Skip to content

Releases: microsoft/autogen

python-v0.4.9

12 Mar 07:21
@ekzhu ekzhu
python-v0.4.9
bb8439c
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.
Compare
Choose a tag to compare
python-v0.4.9 Latest
Latest

What's New

Anthropic Model Client

Native support for Anthropic models. Get your update:
 

pip install -U "autogen-ext[anthropic]"

The new client follows the same interface as OpenAIChatCompletionClient so you can use it directly in your agents and teams.

import asyncio
from autogen_ext.models.anthropic import AnthropicChatCompletionClient
from autogen_core.models import UserMessage


async def main():
    anthropic_client = AnthropicChatCompletionClient(
        model="claude-3-sonnet-20240229",
        api_key="your-api-key",  # Optional if ANTHROPIC_API_KEY is set in environment
    )

    result = await anthropic_client.create([UserMessage(content="What is the capital of France?", source="user")])  # type: ignore
    print(result)


if __name__ == "__main__":
    asyncio.run(main())

You can also load the model client directly from a configuration dictionary:

from autogen_core.models import ChatCompletionClient

config = {
    "provider": "AnthropicChatCompletionClient",
    "config": {"model": "claude-3-sonnet-20240229"},
}

client = ChatCompletionClient.load_component(config)

To use with AssistantAgent and run the agent in a loop to match the behavior of Claude agents, you can use Single-Agent Team.

LlamaCpp Model Client

LlamaCpp is a great project for working with local models. Now we have native support via its official SDK.

pip install -U "autogen-ext[llama-cpp]"

To use a local model file:

import asyncio

from autogen_core.models import UserMessage
from autogen_ext.models.llama_cpp import LlamaCppChatCompletionClient


async def main():
    llama_client = LlamaCppChatCompletionClient(model_path="/path/to/your/model.gguf")
    result = await llama_client.create([UserMessage(content="What is the capital of France?", source="user")])
    print(result)


asyncio.run(main())

To use it with a Hugging Face model:

import asyncio

from autogen_core.models import UserMessage
from autogen_ext.models.llama_cpp import LlamaCppChatCompletionClient


async def main():
    llama_client = LlamaCppChatCompletionClient(
        repo_id="unsloth/phi-4-GGUF", filename="phi-4-Q2_K_L.gguf", n_gpu_layers=-1, seed=1337, n_ctx=5000
    )
    result = await llama_client.create([UserMessage(content="What is the capital of France?", source="user")])
    print(result)


asyncio.run(main())

Task-Centric Memory (Experimental)

Task-Centric memory is an experimental module that can give agents the ability to:

  • Accomplish general tasks more effectively by learning quickly and continually beyond context-window limitations.
  • Remember guidance, corrections, plans, and demonstrations provided by users (teachability)
  • Learn through the agent's own experience and adapt quickly to changing circumstances (self-improvement)
  • Avoid repeating mistakes on tasks that are similar to those previously encountered.

For example, you can use Teachability as a memory for AssistantAgent so your agent can learn from user teaching.

from autogen_agentchat.agents import AssistantAgent
from autogen_agentchat.ui import Console
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_ext.experimental.task_centric_memory import MemoryController
from autogen_ext.experimental.task_centric_memory.utils import Teachability


async def main():
    # Create a client
    client = OpenAIChatCompletionClient(model="gpt-4o-2024-08-06", )

    # Create an instance of Task-Centric Memory, passing minimal parameters for this simple example
    memory_controller = MemoryController(reset=False, client=client)

    # Wrap the memory controller in a Teachability instance
    teachability = Teachability(memory_controller=memory_controller)

    # Create an AssistantAgent, and attach teachability as its memory
    assistant_agent = AssistantAgent(
        name="teachable_agent",
        system_message = "You are a helpful AI assistant, with the special ability to remember user teachings from prior conversations.",
        model_client=client,
        memory=[teachability],
    )

    # Enter a loop to chat with the teachable agent
    print("Now chatting with a teachable agent. Please enter your first message. Type 'exit' or 'quit' to quit.")
    while True:
        user_input = input("\nYou: ")
        if user_input.lower() in ["exit", "quit"]:
            break
        await Console(assistant_agent.run_stream(task=user_input))

if __name__ == "__main__":
    import asyncio
    asyncio.run(main())

Head over to its README for details, and the samples for runnable examples.

New Sample: Gitty (Experimental)

Gitty is an experimental application built to help easing the burden on open-source project maintainers. Currently, it can generate auto reply to issues.

To use:

gitty --repo microsoft/autogen issue 5212

Head over to Gitty to see details.

Improved Tracing and Logging

In this version, we made a number of improvements on tracing and logging.

  • add LLMStreamStartEvent and LLMStreamEndEvent by @EItanya in #5890
  • Allow for tracing via context provider by @EItanya in #5889
  • Fix span structure for tracing by @ekzhu in #5853
  • Add ToolCallEvent and log it from all builtin tools by @ekzhu in #5859

Powershell Support for LocalCommandLineCodeExecutor

  • feat: update local code executor to support powershell by @lspinheiro in #5884

Website Accessibility Improvements

@peterychang has made huge improvements to the accessibility of our documentation website. Thank you @peterychang!

