {"tool_type": "some.module.Tool"}
+        assert _resolve_tools([spec]) == [spec]
+
+
+class TestValidationDoesNotExecuteTools:
+    def _write_project(self, root, tool_line='"custom:landmine"'):
+        agents_dir = root / "agents"
+        agents_dir.mkdir()
+        (agents_dir / "worker.jsonc").write_text(
+            "{\n"
+            '  "role": "Worker",\n'
+            '  "goal": "Work",\n'
+            '  "backstory": "Works hard",\n'
+            f'  "tools": [{tool_line}]\n'
+            "}\n"
+        )
+        crew_path = root / "crew.jsonc"
+        crew_path.write_text(
+            "{\n"
+            '  "agents": ["worker"],\n'
+            '  "tasks": [\n'
+            '    {"name": "t1", "description": "Do work", '
+            '"expected_output": "Done", "agent": "worker"}\n'
+            "  ]\n"
+            "}\n"
+        )
+        return crew_path
+
+    def test_validate_does_not_execute_custom_tool_code(self, tmp_path):
+        from crewai.project.json_loader import validate_crew_project
+
+        sentinel = tmp_path / "executed.txt"
+        tools_dir = tmp_path / "tools"
+        tools_dir.mkdir()
+        (tools_dir / "landmine.py").write_text(
+            f"open({str(sentinel)!r}, 'w').write('boom')\n"
+        )
+        crew_path = self._write_project(tmp_path)
+
+        project = validate_crew_project(crew_path, tmp_path / "agents")
+
+        assert not sentinel.exists(), "validation must not execute tools/<name>.py"
+        assert project.agent_names == ["worker"]
+
+    def test_validate_reports_missing_custom_tool_file(self, tmp_path):
+        from crewai.project.json_loader import (
+            JSONProjectValidationError,
+            validate_crew_project,
+        )
+
+        crew_path = self._write_project(tmp_path)
+
+        with pytest.raises(JSONProjectValidationError) as exc_info:
+            validate_crew_project(crew_path, tmp_path / "agents")
+
+        assert "custom:landmine" in str(exc_info.value)
+        assert "not found" in str(exc_info.value)
+
+    def test_validate_reports_path_escaping_custom_tool(self, tmp_path):
+        from crewai.project.json_loader import (
+            JSONProjectValidationError,
+            validate_crew_project,
+        )
+
+        crew_path = self._write_project(tmp_path, tool_line='"custom:../evil"')
+
+        with pytest.raises(JSONProjectValidationError) as exc_info:
+            validate_crew_project(crew_path, tmp_path / "agents")
+
+        assert "Invalid custom tool name" in str(exc_info.value)
+
+
+class TestCustomToolPathSafety:
+    @pytest.mark.parametrize(
+        "bad_name",
+        ["../evil", "..", "sub/inner", "/etc/passwd", "a-b", "", "name.py"],
+    )
+    def test_unsafe_names_rejected_at_runtime(self, bad_name, tmp_path, monkeypatch):
+        from crewai.project.json_loader import JSONProjectError, _resolve_tools
+
+        monkeypatch.chdir(tmp_path)
+        with pytest.raises(JSONProjectError, match="Invalid custom tool name"):
+            _resolve_tools([f"custom:{bad_name}"])
+
+    def test_resolves_relative_to_project_root_not_cwd(self, tmp_path, monkeypatch):
+        from crewai.project.json_loader import _resolve_tools
+
+        project_root = tmp_path / "project"
+        tools_dir = project_root / "tools"
+        tools_dir.mkdir(parents=True)
+        (tools_dir / "echo.py").write_text(
+            "from crewai.tools.base_tool import BaseTool\n"
+            "\n"
+            "class EchoTool(BaseTool):\n"
+            "    name: str = 'echo'\n"
+            "    description: str = 'echo input'\n"
+            "\n"
+            "    def _run(self, text: str) -> str:\n"
+            "        return text\n"
+        )
+        elsewhere = tmp_path / "elsewhere"
+        elsewhere.mkdir()
+        monkeypatch.chdir(elsewhere)
+
+        tools = _resolve_tools(["custom:echo"], project_root=project_root)
+
+        assert len(tools) == 1
+        assert tools[0].name == "echo"

lib/crewai/tests/rag/embeddings/test_backward_compatibility.py

@@ -46,6 +46,30 @@ class TestModelKeyBackwardCompatibility:
         )
         assert provider.model_name == "text-embedding-3-large"
 
