AutoGen Extensions

0.7.5 · active · verified Mon Apr 13

AutoGen extensions library, version 0.7.5. This library provides additional tools and functionalities to enhance the capabilities of the Microsoft AutoGen multi-agent framework, such as advanced web search, persistent agent memory, and improved communication patterns. It follows the release cadence of its parent `autogen` library, with frequent updates.

Warnings

breaking `autogen-ext` is tightly coupled with specific versions of `autogen`. The current `autogen-ext` (v0.7.5) has a dependency constraint `autogen~=0.2.0` (i.e., `>=0.2.0, <0.3.0`). Using `autogen-ext` with newer versions of `autogen` (e.g., `autogen>=0.3.0` or even later `0.2.x` sub-versions that introduced breaking changes) will likely lead to runtime errors due to API changes in `autogen`.
Fix: Pin your `autogen` version to be compatible, e.g., `pip install autogen~=0.2.0`. Always check `autogen-ext`'s `pyproject.toml` or `setup.py` for the exact `autogen` dependency for your `autogen-ext` version.
gotcha Tools like `WebSearchTool` often require their own dedicated API keys (e.g., `SERPER_API_KEY` for Serper API), separate from the `OPENAI_API_KEY` used for the LLM itself. Misconfiguring or omitting the tool-specific API key will result in search operations failing silently or with generic errors.
Fix: Always check tool-specific documentation for required API keys and ensure they are correctly provided, typically through environment variables or directly in the tool's `config` dictionary (e.g., `WebSearchTool(config={"api_key": os.environ.get("SERPER_API_KEY")})`).

Install

pip install autogen-ext Install autogen-ext

Imports

WebSearchTool
```
from autogen_extensions.web_search import WebSearchTool
```
WebSearchTool is part of the autogen-ext library, not directly from autogen's core modules.

AgentMemory

from autogen_extensions.agent_memory import AgentMemory

Quickstart

This quickstart demonstrates how to integrate the `WebSearchTool` from `autogen-ext` into an AutoGen multi-agent conversation. It requires setting `OPENAI_API_KEY` and `SERPER_API_KEY` (e.g., from Serper.dev) as environment variables or directly in the code.

import os
import autogen
from autogen_extensions.web_search import WebSearchTool

# Configure LLM (replace with your actual configuration)
llm_config = {
    "model": "gpt-4o", # or "gpt-4", "gpt-3.5-turbo"
    "api_key": os.environ.get("OPENAI_API_KEY", "YOUR_OPENAI_API_KEY"),
}

# Create an AssistantAgent
assistant = autogen.AssistantAgent(
    name="assistant",
    llm_config=llm_config,
    is_termination_msg=lambda x: x.get("content", "").rstrip().endswith("TERMINATE"),
)

# Create a UserProxyAgent
user_proxy = autogen.UserProxyAgent(
    name="user_proxy",
    is_termination_msg=lambda x: x.get("content", "").rstrip().endswith("TERMINATE"),
    human_input_mode="NEVER",
    max_consecutive_auto_reply=10,
    llm_config=llm_config, # UserProxyAgent can also use LLM config for self-reflection
    code_execution_config={"work_dir": "autogen_extension_work_dir"},
)

# Register the WebSearchTool (requires SERPER_API_KEY or similar)
# Get an API key from https://serper.dev or a similar search API provider.
web_search_tool = WebSearchTool(config={"api_key": os.environ.get("SERPER_API_KEY", None)})
user_proxy.register_for_execution(web_search_tool)
assistant.register_for_llm(web_search_tool)

# Start the conversation
print("\n--- Initiating chat with WebSearchTool ---\n")
user_proxy.initiate_chat(
    assistant,
    message="What's the capital of France and what are its main attractions? Use web search to find out.",
)
print("\n--- Chat finished ---\n")

view raw JSON →