Bug Fixes

  • fix: save_state should not require the team to be stopped. by @ekzhu in #5885
  • fix: remove max_tokens from az ai client create call when stream=True by @ekzhu in #5860
  • fix: add plugin to kernel by @lspinheiro in #5830
  • fix: warn when using reflection on tool use with Claude models by @ekzhu in #5829

Other Python Related Changes

  • doc: update termination tutorial to include FunctionCallTermination condition and fix formatting by @ekzhu in #5813
  • docs: Add note recommending PythonCodeExecutionTool as an alternative to CodeExecutorAgent by @ekzhu in #5809
  • Update quickstart.ipynb by @taswar in #5815
  • Fix warning in selector gorup chat guide by @ekzhu in #5849
  • Support for external agent runtime in AgentChat by @ekzhu in #5843
  • update ollama usage docs by @ekzhu in #5854
  • Update markitdown requirements to >= 0.0.1, while still in the 0.0.x range by @afourney in #5864
  • Add client close by @afourney in #5871
  • Update README to clarify Web Browsing Agent Team usage, and use animated Chromium browser by @ekzhu in #5861
  • Add author name before their message in Chainlit team sample by @DavidYu00 in #5878
  • Bump axios from 1.7.9 to 1.8.2 in /python/packages/autogen-studio/frontend by @dependabot in #5874
  • Add an optional base path to FileSurfer by @husseinmozannar in #5886
  • feat: Pause and Resume for AgentChat Teams and Agents by @ekzhu in #5887
  • update version to v0.4.9 by @ekzhu in #5903

New Contributors

**Full Chang...

Read more

Contributors

ekzhu, taswar, and 11 other contributors
Assets 2
Loading

python-v0.4.8.2

07 Mar 23:50
@ekzhu ekzhu
python-v0.4.8.2
c6c2c8e
Compare
Choose a tag to compare
python-v0.4.8.2

Patch Fixes

  • fix: Remove max_tokens=20 from AzureAIChatCompletionClient.create_stream's create call when stream=True #5860
  • fix: Add close() method to built-in model clients to ensure the async event loop is closed when program exits. This should fix the "ResourceWarning: unclosed transport when importing web_surfer" errors. #5871

Full Changelog: python-v0.4.8.1...python-v0.4.8.2

Loading

python-v0.4.8.1

05 Mar 17:17
@ekzhu ekzhu
python-v0.4.8.1
3b9aa01
Compare
Choose a tag to compare
python-v0.4.8.1

Patch fixes to v0.4.8:

  • Fixing SKChatCompletionAdapter bug that disabled tool use #5830

Full Changelog: python-v0.4.8...python-v0.4.8.1

Loading

python-v0.4.8

04 Mar 06:47
@ekzhu ekzhu
python-v0.4.8
e7b4770
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.
Compare
Choose a tag to compare
python-v0.4.8

What's New

Ollama Chat Completion Client

To use the new Ollama Client:

pip install -U "autogen-ext[ollama]"

from autogen_ext.models.ollama import OllamaChatCompletionClient
from autogen_core.models import UserMessage

ollama_client = OllamaChatCompletionClient(
    model="llama3",
)

result = await ollama_client.create([UserMessage(content="What is the capital of France?", source="user")])  # type: ignore
print(result)

To load a client from configuration:

from autogen_core.models import ChatCompletionClient

config = {
    "provider": "OllamaChatCompletionClient",
    "config": {"model": "llama3"},
}

client = ChatCompletionClient.load_component(config)

It also supports structured output:

from autogen_ext.models.ollama import OllamaChatCompletionClient
from autogen_core.models import UserMessage
from pydantic import BaseModel


class StructuredOutput(BaseModel):
    first_name: str
    last_name: str


ollama_client = OllamaChatCompletionClient(
    model="llama3",
    response_format=StructuredOutput,
)
result = await ollama_client.create([UserMessage(content="Who was the first man on the moon?", source="user")])  # type: ignore
print(result)

New Required name Field in FunctionExecutionResult

Now name field is required in FunctionExecutionResult:

exec_result = FunctionExecutionResult(call_id="...", content="...", name="...", is_error=False)
  • fix: Update SKChatCompletionAdapter message conversion by @lspinheiro in #5749

Using thought Field in CreateResult and ThoughtEvent

Now CreateResult uses the optional thought field for the extra text content generated as part of a tool call from model. It is currently supported by OpenAIChatCompletionClient.

When available, the thought content will be emitted by AssistantAgent as a ThoughtEvent message.

  • feat: Add thought process handling in tool calls and expose ThoughtEvent through stream in AgentChat by @ekzhu in #5500

New metadata Field in AgentChat Message Types

Added a metadata field for custom message content set by applications.

Exception in AgentChat Agents is now fatal

Now, if there is an exception raised within an AgentChat agent such as the AssistantAgent, instead of silently stopping the team, it will raise the exception.

New Termination Conditions

New termination conditions for better control of agents.

See how you use TextMessageTerminationCondition to control a single agent team running in a loop: https://microsoft.github.io/autogen/stable/user-guide/agentchat-user-guide/tutorial/teams.html#single-agent-team.

FunctionCallTermination is also discussed as an example for custom termination condition: https://microsoft.github.io/autogen/stable/user-guide/agentchat-user-guide/tutorial/termination.html#custom-termination-condition

  • TextMessageTerminationCondition for agentchat by @EItanya in #5742
  • FunctionCallTermination condition by @ekzhu in #5808

Docs Update

The ChainLit sample contains UserProxyAgent in a team, and shows you how to use it to get user input from UI. See: https://github.com/microsoft/autogen/tree/main/python/samples/agentchat_chainlit

  • doc & sample: Update documentation for human-in-the-loop and UserProxyAgent; Add UserProxyAgent to ChainLit sample; by @ekzhu in #5656
  • docs: Add logging instructions for AgentChat and enhance core logging guide by @ekzhu in #5655
  • doc: Enrich AssistantAgent API documentation with usage examples. by @ekzhu in #5653
  • doc: Update SelectorGroupChat doc on how to use O3-mini model. by @ekzhu in #5657
  • update human in the loop docs for agentchat by @victordibia in #5720
  • doc: update guide for termination condition and tool usage by @ekzhu in #5807
  • Add examples for custom model context in AssistantAgent and ChatCompletionContext by @ekzhu in #5810

Bug Fixes

  • Initialize BaseGroupChat before reset by @gagb in #5608
  • fix: Remove R1 model family from is_openai function by @ekzhu in #5652
  • fix: Crash in argument parsing when using Openrouter by @philippHorn in #5667
  • Fix: Add support for custom headers in HTTP tool requests by @linznin in #5660
  • fix: Structured output with tool calls for OpenAIChatCompletionClient by @ekzhu in #5671
  • fix: Allow background exceptions to be fatal by @jackgerrits in #5716
  • Fix: Auto-Convert Pydantic and Dataclass Arguments in AutoGen Tool Calls by @mjunaidca in #5737

Other Python Related Changes

N...