+    def test_openai_provider_ignores_chat_model_env(self, monkeypatch):
+        """Test OpenAI embeddings don't inherit the chat model env var."""
+        monkeypatch.setenv("OPENAI_MODEL_NAME", "gpt-5.5")
+        monkeypatch.setenv("MODEL", "gpt-5.5")
+        monkeypatch.delenv("EMBEDDINGS_OPENAI_MODEL_NAME", raising=False)
+
+        provider = OpenAIProvider(api_key="test-key")
+
+        assert provider.model_name == "text-embedding-3-large"
+
+    def test_azure_provider_ignores_openai_chat_model_env(self, monkeypatch):
+        """Test Azure embeddings don't inherit the OpenAI chat model env var."""
+        monkeypatch.setenv("OPENAI_MODEL_NAME", "gpt-5.5")
+        monkeypatch.setenv("MODEL", "gpt-5.5")
+        monkeypatch.delenv("EMBEDDINGS_OPENAI_MODEL_NAME", raising=False)
+        monkeypatch.delenv("AZURE_OPENAI_MODEL_NAME", raising=False)
+
+        provider = AzureProvider(
+            api_key="test-key",
+            deployment_id="test-deployment",
+        )
+
+        assert provider.model_name == "text-embedding-3-large"
+
     def test_cohere_provider_accepts_model_key(self):
         """Test Cohere provider accepts 'model' as alias for 'model_name'."""
         provider = CohereProvider(
@@ -361,4 +385,4 @@ class TestLegacyConfigurationFormats:
             deployment_id="test-deployment",
             model="text-embedding-3-large",
         )
-        assert provider.model_name == "text-embedding-3-large"
+        assert provider.model_name == "text-embedding-3-large"

lib/crewai/tests/telemetry/test_telemetry.py

@@ -1,11 +1,10 @@
 import os
 import threading
-from unittest.mock import patch
+from unittest.mock import Mock, patch
 
 import pytest
 from crewai import Agent, Crew, Task
 from crewai.telemetry import Telemetry
-from opentelemetry import trace
 from opentelemetry.sdk.trace import TracerProvider
 
 
@@ -71,6 +70,32 @@ def test_set_tracer_skips_when_provider_already_configured():
     assert telemetry.trace_set is True
 
 
+def test_flow_execution_span_records_crewai_version():
+    tracer = Mock()
+    span = Mock()
+    tracer.start_span.return_value = span
+
+    with (
+        patch.dict(
+            os.environ,
+            {
+                "CREWAI_DISABLE_TELEMETRY": "false",
+                "CREWAI_DISABLE_TRACKING": "false",
+                "OTEL_SDK_DISABLED": "false",
+            },
+        ),
+        patch("crewai.telemetry.telemetry.TracerProvider"),
+        patch("crewai.telemetry.telemetry.trace.get_tracer", return_value=tracer),
+        patch("crewai.telemetry.telemetry.version", return_value="9.9.9"),
+    ):
+        telemetry = Telemetry()
+        telemetry.flow_execution_span("ResearchFlow", ["start", "finish"])
+
+    tracer.start_span.assert_called_once_with("Flow Execution")
+    span.set_attribute.assert_any_call("crewai_version", "9.9.9")
+    span.set_attribute.assert_any_call("flow_name", "ResearchFlow")
+
+
 @patch("crewai.telemetry.telemetry.logger.error")
 @patch(
     "opentelemetry.exporter.otlp.proto.http.trace_exporter.OTLPSpanExporter.export",
@@ -85,9 +110,8 @@ def test_telemetry_fails_due_connect_timeout(export_mock, logger_mock):
         os.environ, {"CREWAI_DISABLE_TELEMETRY": "false", "OTEL_SDK_DISABLED": "false"}
     ):
         telemetry = Telemetry()
-        telemetry.set_tracer()
 
