Eric Zhu 7e5c1154cf
Support for external agent runtime in AgentChat (#5843)
Resolves #4075

1. Introduce custom runtime parameter for all AgentChat teams
(RoundRobinGroupChat, SelectorGroupChat, etc.). This is done by making
sure each team's topics are isolated from other teams, and decoupling
state from agent identities. Also, I removed the closure agent from the
BaseGroupChat and use the group chat manager agent to relay messages to
the output message queue.
2. Added unit tests to test scenarios with custom runtimes by using
pytest fixture
3. Refactored existing unit tests to use ReplayChatCompletionClient with
a few improvements to the client.
4. Fix a one-liner bug in AssistantAgent that caused deserialized agent
to have handoffs.

How to use it? 

```python
import asyncio
from autogen_core import SingleThreadedAgentRuntime
from autogen_agentchat.agents import AssistantAgent
from autogen_agentchat.teams import RoundRobinGroupChat
from autogen_agentchat.conditions import TextMentionTermination
from autogen_ext.models.replay import ReplayChatCompletionClient

async def main() -> None:
    # Create a runtime
    runtime = SingleThreadedAgentRuntime()
    runtime.start()

    # Create a model client.
    model_client = ReplayChatCompletionClient(
        ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10"],
    )

    # Create agents
    agent1 = AssistantAgent("assistant1", model_client=model_client, system_message="You are a helpful assistant.")
    agent2 = AssistantAgent("assistant2", model_client=model_client, system_message="You are a helpful assistant.")

    # Create a termination condition
    termination_condition = TextMentionTermination("10", sources=["assistant1", "assistant2"])

    # Create a team
    team = RoundRobinGroupChat([agent1, agent2], runtime=runtime, termination_condition=termination_condition)

    # Run the team
    stream = team.run_stream(task="Count to 10.")
    async for message in stream:
        print(message)
    
    # Save the state.
    state = await team.save_state()

    # Load the state to an existing team.
    await team.load_state(state)

    # Run the team again
    model_client.reset()
    stream = team.run_stream(task="Count to 10.")
    async for message in stream:
        print(message)

    # Create a new team, with the same agent names.
    agent3 = AssistantAgent("assistant1", model_client=model_client, system_message="You are a helpful assistant.")
    agent4 = AssistantAgent("assistant2", model_client=model_client, system_message="You are a helpful assistant.")
    new_team = RoundRobinGroupChat([agent3, agent4], runtime=runtime, termination_condition=termination_condition)

    # Load the state to the new team.
    await new_team.load_state(state)

    # Run the new team
    model_client.reset()
    new_stream = new_team.run_stream(task="Count to 10.")
    async for message in new_stream:
        print(message)
    
    # Stop the runtime
    await runtime.stop()

asyncio.run(main())
```

TODOs as future PRs:
1. Documentation.
2. How to handle errors in custom runtime when the agent has exception?

---------

Co-authored-by: Ryan Sweet <rysweet@microsoft.com>
2025-03-06 10:32:52 -08:00
..
2025-03-04 09:56:49 -08:00
2025-03-04 09:56:49 -08:00
2025-03-04 09:56:49 -08:00

AutoGen Python packages

0.4 Docs PyPi autogen-core PyPi autogen-agentchat PyPi autogen-ext

This directory works as a single uv workspace containing all project packages. See packages to discover all project packages.

Migrating from 0.2.x?

Please refer to the migration guide for how to migrate your code from 0.2.x to 0.4.x.

Development

TL;DR, run all checks with:

uv sync --all-extras
source .venv/bin/activate
poe check

Setup

uv is a package manager that assists in creating the necessary environment and installing packages to run AutoGen.

Note: To prevent incompatibilities between versions the same UV version as is running in CI should be used. Check the version in CI by looking the setup-uv action, here for example.

For example, to change your version to 0.5.18, run:

uv self update 0.5.18

Virtual Environment

During development, you may need to test changes made to any of the packages.
To do so, create a virtual environment where the AutoGen packages are installed based on the current state of the directory.
Run the following commands at the root level of the Python directory:

uv sync --all-extras
source .venv/bin/activate
  • uv sync --all-extras will create a .venv directory at the current level and install packages from the current directory along with any other dependencies. The all-extras flag adds optional dependencies.
  • source .venv/bin/activate activates the virtual environment.

Common Tasks

To create a pull request (PR), ensure the following checks are met. You can run each check individually:

  • Format: poe format
  • Lint: poe lint
  • Test: poe test
  • Mypy: poe mypy
  • Pyright: poe pyright
  • Build docs: poe --directory ./packages/autogen-core/ docs-build
  • Auto rebuild+serve docs: poe --directory ./packages/autogen-core/ docs-serve
  • Check samples in python/samples: poe samples-code-check Alternatively, you can run all the checks with:
  • poe check

Note

These need to be run in the virtual environment.

Syncing Dependencies

When you pull new changes, you may need to update the dependencies. To do so, first make sure you are in the virtual environment, and then in the python directory, run:

uv sync --all-extras

This will update the dependencies in the virtual environment.

Creating a New Package

To create a new package, similar to autogen-core or autogen-chat, use the following:

uv sync --python 3.12
source .venv/bin/activate
cookiecutter ./templates/new-package/