Read more

Contributors

weijen, ekzhu, and 25 other contributors
Loading

python-v0.4.7

17 Feb 06:41
@ekzhu ekzhu
python-v0.4.7
b04775f
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.
Compare
Choose a tag to compare
python-v0.4.7

Overview

This release contains various bug fixes and feature improvements for the Python API.

Related news: our .NET API website is up and running: https://microsoft.github.io/autogen/dotnet/dev/. Our .NET Core API now has dev releases. Check it out!  

Important

Starting from v0.4.7, ModelInfo's required fields will be enforced. So please include all required fields when you use model_info when creating model clients. For example,

from autogen_core.models import UserMessage
from autogen_ext.models.openai import OpenAIChatCompletionClient

model_client = OpenAIChatCompletionClient(
    model="llama3.2:latest",
    base_url="http://localhost:11434/v1",
    api_key="placeholder",
    model_info={
        "vision": False,
        "function_calling": True,
        "json_output": False,
        "family": "unknown",
    },
)

response = await model_client.create([UserMessage(content="What is the capital of France?", source="user")])
print(response)

See ModelInfo for more details.
 

New Features

  • DockerCommandLineCodeExecutor support for additional volume mounts, exposed host ports by @andrejpk in #5383
  • Remove and get subscription APIs for Python GrpcWorkerAgentRuntime by @jackgerrits in #5365
  • Add strict mode support to BaseTool, ToolSchema and FunctionTool to allow tool calls to be used together with structured output mode by @ekzhu in #5507
  • Make CodeExecutor components serializable by @victordibia in #5527

Bug Fixes

  • fix: Address tool call execution scenario when model produces empty tool call ids by @ekzhu in #5509
  • doc & fix: Enhance AgentInstantiationContext with detailed documentation and examples for agent instantiation; Fix a but that caused value error when the expected class is not provided in register_factory by @ekzhu in #5555
  • fix: Add model info validation and improve error messaging by @ekzhu in #5556
  • fix: Add warning and doc for Windows event loop policy to avoid subprocess issues in web surfer and local executor by @ekzhu in #5557

Doc Updates

  • doc: Update API doc for MCP tool to include installation instructions by @ekzhu in #5482
  • doc: Update AgentChat quickstart guide to enhance clarity and installation instructions by @ekzhu in #5499
  • doc: API doc example for langchain database tool kit by @ekzhu in #5498
  • Update Model Client Docs to Mention API Key from Environment Variables by @victordibia in #5515
  • doc: improve tool guide in Core API doc by @ekzhu in #5546

Other Python Related Changes

  • Update website version v0.4.6 by @ekzhu in #5481
  • Reduce number of doc jobs for old releases by @jackgerrits in #5375
  • Fix class name style in document by @weijen in #5516
  • Update custom-agents.ipynb by @yosuaw in #5531
  • fix: update 0.2 deployment workflow to use tag input instead of branch by @ekzhu in #5536
  • fix: update help text for model configuration argument by @gagb in #5533
  • Update python version to v0.4.7 by @ekzhu in #5558

New Contributors

Full Changelog: python-v0.4.6...python-v0.4.7

Contributors

weijen, ekzhu, and 5 other contributors
Loading

python-v0.4.6

11 Feb 02:55
@ekzhu ekzhu
python-v0.4.6
cd085e6
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.
Compare
Choose a tag to compare
python-v0.4.6

Features and Improvements

MCP Tool

In this release we added a new built-in tool by @richard-gyiko for using Model Context Protocol (MCP) servers. MCP is an open protocol that allows agents to tap into an ecosystem of tools, from browsing file system to Git repo management.

Here is an example of using the mcp-server-fetch tool for fetching web content as Markdown.

# pip install mcp-server-fetch autogen-ext[mcp]

import asyncio

from autogen_agentchat.agents import AssistantAgent
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_ext.tools.mcp import StdioServerParams, mcp_server_tools


async def main() -> None:
    # Get the fetch tool from mcp-server-fetch.
    fetch_mcp_server = StdioServerParams(command="uvx", args=["mcp-server-fetch"])
    tools = await mcp_server_tools(fetch_mcp_server)

    # Create an agent that can use the fetch tool.
    model_client = OpenAIChatCompletionClient(model="gpt-4o")
    agent = AssistantAgent(name="fetcher", model_client=model_client, tools=tools, reflect_on_tool_use=True)  # type: ignore

    # Let the agent fetch the content of a URL and summarize it.
    result = await agent.run(task="Summarize the content of https://en.wikipedia.org/wiki/Seattle")
    print(result.messages[-1].content)


asyncio.run(main())

HTTP Tool

In this release we introduce a new built-in tool built by @EItanya for querying HTTP-based API endpoints. This lets agent call remotely hosted tools through HTTP.

Here is an example of using the httpbin.org API for base64 decoding.

