New AutoGen Release - v0.4.4: Serializable Agent Configuration, Support for Azure Hosted Models and More

We just released v0.4.4 for the AutoGen library with several important developer experience improvements:

Full release announcement here.

ICYMI - we showed how to replicate capabilities of the OpenAI Operator Agent in < 50 lines of code with AutoGen ! Code here.

Serializable Configuration and State Management

You can now serialize and deserialize both configurations and states of agents and teams. This enables:

• Saving agent configurations as JSON for version control and deployment
• Maintaining persistent chat sessions across server restarts
• Managing state in distributed applications
• Loading pre-configured agent teams, improving reproducibility

The feature works with both individual agents and team configurations, supporting all standard AutoGen components including RoundRobinGroupChat.

model_client = OpenAIChatCompletionClient(model="gpt-4o")
    assistant = AssistantAgent(
        "assistant",
        model_client=model_client,
        system_message="You are a helpful assistant.",
    )
    critic = AssistantAgent(
        "critic",
        model_client=model_client,
        system_message="Provide feedback. Reply with 'APPROVE' if the feedback has been addressed.",
    )
    termination = TextMentionTermination("APPROVE", sources=["critic"])
    group_chat = RoundRobinGroupChat(
        [assistant, critic], termination_condition=termination
    )
    # Run the group chat. 
    # Dump the team configuration to a JSON file. !!
    config = group_chat.dump_component()        

I am excited to see how people use this feature!

New Client for Azure AI Hosted Models

The new Azure AI client provides direct access to Azure and GitHub-hosted models, including:

• Phi-4
• Mistral models
• Cohere models

Implementation requires only an Azure endpoint and GitHub token for authentication. The client handles model-specific configurations and API interactions automatically. Note that we already support AzureOpenAI models, but this expands to support for all BYOM models on Azure.

client = AzureAIChatCompletionClient(
        model="Phi-4",
        endpoint="https://models.inference.ai.azure.com",
        # To authenticate with the model you will need to generate a personal access token (PAT) in your GitHub settings.
        # Create your PAT token by following instructions here: https://docs.github.com/en/authentication/keeping-your-account-and-data-secure/managing-your-personal-access-tokens
        credential=AzureKeyCredential(os.environ["GITHUB_TOKEN"]),
        model_info={
            "json_output": False,
            "function_calling": False,
            "vision": False,
            "family": "unknown",
        },
    )        

Improved Model Client Caching

A new default in-memory cache is now available for ChatCompletionClient. Key features:

• No external cache service required
• Automatic caching of repeated queries
• Cache state persistence across sessions
• Simple implementation with existing model clients

# Create a model client.
    client = OpenAIChatCompletionClient(model="gpt-4o")

    # Create a cached wrapper around the model client.
    cached_client = ChatCompletionCache(client)

    # Call the cached client.
    result = await cached_client.create(
        [UserMessage(content="What is the capital of France?", source="user")]
    )
    print(result.content, result.cached)        

Other Updates

• Improve Console for Magentic One.
• Documentation Updates
• Bug fixes

Full release details here - https://github.com/microsoft/autogen/releases/tag/v0.4.4