-        tracer = trace.get_tracer(__name__)
+        tracer = telemetry.provider.get_tracer(__name__)
         with tracer.start_as_current_span("test-span"):
             agent = Agent(
                 role="agent",
@@ -103,7 +127,7 @@ def test_telemetry_fails_due_connect_timeout(export_mock, logger_mock):
             crew = Crew(agents=[agent], tasks=[task], name="TestCrew")
             crew.kickoff()
 
-        trace.get_tracer_provider().force_flush()
+        telemetry.provider.force_flush()
 
     assert export_mock.called
     assert logger_mock.call_count == export_mock.call_count

lib/crewai/tests/test_async_human_feedback.py

@@ -1066,7 +1066,11 @@ class TestLLMObjectPreservedInContext:
             persistence = SQLiteFlowPersistence(db_path)
 
             from crewai.llm import LLM
-            mock_llm_obj = LLM(model="gemini-2.0-flash", provider="gemini")
+            mock_llm_obj = LLM(
+                model="llama3",
+                provider="ollama",
+                base_url="http://localhost:11434",
+            )
 
             class PausingProvider:
                 def __init__(self, persistence: SQLiteFlowPersistence):
@@ -1116,19 +1120,19 @@ class TestLLMObjectPreservedInContext:
 
             assert provider.captured_context is not None
             assert isinstance(provider.captured_context.llm, dict)
-            assert provider.captured_context.llm["model"] == "gemini/gemini-2.0-flash"
+            assert provider.captured_context.llm["model"] == "ollama/llama3"
 
             flow_id = result.context.flow_id
             loaded = persistence.load_pending_feedback(flow_id)
             assert loaded is not None
             _, loaded_context = loaded
             assert isinstance(loaded_context.llm, dict)
-            assert loaded_context.llm["model"] == "gemini/gemini-2.0-flash"
+            assert loaded_context.llm["model"] == "ollama/llama3"
 
             flow2 = TestFlow.from_pending(flow_id, persistence)
             assert flow2._pending_feedback_context is not None
             assert isinstance(flow2._pending_feedback_context.llm, dict)
-            assert flow2._pending_feedback_context.llm["model"] == "gemini/gemini-2.0-flash"
+            assert flow2._pending_feedback_context.llm["model"] == "ollama/llama3"
 
             with patch.object(flow2, "_collapse_to_outcome", return_value="approved") as mock_collapse:
                 flow2.resume("this looks good, proceed!")
@@ -1140,7 +1144,7 @@ class TestLLMObjectPreservedInContext:
             assert call_kwargs.kwargs["outcomes"] == ["needs_changes", "approved"]
             # LLM should be a live object (from _human_feedback_llm) or reconstructed, not None
             assert call_kwargs.kwargs["llm"] is not None
-            assert getattr(call_kwargs.kwargs["llm"], "model", None) == "gemini-2.0-flash"
+            assert getattr(call_kwargs.kwargs["llm"], "model", None) == "llama3"
             assert flow2.last_human_feedback.outcome == "approved"
             assert flow2.result_path == "approved"
 
@@ -1172,20 +1176,24 @@ class TestLLMObjectPreservedInContext:
         from crewai.flow.human_feedback import _serialize_llm_for_context
         from crewai.llm import LLM
 