# pip install autogen-ext[http-tool]

import asyncio

from autogen_agentchat.agents import AssistantAgent
from autogen_agentchat.messages import TextMessage
from autogen_core import CancellationToken
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_ext.tools.http import HttpTool

# Define a JSON schema for a base64 decode tool
base64_schema = {
    "type": "object",
    "properties": {
        "value": {"type": "string", "description": "The base64 value to decode"},
    },
    "required": ["value"],
}

# Create an HTTP tool for the httpbin API
base64_tool = HttpTool(
    name="base64_decode",
    description="base64 decode a value",
    scheme="https",
    host="httpbin.org",
    port=443,
    path="/base64/{value}",
    method="GET",
    json_schema=base64_schema,
)


async def main():
    # Create an assistant with the base64 tool
    model = OpenAIChatCompletionClient(model="gpt-4")
    assistant = AssistantAgent("base64_assistant", model_client=model, tools=[base64_tool])

    # The assistant can now use the base64 tool to decode the string
    response = await assistant.on_messages(
        [TextMessage(content="Can you base64 decode the value 'YWJjZGU=', please?", source="user")],
        CancellationToken(),
    )
    print(response.chat_message.content)


asyncio.run(main())
  • Adding declarative HTTP tools to autogen ext by @EItanya in #5181

MagenticOne Improvement

We introduced several improvements to MagenticOne (M1) and its agents. We made M1 work with text-only models that can't read screenshots, and prompt changes to make it work better with smaller models.

Do you know now you can configure m1 CLI tool with a YAML configuration file?

SelectorGroupChat Improvement

In this release we made several improvements to make SelectorGroupChat work well with smaller models such as LLama 13B, and hosted models that do not support the name field in Chat Completion messages.

Do you know you can use models served through Ollama directly through the OpenAIChatCompletionClient? See: https://microsoft.github.io/autogen/stable/user-guide/agentchat-user-guide/tutorial/models.html#ollama

  • Get SelectorGroupChat working for Llama models. by @afourney in #5409
  • Mitigates #5401 by optionally prepending names to messages. by @afourney in #5448
  • fix: improve speaker selection in SelectorGroupChat for weaker models by @ekzhu in #5454

Gemini Model Client

We enhanced our support for Gemini models. Now you can use Gemini models without passing in model_info and base_url.

from autogen_core.models import UserMessage
from autogen_ext.models.openai import OpenAIChatCompletionClient

model_client = OpenAIChatCompletionClient(
    model="gemini-1.5-flash-8b",
    # api_key="GEMINI_API_KEY",
)

response = await model_client.create([UserMessage(content="What is the capital of France?", source="user")])
print(response)
  • feat: add gemini model families, enhance group chat selection for Gemini model and add tests by @ekzhu in #5334
  • feat: enhance Gemini model support in OpenAI client and tests by @ekzhu in #5461

AGBench Update

New Sample

Interested in integration with FastAPI? We have a new sample: https://github.com/microsoft/autogen/blob/main/python/samples/agentchat_fastapi

  • Add sample chat application with FastAPI by @ekzhu in #5433
  • docs: enhance human-in-the-loop tutorial with FastAPI websocket example by @ekzhu in #5455

Bug Fixes

  • Fix reading string args from m1 cli by @afourney in #5343
  • Fix summarize_page in a text-only context, and for unknown models. by @afourney in #5388
  • fix: warn on empty chunks, don't error out by @MohMaz in #5332
  • fix: add state management for oai assistant by @lspinheiro in #5352
  • fix: streaming token mode cannot work in function calls and will infi… by @so2liu in #5396
  • fix: do not count agent event in MaxMessageTermination condition by @ekzhu in #5436
  • fix: remove sk tool adapter plugin name by @lspinheiro in #5444
  • fix & doc: update selector prompt documentation and remove validation checks by @ekzhu in #5456
  • fix: update SK adapter stream tool call processing. by @lspinheiro in #5449
  • fix: Update SK kernel from tool to use method. by @lspinheiro in #5469

Other Python Changes

  • Update Python website to v0.4.5 by @ekzhu in #5316
  • Adding o3 family: o3-mini by @razvanvalca in #5325
  • Ensure ModelInfo field is serialized for OpenAIChatCompletionClient by @victordibia in #5315
  • docs(core_distributed-group-chat): fix the typos in the docs in the README.md by @jsburckhardt in #5347
  • Assistant agent drop images when not provided with a vision-capable model. by @afourney in #5351
  • docs(python): add instructions for syncing dependencies and checking samples by @ekzhu in #5362
  • Fix typo by @weijen in #5361
  • docs: add blog link to README for updates and resources by @gagb in #5368
  • Memory component base by @EItanya in #5380
  • Fixed example code in doc:Custom Agents by @weijen in #5381
  • Various web surfer fixes. by @afourney in #5393
  • Refactor grpc channel connection in servicer by @jackgerrits in #5402
  • Updates to proto for state apis by @jackgerrits in #5407
  • feat: add integration workflow for testing multiple packages by @ekzhu in #5412
  • Flush console output after every message. by @afourney in #5415
  • Use a root json element instead of dict by @jackgerrits in #5430
  • Split out GRPC tests by @jackgerrits in #5431
  • feat: enhance AzureAIChatCompletionClient validation and add unit tests by @ekzhu in #5417
  • Fix typo in Swarm doc by @weijen in #5435
  • Update teams.ipynb : In the sample code the termination condition is set to the text "APPROVE" but the documentation mentions "TERMIN...
Read more

Contributors

weijen, ekzhu, and 14 other contributors
Loading

autogenstudio-v0.4.1

08 Feb 00:22
@victordibia victordibia
autogenstudio-v0.4.1
9494ac9
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.
Compare
Choose a tag to compare
autogenstudio-v0.4.1 Pre-release
Pre-release

Whats New

AutoGen Studio Declarative Configuration

  • in #5172, you can now build your agents in python and export to a json format that works in autogen studio

AutoGen studio now used the same declarative configuration interface as the rest of the AutoGen library. This means you can create your agent teams in python and then dump_component() it into a JSON spec that can be directly used in AutoGen Studio! This eliminates compatibility (or feature inconsistency) errors between AGS/AgentChat Python as the exact same specs can be used across.

