> ## Documentation Index
> Fetch the complete documentation index at: https://docs.cascadeflow.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# CrewAI

> Hook-based harness integration for CrewAI with budget gating, metrics tracking, and decision traces across crew steps.

cascadeflow integrates with CrewAI through the native `llm_hooks` system. Call `enable()` to register global hooks that track crew execution where the real cost and control decisions happen: across agent steps inside the crew, not at the request edge.

## Install

```bash theme={null}
pip install "cascadeflow[crewai]"
```

## Quick Start

```python theme={null}
from crewai import Agent, Crew, Process, Task
import cascadeflow
from cascadeflow.integrations.crewai import CrewAIHarnessConfig, enable

cascadeflow.init(mode="observe")

# Enable harness hooks
config = CrewAIHarnessConfig(
    fail_open=True,
    budget_gate=True,
)
enable(config=config)

# Define agents and tasks as usual
researcher = Agent(
    role="Researcher",
    goal="Find relevant information",
    llm="gpt-4o-mini",
)

task = Task(
    description="Research the topic of AI agent frameworks",
    agent=researcher,
)

crew = Crew(
    agents=[researcher],
    tasks=[task],
    process=Process.sequential,
)

# Run with budget tracking
with cascadeflow.run(budget=1.00) as session:
    result = crew.kickoff()
    print(session.summary())
    for record in session.trace():
        print(f"Step {record['step']}: {record['action']} — {record['reason']}")
```

## Configuration

```python theme={null}
config = CrewAIHarnessConfig(
    fail_open=True,    # Continue on harness errors
    budget_gate=True,  # Enforce budget caps
)
```

## Features

* Tracks all crew steps automatically via `llm_hooks`
* Budget gating stops crew execution when budget is exceeded
* Full decision trace across all agents in the crew
* Fail-open mode for production safety
* No changes to existing CrewAI agent or task definitions

## Why This Integration Matters

* Crew-level workflows often hide expensive multi-step loops
* Hooks make those loops measurable and governable without rewriting crew logic
* Decision traces help explain runtime behavior across multiple agents

## Limitations

* Tool-level gating is not currently applied (CrewAI hooks operate at the LLM call level)
* Model switching depends on CrewAI's model configuration

<Tip>
  **Example on GitHub:** [integrations/crewai\_harness.py](https://github.com/lemony-ai/cascadeflow/blob/main/examples/integrations/crewai_harness.py)
</Tip>