-        llm = LLM(model="gemini-2.0-flash", provider="gemini")
+        llm = LLM(
+            model="llama3",
+            provider="ollama",
+            base_url="http://localhost:11434",
+        )
         result = _serialize_llm_for_context(llm)
         assert isinstance(result, dict)
-        assert result["model"] == "gemini/gemini-2.0-flash"
+        assert result["model"] == "ollama/llama3"
 
     def test_provider_prefix_not_doubled_when_already_present(self) -> None:
         """Test that provider prefix is not added when model already has a slash."""
         from crewai.flow.human_feedback import _serialize_llm_for_context
         from crewai.llm import LLM
 
-        llm = LLM(model="gemini/gemini-2.0-flash")
+        llm = LLM(model="ollama/llama3", base_url="http://localhost:11434")
         result = _serialize_llm_for_context(llm)
         assert isinstance(result, dict)
-        assert result["model"] == "gemini/gemini-2.0-flash"
+        assert result["model"] == "ollama/llama3"
 
     def test_no_provider_attr_falls_back_to_bare_model(self) -> None:
         """Test that objects without to_config_dict fall back to model string."""

lib/crewai/tests/test_flow_definition.py

@@ -44,6 +44,8 @@ def test_flow_public_exports_are_explicit():
         "FlowDefinition",
         "FlowDefinitionCondition",
         "FlowDefinitionDiagnostic",
+        "FlowEachActionDefinition",
+        "FlowEachInnerActionDefinition",
         "FlowExpressionActionDefinition",
         "FlowHumanFeedbackDefinition",
         "FlowMethodDefinition",
@@ -432,6 +434,73 @@ def test_flow_definition_round_trips_json_and_yaml():
     assert yaml_round_trip.methods["decide"].listen == "begin"
 
 
+def test_each_action_round_trips_json_and_yaml():
+    definition = flow_definition.FlowDefinition.from_dict(
+        {
+            "schema": "crewai.flow/v1",
+            "name": "EachFlow",
+            "methods": {
+                "process_rows": {
+                    "description": "Process every loaded row.",
+                    "start": True,
+                    "do": {
+                        "call": "each",
+                        "in": "state.rows",
+                        "do": [
+                            {
+                                "normalize": {
+                                    "call": "tool",
+                                    "ref": "my_tools:NormalizeRowTool",
+                                    "with": {"row": "${ item }"},
+                                }
+                            },
+                            {
+                                "save": {
+                                    "call": "code",
+                                    "ref": "my_flow:save_row",
+                                    "with": {
+                                        "row": "${ item }",
+                                        "normalized": "${ outputs.normalize }",
+                                    },
+                                }
+                            },
+                        ],
+                    },
+                }
+            },
+        }
+    )
+
+    json_round_trip = flow_definition.FlowDefinition.from_json(definition.to_json())
+    yaml_round_trip = flow_definition.FlowDefinition.from_yaml(definition.to_yaml())
+
+    assert json_round_trip.to_dict() == definition.to_dict()
+    assert yaml_round_trip.to_dict() == definition.to_dict()
+    assert yaml_round_trip.methods["process_rows"].description == (
+        "Process every loaded row."
+    )
+    assert yaml_round_trip.methods["process_rows"].do.call == "each"
+
+
+def test_flow_definition_rejects_invalid_method_names():
+    with pytest.raises(ValueError, match="Flow method names must match"):
+        flow_definition.FlowDefinition.from_dict(
+            {
+                "schema": "crewai.flow/v1",
+                "name": "InvalidMethodNameFlow",
+                "methods": {
+                    "process-rows": {
+                        "start": True,
+                        "do": {
+                            "call": "expression",
+                            "expr": "'done'",
+                        },
+                    }
+                },
+            }
+        )
+
+
 def test_flow_definition_detects_persist_metadata():
     @persist(verbose=True)
     class PersistedFlow(Flow[dict]):

lib/crewai/tests/test_flow_from_definition.py

@@ -67,6 +67,26 @@ class ToolInputFlow(Flow):
         return {"query": "ai agents", "suffix": " news"}
 