See a video tutorial on AutoGen Studio v0.4 (02/25) - https://youtu.be/oum6EI7wohM

A Friendly Introduction to AutoGen Studio v0.4

Here's an example of an agent team and how it is converted to a JSON file:

from autogen_agentchat.agents import AssistantAgent
from autogen_agentchat.teams import RoundRobinGroupChat
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_agentchat.conditions import  TextMentionTermination

agent = AssistantAgent(
        name="weather_agent",
        model_client=OpenAIChatCompletionClient(
            model="gpt-4o-mini",
        ),
    )

agent_team = RoundRobinGroupChat([agent], termination_condition=TextMentionTermination("TERMINATE"))
config = agent_team.dump_component()
print(config.model_dump_json())
{
  "provider": "autogen_agentchat.teams.RoundRobinGroupChat",
  "component_type": "team",
  "version": 1,
  "component_version": 1,
  "description": "A team that runs a group chat with participants taking turns in a round-robin fashion\n    to publish a message to all.",
  "label": "RoundRobinGroupChat",
  "config": {
    "participants": [
      {
        "provider": "autogen_agentchat.agents.AssistantAgent",
        "component_type": "agent",
        "version": 1,
        "component_version": 1,
        "description": "An agent that provides assistance with tool use.",
        "label": "AssistantAgent",
        "config": {
          "name": "weather_agent",
          "model_client": {
            "provider": "autogen_ext.models.openai.OpenAIChatCompletionClient",
            "component_type": "model",
            "version": 1,
            "component_version": 1,
            "description": "Chat completion client for OpenAI hosted models.",
            "label": "OpenAIChatCompletionClient",
            "config": { "model": "gpt-4o-mini" }
          },
          "tools": [],
          "handoffs": [],
          "model_context": {
            "provider": "autogen_core.model_context.UnboundedChatCompletionContext",
            "component_type": "chat_completion_context",
            "version": 1,
            "component_version": 1,
            "description": "An unbounded chat completion context that keeps a view of the all the messages.",
            "label": "UnboundedChatCompletionContext",
            "config": {}
          },
          "description": "An agent that provides assistance with ability to use tools.",
          "system_message": "You are a helpful AI assistant. Solve tasks using your tools. Reply with TERMINATE when the task has been completed.",
          "model_client_stream": false,
          "reflect_on_tool_use": false,
          "tool_call_summary_format": "{result}"
        }
      }
    ],
    "termination_condition": {
      "provider": "autogen_agentchat.conditions.TextMentionTermination",
      "component_type": "termination",
      "version": 1,
      "component_version": 1,
      "description": "Terminate the conversation if a specific text is mentioned.",
      "label": "TextMentionTermination",
      "config": { "text": "TERMINATE" }
    }
  }
}

Note: If you are building custom agents and want to use them in AGS, you will need to inherit from the AgentChat BaseChat agent and Component class.

Note: This is a breaking change in AutoGen Studio. You will need to update your AGS specs for any teams created with version autogenstudio <0.4.1

Ability to Test Teams in Team Builder

  • in #5392, you can now test your teams as you build them. No need to switch between team builder and playground sessions to test.

You can now test teams directly as you build them in the team builder UI. As you edit your team (either via drag and drop or by editing the JSON spec)

Image Image

New Default Agents in Gallery (Web Agent Team, Deep Research Team)

  • in #5416, adds an implementation of a Web Agent Team and Deep Research Team in the default gallery.

The default gallery now has two additional default agents that you can build on and test:

  • Web Agent Team - A team with 3 agents - a Web Surfer agent that can browse the web, a Verification Assistant that verifies and summarizes information, and a User Proxy that provides human feedback when needed.
  • Deep Research Team - A team with 3 agents - a Research Assistant that performs web searches and analyzes information, a Verifier that ensures research quality and completeness, and a Summary Agent that provides a detailed markdown summary of the research as a report to the user.

Other Improvements

Older features that are currently possible in v0.4.1

  • Real-time agent updates streaming to the frontend
  • Run control: You can now stop agents mid-execution if they're heading in the wrong direction, adjust the team, and continue
  • Interactive feedback: Add a UserProxyAgent to get human input through the UI during team runs
  • Message flow visualization: See how agents communicate with each other
  • Ability to import specifications from external galleries
  • Ability to wrap agent teams into an API using the AutoGen Studio CLI

To update to the latest version:

pip install -U autogenstudio

Overall roadmap for AutoGen Studion is here #4006 .
Contributions welcome!

Loading

python-v0.4.5

01 Feb 04:11
@ekzhu ekzhu
python-v0.4.5
756e2a4
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.
Compare
Choose a tag to compare
python-v0.4.5

What's New

Streaming for AgentChat agents and teams

  • Introduce ModelClientStreamingChunkEvent for streaming model output and update handling in agents and console by @ekzhu in #5208

To enable streaming from an AssistantAgent, set model_client_stream=True when creating it. The token stream will be available when you run the agent directly, or as part of a team when you call run_stream.

If you want to see tokens streaming in your console application, you can use Console directly.

import asyncio
from autogen_agentchat.agents import AssistantAgent
from autogen_agentchat.ui import Console
from autogen_ext.models.openai import OpenAIChatCompletionClient

async def main() -> None:
    agent = AssistantAgent("assistant", OpenAIChatCompletionClient(model="gpt-4o"), model_client_stream=True)
    await Console(agent.run_stream(task="Write a short story with a surprising ending."))

asyncio.run(main())

If you are handling the messages yourself and streaming to the frontend, you can handle
autogen_agentchat.messages.ModelClientStreamingChunkEvent message.

