If you need an agent to perform Wikipedia searches and retrieve relevant information, WikipediaAgent can help you.

​
Installation

Install AG2 with the LLM model provider and Wikipedia platform extra: 
pip install ag2[openai,wikipedia]
If you have been using autogen or ag2, you can upgrade using: 
pip install -U "autogen[wikipedia,openai]"
or 
pip install -U "ag2[wikipedia,openai]"
as autogen and ag2 are aliases for the same PyPI package.

​
Capabilities

WikipediaAgent can:
  • Perform searches on Wikipedia using natural language queries.
  • Retrieve summaries and relevant information from Wikipedia articles.
  • Integrate Wikipedia search results into conversations or workflows.
It utilizes the WikipediaQueryRunTool to execute searches and fetch information directly from Wikipedia.

​
Built-In Tools

The WikipediaAgent has the following in-built tool:
  • WikipediaQueryRunTool Enables the agent to perform Wikipedia searches and retrieve relevant summaries.
  • WikipediaPageLoadTool Retrieves the full content of a specified Wikipedia page for more in-depth responses.

​
Platform Configuration

No special configuration is required to use WikipediaAgent. It operates using the Wikipedia API, which is publicly accessible. Ensure that your environment has internet access to allow the agent to fetch data from Wikipedia.

​
Code Example

Here’s a simple example using the WikipediaAgent to search for information about Australia: 
from autogen import LLMConfig
from autogen.agentchat import UserProxyAgent
from autogen.agents.experimental import WikipediaAgent

# Configure the LLM
llm_config = LLMConfig(api_type="openai", model="gpt-4o-mini")

# Initialize the WikipediaAgent
wiki_agent = WikipediaAgent(name="wiki-agent", llm_config=llm_config)

# Initialize the UserProxyAgent
user_proxy = UserProxyAgent(name="user_proxy", human_input_mode="NEVER", code_execution_config=False)

# Register WikipediaAgent's tools with the UserProxyAgent
for tool in wiki_agent.tools:
    tool.register_for_execution(user_proxy)

# Start the conversation
response = user_proxy.initiate_chat(
    recipient=wiki_agent,
    message="What's the population of Australia?",
    max_turns=2,
)
This setup allows the UserProxyAgent to delegate the query to the WikipediaAgent, which then performs the search and returns the information.

​
Tool Execution

In AG2, tool execution is typically handled by a separate agent that follows the main agent in the conversation. In this example, the UserProxyAgent is responsible for executing the tools registered by the WikipediaAgent. This separation ensures a clear distinction between decision-making (handled by WikipediaAgent) and action execution (handled by UserProxyAgent). For more details on integrating Wikipedia search tools into your agents, refer to the Wikipedia Search Tools Notebook.
Edit this page