 
+class EachActionFlow(Flow):
+    def normalize_row(self, row: str, prefix: str = "normalized") -> str:
+        return f"{prefix}:{row}"
+
+    def save_row(self, row: str, normalized: str) -> dict[str, str]:
+        return {"row": row, "normalized": normalized}
+
+    def keyword_code(self, name: str, punctuation: str) -> str:
+        return f"{name}{punctuation}"
+
+    def fail_on_bad_row(self, row: str) -> str:
+        if row == "bad":
+            raise RuntimeError("bad row")
+        return row
+
+    def after_each(self) -> str:
+        self.state["after_count"] = self.state.get("after_count", 0) + 1
+        return f"after:{self.state['after_count']}"
+
+
 CHAIN_YAML = f"""
 schema: crewai.flow/v1
 name: ChainFlow
@@ -727,6 +747,313 @@ methods:
         flow.kickoff()
 
 
+def test_code_action_renders_keyword_inputs():
+    yaml_str = f"""
+schema: crewai.flow/v1
+name: CodeWithFlow
+methods:
+  greet:
+    do:
+      call: code
+      ref: {__name__}:EachActionFlow.keyword_code
+      with:
+        name: "${{state.name}}"
+        punctuation: "!"
+    start: true
+"""
+
+    flow = Flow.from_definition(FlowDefinition.from_yaml(yaml_str))
+
+    assert flow.kickoff(inputs={"name": "hello"}) == "hello!"
+
+
+def test_each_action_executes_one_nested_code_action():
+    yaml_str = f"""
+schema: crewai.flow/v1
+name: EachFlow
+methods:
+  process_rows:
+    do:
+      call: each
+      in: state.rows
+      do:
+        - normalize:
+            call: code
+            ref: {__name__}:EachActionFlow.normalize_row
+            with:
+              row: "${{item}}"
+    start: true
+"""
+
+    flow = Flow.from_definition(FlowDefinition.from_yaml(yaml_str))
+
+    assert flow.kickoff(inputs={"rows": ["a", "b"]}) == [
+        "normalized:a",
+        "normalized:b",
+    ]
+
+
+def test_each_action_uses_iteration_outputs_between_nested_actions():
+    yaml_str = f"""
+schema: crewai.flow/v1
+name: EachFlow
+methods:
+  process_rows:
+    do:
+      call: each
+      in: state.rows
+      do:
+        - normalize:
+            call: code
+            ref: {__name__}:EachActionFlow.normalize_row
+            with:
+              row: "${{item}}"
+              prefix: saved
+        - save:
+            call: code
+            ref: {__name__}:EachActionFlow.save_row
+            with:
+              row: "${{item}}"
+              normalized: "${{outputs.normalize}}"
+    start: true
+"""
+
+    flow = Flow.from_definition(FlowDefinition.from_yaml(yaml_str))
+
+    assert flow.kickoff(inputs={"rows": ["a", "b"]}) == [
+        {"row": "a", "normalized": "saved:a"},
+        {"row": "b", "normalized": "saved:b"},
+    ]
+
+
+def test_each_action_resets_inner_outputs_between_iterations():
+    yaml_str = """
+schema: crewai.flow/v1
+name: EachFlow
+methods:
+  process_rows:
+    do:
+      call: each
+      in: state.rows
+      do:
+        - leak_check:
+            call: expression
+            expr: "has(outputs.previous) ? outputs.previous : 'empty'"
+        - previous:
+            call: expression
+            expr: item
+    start: true
+"""
+
+    flow = Flow.from_definition(FlowDefinition.from_yaml(yaml_str))
+
+    assert flow.kickoff(inputs={"rows": ["a", "b"]}) == ["a", "b"]
+    assert flow._method_outputs == [
+        {"method": "process_rows", "output": ["a", "b"]}
+    ]
+
+
+def test_each_action_preserves_flow_outputs_and_prefers_inner_outputs():
+    yaml_str = """
+schema: crewai.flow/v1
+name: EachFlow
+methods:
+  seed:
+    do:
+      call: expression
+      expr: "'global'"
+    start: true
+  process_rows:
+    do:
+      call: each
+      in: state.rows
+      do:
+        - before_shadow:
+            call: expression
+            expr: "outputs.seed + ':' + item"
+        - seed:
+            call: expression
+            expr: "'local:' + item"
+        - after_shadow:
+            call: expression
+            expr: "outputs.seed"
+    listen: seed
+"""
+
+    flow = Flow.from_definition(FlowDefinition.from_yaml(yaml_str))
+
+    assert flow.kickoff(inputs={"rows": ["a", "b"]}) == [
+        "local:a",
+        "local:b",
+    ]
+    assert flow._method_outputs == [
+        {"method": "seed", "output": "global"},
+        {"method": "process_rows", "output": ["local:a", "local:b"]},
+    ]
+
+
+def test_each_action_empty_list_returns_empty_and_listener_runs_once():
+    yaml_str = f"""
+schema: crewai.flow/v1
+name: EachFlow
+methods:
+  process_rows:
+    do:
+      call: each
+      in: state.rows
+      do:
+        - normalize:
+            call: code
+            ref: {__name__}:EachActionFlow.normalize_row
+            with:
+              row: "${{item}}"
+    start: true