import asyncio
from autogen_agentchat.agents import AssistantAgent
from autogen_ext.models.openai import OpenAIChatCompletionClient

async def main() -> None:
    agent = AssistantAgent("assistant", OpenAIChatCompletionClient(model="gpt-4o"), model_client_stream=True)
    async for message in agent.run_stream(task="Write 3 line poem."):
        print(message)

asyncio.run(main())
source='user' models_usage=None content='Write 3 line poem.' type='TextMessage'
source='assistant' models_usage=None content='Silent' type='ModelClientStreamingChunkEvent'
source='assistant' models_usage=None content=' whispers' type='ModelClientStreamingChunkEvent'
source='assistant' models_usage=None content=' glide' type='ModelClientStreamingChunkEvent'
source='assistant' models_usage=None content=',' type='ModelClientStreamingChunkEvent'
source='assistant' models_usage=None content='  \n' type='ModelClientStreamingChunkEvent'
source='assistant' models_usage=None content='Moon' type='ModelClientStreamingChunkEvent'
source='assistant' models_usage=None content='lit' type='ModelClientStreamingChunkEvent'
source='assistant' models_usage=None content=' dreams' type='ModelClientStreamingChunkEvent'
source='assistant' models_usage=None content=' dance' type='ModelClientStreamingChunkEvent'
source='assistant' models_usage=None content=' through' type='ModelClientStreamingChunkEvent'
source='assistant' models_usage=None content=' the' type='ModelClientStreamingChunkEvent'
source='assistant' models_usage=None content=' night' type='ModelClientStreamingChunkEvent'
source='assistant' models_usage=None content=',' type='ModelClientStreamingChunkEvent'
source='assistant' models_usage=None content='  \n' type='ModelClientStreamingChunkEvent'
source='assistant' models_usage=None content='Stars' type='ModelClientStreamingChunkEvent'
source='assistant' models_usage=None content=' watch' type='ModelClientStreamingChunkEvent'
source='assistant' models_usage=None content=' from' type='ModelClientStreamingChunkEvent'
source='assistant' models_usage=None content=' above' type='ModelClientStreamingChunkEvent'
source='assistant' models_usage=None content='.' type='ModelClientStreamingChunkEvent'
source='assistant' models_usage=RequestUsage(prompt_tokens=0, completion_tokens=0) content='Silent whispers glide,  \nMoonlit dreams dance through the night,  \nStars watch from above.' type='TextMessage'
TaskResult(messages=[TextMessage(source='user', models_usage=None, content='Write 3 line poem.', type='TextMessage'), TextMessage(source='assistant', models_usage=RequestUsage(prompt_tokens=0, completion_tokens=0), content='Silent whispers glide,  \nMoonlit dreams dance through the night,  \nStars watch from above.', type='TextMessage')], stop_reason=None)

Read more here: https://microsoft.github.io/autogen/stable/user-guide/agentchat-user-guide/tutorial/agents.html#streaming-tokens

Also, see the sample showing how to stream a team's messages to ChainLit frontend: https://github.com/microsoft/autogen/tree/python-v0.4.5/python/samples/agentchat_chainlit

R1-style reasoning output

  • Support R1 reasoning text in model create result; enhance API docs by @ekzhu in #5262
import asyncio
from autogen_core.models import UserMessage, ModelFamily
from autogen_ext.models.openai import OpenAIChatCompletionClient

async def main() -> None:
    model_client = OpenAIChatCompletionClient(
        model="deepseek-r1:1.5b",
        api_key="placeholder",
        base_url="http://localhost:11434/v1",
        model_info={
            "function_calling": False,
            "json_output": False,
            "vision": False,
            "family": ModelFamily.R1,
        }
    )

    # Test basic completion with the Ollama deepseek-r1:1.5b model.
    create_result = await model_client.create(
        messages=[
            UserMessage(
                content="Taking two balls from a bag of 10 green balls and 20 red balls, "
                "what is the probability of getting a green and a red balls?",
                source="user",
            ),
        ]
    )

    # CreateResult.thought field contains the thinking content.
    print(create_result.thought)
    print(create_result.content)

asyncio.run(main())

Streaming is also supported with R1-style reasoning output.

See the sample showing R1 playing chess: https://github.com/microsoft/autogen/tree/python-v0.4.5/python/samples/agentchat_chess_game

FunctionTool for partial functions

Now you can define function tools from partial functions, where some parameters have been set before hand.

import json
from functools import partial
from autogen_core.tools import FunctionTool


def get_weather(country: str, city: str) -> str:
    return f"The temperature in {city}, {country} is 75°"


partial_function = partial(get_weather, "Germany")
tool = FunctionTool(partial_function, description="Partial function tool.")

print(json.dumps(tool.schema, indent=2))
{
  "name": "get_weather",
  "description": "Partial function tool.",
  "parameters": {
    "type": "object",
    "properties": {
      "city": {
        "description": "city",
        "title": "City",
        "type": "string"
      }
    },
    "required": [
      "city"
    ]
  }
}

CodeExecutorAgent update

  • Added an optional sources parameter to CodeExecutorAgent by @afourney in #5259

New Samples

  • Streamlit + AgentChat sample by @husseinkorly in #5306
  • ChainLit + AgentChat sample with streaming by @ekzhu in #5304
  • Chess sample showing R1-Style reasoning for planning and strategizing by @ekzhu in #5285

Documentation update:

  • Add Semantic Kernel Adapter documentation and usage examples in user guides by @ekzhu in #5256
  • Update human-in-the-loop tutorial with better system message to signal termination condition by @ekzhu in #5253

Moves

Bug Fixes

  • fix: handle non-string function arguments in tool calls and add corresponding warnings by @ekzhu in #5260
  • Add default_header support by @nour-bouzid in #5249
  • feat: update OpenAIAssistantAgent to support AsyncAzureOpenAI client by @ekzhu in #5312

All Other Python Related Changes

  • Update website for v0.4.4 by @ekzhu in #5246
  • update dependencies to work with protobuf 5 by @MohMaz in #5195
  • Adjusted M1 agent system prompt to remove TERMINATE by @afourney in #5263
    #5270
  • chore: update package versions to 0.4.5 and remove deprecated requirements by @ekzhu in #5280
  • Update Distributed Agent Runtime Cross-platform Sample by @linznin in #5164
  • fix: windows check ci failure by @bassmang in #5287
  • fix: type issues in streamlit sample and add streamlit to dev dependencies by @ekzhu in #5309
  • chore: add asyncio_atexit dependency to docker requirements by @ekzhu in #5307
  • feat: add o3 to model info; update chess example by @ekzhu in #5311

