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

# AutoGen & CrewAI

> Route AutoGen and CrewAI through Xenovia using OpenAI-compatible model settings.

Both AutoGen and CrewAI can target OpenAI-compatible endpoints, which makes Xenovia a good fit. The main job is to point the framework at your Xenovia proxy URL and keep a stable session header for the whole run.

<Callout type="info">
  AutoGen has multiple major-package eras in the wild. The example below uses the current `autogen-agentchat` and `autogen-ext[openai]` packages rather than the older `pyautogen` setup.
</Callout>

## AutoGen

```bash theme={null}
pip install -U "autogen-agentchat" "autogen-ext[openai]"
```

```python theme={null}
import asyncio
import os
from autogen_agentchat.agents import AssistantAgent
from autogen_ext.models.openai import OpenAIChatCompletionClient

model_client = OpenAIChatCompletionClient(
    model="gpt-4o-mini",
    api_key=os.environ["XENOVIA_API_KEY"],
    base_url=f"https://runtime.xenovia.io/a/{os.environ['XENOVIA_PROXY_ID']}/openai/v1"
)

assistant = AssistantAgent(
    "assistant",
    model_client=model_client,
    system_message="You are a helpful assistant."
)

async def main() -> None:
    result = await assistant.run(task="Summarise the key principles of AI governance.")
    print(result.messages[-1].content)

asyncio.run(main())
```

If your AutoGen client version exposes default request headers, add a stable `X-Xenovia-Session-Id` there so every turn lands in the same Xenovia session.

## CrewAI

```bash theme={null}
pip install crewai
```

```python theme={null}
import os, uuid
from crewai import Agent, Task, Crew, LLM

session_id = str(uuid.uuid4())

llm = LLM(
    model="openai/gpt-4o-mini",
    api_key=os.environ["XENOVIA_API_KEY"],
    base_url=f"https://runtime.xenovia.io/a/{os.environ['XENOVIA_PROXY_ID']}/openai/v1",
    extra_headers={"X-Xenovia-Session-Id": session_id}
)

researcher = Agent(
    role="Researcher",
    goal="Find accurate information on the given topic",
    backstory="You are an expert researcher with strong analytical skills.",
    llm=llm
)

writer = Agent(
    role="Writer",
    goal="Write clear and concise reports",
    backstory="You are an experienced technical writer.",
    llm=llm
)

research_task = Task(
    description="Research AI governance frameworks",
    expected_output="A summary of key AI governance principles",
    agent=researcher
)

write_task = Task(
    description="Write a report based on the research",
    expected_output="A structured report on AI governance",
    agent=writer
)

crew = Crew(agents=[researcher, writer], tasks=[research_task, write_task])
result = crew.kickoff()
print(result)
```

## How multi-agent traces work

In multi-agent setups, each agent turn is a separate call through the proxy. Each call:

* Gets its own `X-Xenovia-Trace-Id`
* Is evaluated independently by the request-stage Rego policy
* Is grouped under the shared `X-Xenovia-Session-Id`

Filter by `session_id` in the Traces view to reconstruct the full agent conversation in order, with per-turn policy decisions and intent scores.

## Handling policy blocks

When a request is blocked by policy, AutoGen and CrewAI surface the upstream `403`. Add framework-level error handling around the top-level run call so you can log the failure and trace ID cleanly.