+  after_each:
+    do:
+      call: code
+      ref: {__name__}:EachActionFlow.after_each
+    listen: process_rows
+"""
+
+    flow = Flow.from_definition(FlowDefinition.from_yaml(yaml_str))
+    events = []
+    with crewai_event_bus.scoped_handlers():
+
+        @crewai_event_bus.on(MethodExecutionFinishedEvent)
+        def on_finished(source, event):
+            events.append(event.method_name)
+
+        result = flow.kickoff(inputs={"rows": []})
+
+    assert result == "after:1"
+    assert flow.method_outputs == [[], "after:1"]
+    assert flow.state["after_count"] == 1
+    assert events.count("process_rows") == 1
+    assert events.count("after_each") == 1
+
+
+@pytest.mark.parametrize(
+    ("expr", "inputs"),
+    [
+        ("1", {}),
+        ('"rows"', {}),
+        ("state.rows", {"rows": {"a": 1}}),
+    ],
+)
+def test_each_action_rejects_non_list_inputs(expr, inputs):
+    definition = FlowDefinition.from_dict(
+        {
+            "schema": "crewai.flow/v1",
+            "name": "EachFlow",
+            "methods": {
+                "process_rows": {
+                    "start": True,
+                    "do": {
+                        "call": "each",
+                        "in": expr,
+                        "do": [{"value": {"call": "expression", "expr": "item"}}],
+                    },
+                }
+            },
+        }
+    )
+    flow = Flow.from_definition(definition)
+
+    with pytest.raises(ValueError, match="each.in must evaluate to an array"):
+        flow.kickoff(inputs=inputs)
+
+
+@pytest.mark.parametrize(
+    "action_do",
+    [
+        [],
+        [{"first": {"call": "expression", "expr": "item"}, "second": {"call": "expression", "expr": "item"}}],
+        [{"1bad": {"call": "expression", "expr": "item"}}],
+        [
+            {"same": {"call": "expression", "expr": "item"}},
+            {"same": {"call": "expression", "expr": "item"}},
+        ],
+    ],
+)
+def test_each_action_validates_inner_action_shape(action_do):
+    with pytest.raises(ValidationError):
+        FlowDefinition.from_dict(
+            {
+                "schema": "crewai.flow/v1",
+                "name": "EachFlow",
+                "methods": {
+                    "process_rows": {
+                        "start": True,
+                        "do": {
+                            "call": "each",
+                            "in": "state.rows",
+                            "do": action_do,
+                        },
+                    }
+                },
+            }
+        )
+
+
+def test_each_action_rejects_nested_each_actions():
+    with pytest.raises(ValidationError):
+        FlowDefinition.from_dict(
+            {
+                "schema": "crewai.flow/v1",
+                "name": "EachFlow",
+                "methods": {
+                    "process_rows": {
+                        "start": True,
+                        "do": {
+                            "call": "each",
+                            "in": "state.rows",
+                            "do": [
+                                {
+                                    "nested": {
+                                        "call": "each",
+                                        "in": "state.children",
+                                        "do": [
+                                            {
+                                                "child": {
+                                                    "call": "expression",
+                                                    "expr": "item",
+                                                }
+                                            }
+                                        ],
+                                    }
+                                }
+                            ],
+                        },
+                    }
+                },
+            }
+        )
+
+
+def test_each_action_failure_fails_outer_method():
+    yaml_str = f"""
+schema: crewai.flow/v1
+name: EachFlow
+methods:
+  process_rows:
+    do:
+      call: each
+      in: state.rows
+      do:
+        - validate:
+            call: code
+            ref: {__name__}:EachActionFlow.fail_on_bad_row
+            with:
+              row: "${{item}}"
+    start: true
+"""
+
+    flow = Flow.from_definition(FlowDefinition.from_yaml(yaml_str))
+
+    with pytest.raises(RuntimeError, match="bad row"):
+        flow.kickoff(inputs={"rows": ["ok", "bad"]})
+
+
 def test_expression_action_round_trips():
     definition = FlowDefinition.from_dict(
         {
@@ -830,26 +1157,6 @@ def test_tool_action_requires_module_qualname_ref():
         Flow.from_definition(definition)
 