New Contributors

Full Changelog: v0.4.4...python-v0.4.5

Contributors

ekzhu, husseinkorly, and 5 other contributors
Loading

python-v0.4.4

29 Jan 03:32
@ekzhu ekzhu
v0.4.4
bd5a24b
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.
Compare
Choose a tag to compare
python-v0.4.4

What's New

Serializable Configuration for AgentChat

This new feature allows you to serialize an agent or a team to a JSON string, and deserialize them back into objects. Make sure to also read about save_state and load_state: https://microsoft.github.io/autogen/stable/user-guide/agentchat-user-guide/tutorial/state.html.

You now can serialize and deserialize both the configurations and the state of agents and teams.

For example, create a RoundRobinGroupChat, and serialize its configuration and state.

import asyncio
from autogen_agentchat.agents import AssistantAgent
from autogen_agentchat.teams import RoundRobinGroupChat
from autogen_agentchat.base import Team
from autogen_agentchat.ui import Console
from autogen_agentchat.conditions import TextMentionTermination
from autogen_ext.models.openai import OpenAIChatCompletionClient


async def dump_team_config() -> None:
    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.
    await Console(group_chat.run_stream(task="Write a short poem about winter."))
    # Dump the team configuration to a JSON file.
    config = group_chat.dump_component()
    with open("team_config.json", "w") as f:
        f.write(config.model_dump_json(indent=4))
    # Dump the team state to a JSON file.
    state = await group_chat.save_state()
    with open("team_state.json", "w") as f:
        f.write(json.dumps(state, indent=4))

asyncio.run(dump_team_config())

Produces serialized team configuration and state. Truncated for illustration purpose.

{
    "provider": "autogen_agentchat.teams.RoundRobinGroupChat",
    "component_type": "team",
    "version": 1,
    "component_version": 1,
    "description": "A team that runs a group chat with participants taking turns in a round-robin fashion\n    to publish a message to all.",
    "label": "RoundRobinGroupChat",
    "config": {
        "participants": [
            {
                "provider": "autogen_agentchat.agents.AssistantAgent",
                "component_type": "agent",
                "version": 1,
                "component_version": 1,
                "description": "An agent that provides assistance with tool use.",
                "label": "AssistantAgent",
                "config": {
                    "name": "assistant",
                    "model_client": {
                        "provider": "autogen_ext.models.openai.OpenAIChatCompletionClient",
                        "component_type": "model",
                        "version": 1,
                        "component_version": 1,
                        "description": "Chat completion client for OpenAI hosted models.",
                        "label": "OpenAIChatCompletionClient",
                        "config": {
                            "model": "gpt-4o"
                        }
{
    "type": "TeamState",
    "version": "1.0.0",
    "agent_states": {
        "group_chat_manager/25763eb1-78b2-4509-8607-7224ae383575": {
            "type": "RoundRobinManagerState",
            "version": "1.0.0",
            "message_thread": [
                {
                    "source": "user",
                    "models_usage": null,
                    "content": "Write a short poem about winter.",
                    "type": "TextMessage"
                },
                {
                    "source": "assistant",
                    "models_usage": {
                        "prompt_tokens": 25,
                        "completion_tokens": 150
                    },
                    "content": "Amidst the still and silent air,  \nWhere frost adorns the branches bare,  \nThe world transforms in shades of white,  \nA wondrous, shimmering, quiet sight.\n\nThe whisper of the wind is low,  \nAs snowflakes drift and dance and glow.  \nEach crystal, delicate and bright,  \nFalls gently through the silver night.\n\nThe earth is hushed in pure embrace,  \nA tranquil, glistening, untouched space.  \nYet warmth resides in hearts that roam,  \nFinding solace in the hearth of home.\n\nIn winter\u2019s breath, a promise lies,  \nBeneath the veil of cold, clear skies:  \nThat spring will wake the sleeping land,  \nAnd life will bloom where now we stand.",
                    "type": "TextMessage"

Load the configuration and state back into objects.

import asyncio
import json
from autogen_agentchat.teams import RoundRobinGroupChat
from autogen_agentchat.base import Team

async def load_team_config() -> None:
    # Load the team configuration from a JSON file.
    with open("team_config.json", "r") as f:
        config = json.load(f)
    group_chat = Team.load_component(config)
    # Load the team state from a JSON file.
    with open("team_state.json", "r") as f:
        state = json.load(f)
    await group_chat.load_state(state)
    assert isinstance(group_chat, RoundRobinGroupChat)

asyncio.run(load_team_config())

This new feature allows you to manage persistent sessions across server-client based user interaction.

Azure AI Client for Azure-Hosted Models

This allows you to use Azure and GitHub-hosted models, including Phi-4, Mistral models, and Cohere models.

import asyncio
import os

from autogen_core.models import UserMessage
from autogen_ext.models.azure import AzureAIChatCompletionClient
from azure.core.credentials import AzureKeyCredential


async def main() -> None:
    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",
        },
    )
    result = await client.create(
        [UserMessage(content="What is the capital of France?", source="user")]
    )
    print(result)


asyncio.run(main())

Rich Console UI for Magentic One CLI

You can now enable pretty printed output for m1 command line tool by adding --rich argument.

m1 --rich "Find information about AutoGen"
Screenshot 2025-01-28 191752

Default In-Memory Cache for ChatCompletionCache

This allows you to cache model client calls without specifying an external cache service.

import asyncio

from autogen_core.models import UserMessage
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_ext.models.cache import ChatCompletionCache


async def main() -> None:
    # 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)

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


asyncio.run(main())
The capital of France is Paris. False
The capital of France is Paris. True

