Adding Google Search Capability to AG2

Google Search

The Google Search integration in AG2 allows users to perform real-time web searches within the AG2 framework. This is especially useful for retrieving up-to-date information that may not be available in static datasets.

Installation

To get started with the Google Search integration in AG2, follow these steps:

Install AG2 with the google-search extra. Since our examples also use openai and gemini, install them as well:

pip install -U ag2[openai,gemini,google-search]

Note: If you have been using autogen or ag2, all you need to do is upgrade it using:

pip install -U autogen[openai,gemini,google-search]

or

pip install -U ag2[openai,gemini,google-search]

as autogen, and ag2 are aliases for the same PyPI package.

You’re all set! Now you can start using Google Search with AG2.

Imports

import os

import autogen
from autogen import AssistantAgent
from autogen.tools.experimental import GoogleSearchTool

Using GoogleSearchTool with Gemini GenAI

The GoogleSearchTool enables search functionalities in AG2 and can be configured to use Gemini GenAI for an enhanced search experience. This section covers agent configuration and tool initialization.

Agent Configuration

config_list = autogen.config_list_from_json(
    env_or_file="OAI_CONFIG_LIST",
    filter_dict={
        "model": ["gemini-2.0-flash"],
    },
)

assistant = AssistantAgent(
    name="assistant",
    llm_config={"config_list": config_list},
)

GoogleSearchTool Initialization

For Gemini GenAI, you can use the GenAI prebuilt Google Search tool by setting use_internal_llm_tool_if_available to True.

Note: You cannot combine other tools with the prebuilt Google Search tool. If you need multiple tools, refer to the AG2 implementation section.

gs_tool = GoogleSearchTool(
    use_internal_llm_tool_if_available=True,
)
# Once initialized, register the tool with the assistant
gs_tool.register_for_llm(assistant)

Start the Conversation

With the setup complete, you can now use the assistant to fetch live web search results.

run_response = assistant.run(
    message="What happened with stock prices after deepseek was launched, please search the web.",
    tools=assistant.tools,
    max_turns=2,
    user_input=False,
)
run_response.process()

GoogleSearchTool with AG2 Google Search implementation

For all other LLM providers or if you need to combine Google Search with other tools, follow this section to configure AG2’s implementation.

Setup Google Search Engine and API Key

1. Create a Google Custom Search Engine (CSE):
   • Go to Google Programmable Search Engine
   • Click Get Started and create a search engine.
   • Under Sites to Search, select Search the entire web if you want global search.
   • Copy the Search Engine ID from the CSE dashboard (cx parameter from the url)
2. Get a Google API Key:
   • Go to Google Cloud Console
   • Create a new project.
   • Navigate to APIs & Services > Library, search for Custom Search API and enable it.
   • Go to APIs & Services > Credentials, click on Create Credentials > API key and copy it.
3. Export engine ID and api key bash export GOOGLE_SEARCH_ENGINE_ID="your_engine_id" export GOOGLE_SEARCH_API_KEY="your_api_key"

Agent Configuration

config_list = autogen.config_list_from_json(
    env_or_file="OAI_CONFIG_LIST",
    filter_dict={
        "model": ["gpt-4o-mini"],
    },
)

assistant = AssistantAgent(
    name="assistant",
    llm_config={"config_list": config_list},
)

GoogleSearchTool Initialization

Create GoogleSearchTool with your search_api_key and search_engine_id.

search_api_key = os.getenv("GOOGLE_SEARCH_API_KEY")
search_engine_id = os.getenv("GOOGLE_SEARCH_ENGINE_ID")

assert search_api_key is not None, "Please set GOOGLE_SEARCH_API_KEY environment variable"
assert search_engine_id is not None, "Please set GOOGLE_SEARCH_ENGINE_ID environment variable"

gs_tool = GoogleSearchTool(
    search_api_key=search_api_key,
    search_engine_id=search_engine_id,
)
# Once initialized, register the tool with the assistant:
gs_tool.register_for_llm(assistant)

Start the Conversation

With the setup complete, you can now use the assistant to fetch live web search results.

run_response = assistant.run(
    message="What happened with stock prices after deepseek was launched, please search the web.",
    tools=assistant.tools,
    max_turns=2,
    user_input=False,
)
run_response.process()

Edit this page