 
-def test_code_action_rejects_tool_inputs():
-    with pytest.raises(ValidationError):
-        FlowDefinition.from_dict(
-            {
-                "schema": "crewai.flow/v1",
-                "name": "InvalidCodeActionFlow",
-                "methods": {
-                    "begin": {
-                        "start": True,
-                        "do": {
-                            "call": "code",
-                            "ref": f"{__name__}:ChainFlow.begin",
-                            "with": {"search_query": "ai agents"},
-                        },
-                    }
-                },
-            }
-        )
-
-
 def test_pydantic_state_from_ref_parity():
     flow, result = assert_parity(PydanticStateFlow, PYDANTIC_STATE_YAML)
     assert result == "count=1"

lib/crewai/tests/test_human_feedback_decorator.py

@@ -597,7 +597,7 @@ class TestHumanFeedbackLearn:
         flow.memory.remember_many.assert_not_called()
 
     def test_learn_true_uses_default_llm(self):
-        """When learn=True and llm is not explicitly set, the default gpt-4o-mini is used."""
+        """When learn=True and llm is not explicitly set, the default gpt-5.4-mini is used."""
 
         @human_feedback(message="Review:", learn=True)
         def test_method(self):
@@ -606,8 +606,8 @@ class TestHumanFeedbackLearn:
         config = test_method.__human_feedback_config__
         assert config is not None
         assert config.learn is True
-        # llm defaults to "gpt-4o-mini" at the function level
-        assert config.llm == "gpt-4o-mini"
+        # llm defaults to "gpt-5.4-mini" at the function level
+        assert config.llm == "gpt-5.4-mini"
 
     def test_pre_review_failure_logs_and_returns_raw_output(self, caplog):
         """Pre-review LLM failure falls back to raw output AND logs a warning."""

lib/crewai/tests/test_human_feedback_integration.py

@@ -850,24 +850,22 @@ class TestLLMConfigPreservation:
         assert _deserialize_llm_from_context(None) is None
 
     def test_serialize_llm_preserves_provider_specific_fields(self):
-        """Test that provider-specific fields like project/location are serialized."""
+        """Test that provider-specific fields like base_url are serialized."""
         from crewai.flow.human_feedback import _serialize_llm_for_context
         from crewai.llm import LLM
 