Docs Update

  • Update model client documentation add Ollama, Gemini, Azure AI models by @ekzhu in #5196
  • Add Model Client Cache section to migration guide by @ekzhu in #5197
  • docs: Enhance documentation for SingleThreadedAgentRuntime with usage examples and clarifications; undeprecate process_next by @ekzhu in #5230
  • docs: Update user guide notebooks to enhance clarity and add structured output by @ekzhu in #5224
  • docs: Core API doc update: split out model context from model clients; separate framework and components by @ekzhu in #5171
  • docs: Add a helpful comment to swarm.ipynb by @withsmilo in https://github.com/microsoft/autogen...
Read more

Contributors

ekzhu, bih, and 15 other contributors
Loading

python-v0.4.3

22 Jan 16:14
@jackgerrits jackgerrits
v0.4.3
da1c2bf
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.
Compare
Choose a tag to compare
python-v0.4.3

What's new

This is the first release since 0.4.0 with significant new features! We look forward to hearing feedback and suggestions from the community.

Chat completion model cache

One of the big missing features from 0.2 was the ability to seamlessly cache model client completions. This release adds ChatCompletionCache which can wrap any other ChatCompletionClient and cache completions.

There is a CacheStore interface to allow for easy implementation of new caching backends. The currently available implementations are:

import asyncio
import tempfile

from autogen_core.models import UserMessage
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_ext.models.cache import ChatCompletionCache, CHAT_CACHE_VALUE_TYPE
from autogen_ext.cache_store.diskcache import DiskCacheStore
from diskcache import Cache


async def main():
    with tempfile.TemporaryDirectory() as tmpdirname:
        openai_model_client = OpenAIChatCompletionClient(model="gpt-4o")

        cache_store = DiskCacheStore[CHAT_CACHE_VALUE_TYPE](Cache(tmpdirname))
        cache_client = ChatCompletionCache(openai_model_client, cache_store)

        response = await cache_client.create([UserMessage(content="Hello, how are you?", source="user")])
        print(response)  # Should print response from OpenAI
        response = await cache_client.create([UserMessage(content="Hello, how are you?", source="user")])
        print(response)  # Should print cached response


asyncio.run(main())

ChatCompletionCache is not yet supported by the declarative component config, see the issue to track progress.

#4924 by @srjoglekar246

GraphRAG

This releases adds support for GraphRAG as a tool agents can call. You can find a sample for how to use this integration here, and docs for LocalSearchTool and GlobalSearchTool.

import asyncio
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_agentchat.ui import Console
from autogen_ext.tools.graphrag import GlobalSearchTool
from autogen_agentchat.agents import AssistantAgent


async def main():
    # Initialize the OpenAI client
    openai_client = OpenAIChatCompletionClient(
        model="gpt-4o-mini",
    )

    # Set up global search tool
    global_tool = GlobalSearchTool.from_settings(settings_path="./settings.yaml")

    # Create assistant agent with the global search tool
    assistant_agent = AssistantAgent(
        name="search_assistant",
        tools=[global_tool],
        model_client=openai_client,
        system_message=(
            "You are a tool selector AI assistant using the GraphRAG framework. "
            "Your primary task is to determine the appropriate search tool to call based on the user's query. "
            "For broader, abstract questions requiring a comprehensive understanding of the dataset, call the 'global_search' function."
        ),
    )

    # Run a sample query
    query = "What is the overall sentiment of the community reports?"
    await Console(assistant_agent.run_stream(task=query))


if __name__ == "__main__":
    asyncio.run(main())

#4612 by @lspinheiro

Semantic Kernel model adapter

Semantic Kernel has an extensive collection of AI connectors. In this release we added support to adapt a Semantic Kernel AI Connector to an AutoGen ChatCompletionClient using the SKChatCompletionAdapter.

Currently this requires passing the kernel during create, and so cannot be used with AssistantAgent directly yet. This will be fixed in a future release (#5144).

#4851 by @lspinheiro

AutoGen to Semantic Kernel tool adapter

We also added a tool adapter, but this time to allow AutoGen tools to be added to a Kernel, called KernelFunctionFromTool.

#4851 by @lspinheiro

Jupyter Code Executor

This release also brings forward Jupyter code executor functionality that we had in 0.2, as the JupyterCodeExecutor.

Please note that this currently on supports local execution and should be used with caution.

#4885 by @Leon0402

Memory

It's still early on but we merged the interface for agent memory in this release. This allows agents to enrich their context from a memory store and save information to it. The interface is defined in core and AssistantAgent in agentchat accepts memory as a parameter now. There is an initial example memory implementation which simply injects all memories as system messages for the agent. The intention is for the memory interface to be able to be used for both RAG and agent memory systems going forward.

#4438 by @victordibia, #5053 by @ekzhu

Declarative config

We're continuing to expand support for declarative configs throughout the framework. In this release, we've added support for termination conditions and base chat agents. Once we're done with this, you'll be able to configure and entire team of agents with a single config file and have it work seamlessly with AutoGen studio. Stay tuned!

#4984, #5055 by @victordibia

Other

  • Add sources field to TextMentionTermination by @Leon0402 in #5106
  • Update gpt-4o model version to 2024-08-06 by @ekzhu in #5117

Bug fixes

  • Retry multiple times when M1 selects an invalid agent. Make agent sel… by @afourney in #5079
  • fix: normalize finish reason in CreateResult response by @ekzhu in #5085
  • Pass context between AssistantAgent for handoffs by @ekzhu in #5084
  • fix: ensure proper handling of structured output in OpenAI client and improve test coverage for structured output by @ekzhu in #5116
  • fix: use tool_calls field to detect tool calls in OpenAI client; add integration tests for OpenAI and Gemini by @ekzhu in #5122

Other changes

Read more

Contributors

ekzhu, victordibia, and 12 other contributors
Loading