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

# CrewAI

> Auto-track all LLM calls in CrewAI crews

## Installation

```bash theme={null}
pip install meshai-sdk[crewai]
```

## Usage

```python theme={null}
from meshai import MeshAI
from meshai.integrations.crewai import track_crewai

client = MeshAI(api_key="msh_...", agent_name="my-crew")
client.register(framework="crewai")

# Enable global tracking, applies to ALL crews
track_crewai(client)

# Run your crew as normal
from crewai import Agent, Task, Crew

researcher = Agent(role="Researcher", llm="gpt-4o", ...)
writer = Agent(role="Writer", llm="claude-sonnet-4-6", ...)

crew = Crew(agents=[researcher, writer], tasks=[...])
result = crew.kickoff()
# Each agent's LLM calls tracked with their specific model
```

## How It Works

MeshAI registers a global `after_llm_call` hook with CrewAI. After every LLM interaction, the hook:

1. Extracts the **model name** from the LLM context
2. Extracts **token counts** from the response
3. Infers the **provider** from the model name
4. Sends the usage event to MeshAI (buffered, batched)

This means each agent in your crew is tracked with its own model, no hardcoding needed.

## Multi-Model Crews

CrewAI crews often use different models per agent. MeshAI tracks each one separately:

```
Dashboard shows:
  researcher (gpt-4o)      → 15,000 tokens, $0.45
  writer (claude-sonnet)   → 8,000 tokens, $0.24
  reviewer (gpt-4o-mini)   → 3,000 tokens, $0.01
```