-        # Create a Gemini-style LLM with project and non-default location
         llm = LLM(
-            model="gemini-2.0-flash",
-            provider="gemini",
-            project="my-project",
-            location="europe-west1",
+            model="llama3",
+            provider="ollama",
+            base_url="http://localhost:11434",
             temperature=0.3,
         )
 
         serialized = _serialize_llm_for_context(llm)
 
         assert isinstance(serialized, dict)
-        assert serialized.get("project") == "my-project"
-        assert serialized.get("location") == "europe-west1"
+        assert serialized.get("model") == "ollama/llama3"
+        assert serialized.get("base_url") == "http://localhost:11434/v1"
         assert serialized.get("temperature") == 0.3
 
     def test_config_preserved_through_full_flow_execution(self):

lib/crewai/tests/test_llm.py

@@ -463,6 +463,7 @@ def test_anthropic_message_formatting(anthropic_llm, system_message, user_messag
         anthropic_llm._format_messages_for_anthropic([{"invalid": "message"}])
 
 
+@pytest.mark.vcr()
 def test_deepseek_r1_with_open_router():
     if not os.getenv("OPEN_ROUTER_API_KEY"):
         pytest.skip("OPEN_ROUTER_API_KEY not set; skipping test.")

lib/crewai/tests/test_streaming_integration.py

@@ -247,8 +247,9 @@ class TestStreamingFlowIntegration:
         result = streaming.result
         assert result is not None
 
-    @pytest.mark.vcr()
     @pytest.mark.asyncio
+    @pytest.mark.timeout(180)
+    @pytest.mark.vcr()
     async def test_async_flow_streaming_from_docs(self) -> None:
         """Test async flow streaming example from documentation."""
 

lib/crewai/tests/utilities/test_llm_utils.py

@@ -23,6 +23,49 @@ def test_create_llm_with_valid_model_string() -> None:
         assert llm.model == "gpt-4o"
 
 
+def test_create_llm_with_config_dict() -> None:
+    with patch.dict(os.environ, {}, clear=True):
+        llm = create_llm(
+            llm_value={
+                "model": "llama3",
+                "provider": "ollama",
+                "temperature": 0.2,
+                "base_url": "http://localhost:11434",
+            }
+        )
+
+    assert isinstance(llm, BaseLLM)
+    assert llm.model == "llama3"
+    assert llm.provider == "ollama"
+    assert llm.temperature == 0.2
+    assert llm.base_url == "http://localhost:11434/v1"
+
+
+@pytest.mark.parametrize(
+    ("model_key", "model_value"),
+    [
+        ("model_name", "llama3"),
+        ("deployment_name", "custom-deployment"),
+    ],
+)
+def test_create_llm_with_config_dict_model_aliases(
+    model_key: str,
+    model_value: str,
+) -> None:
+    with patch.dict(os.environ, {}, clear=True):
+        llm = create_llm(
+            llm_value={
+                model_key: model_value,
+                "provider": "ollama",
+                "base_url": "http://localhost:11434",
+            }
+        )
+
+    assert isinstance(llm, BaseLLM)
+    assert llm.model == model_value
+    assert llm.provider == "ollama"
+
+
 def test_create_llm_with_invalid_model_string() -> None:
     with patch.dict(os.environ, {"OPENAI_API_KEY": "fake-key"}, clear=True):
         # For invalid model strings, create_llm succeeds but call() fails with API error

lib/crewai/tests/utilities/test_planning_handler.py

@@ -66,7 +66,7 @@ class TestInternalCrewPlanner:
                 ),
             )
             result = crew_planner._handle_crew_planning()
-            assert crew_planner.planning_agent_llm == "gpt-4o-mini"
+            assert crew_planner.planning_agent_llm == "gpt-5.4-mini"
             assert isinstance(result, PlannerTaskPydanticOutput)
             assert len(result.list_of_plans_per_task) == len(crew_planner.tasks)
             execute.assert_called_once()