Skip to content

Tavily Search

The Tavily Search integration allows AG2 agents to perform real-time, AI-powered web searches. Tavily is a search API purpose-built for AI agents and LLMs, delivering accurate and factual results optimized for retrieval-augmented generation (RAG) workflows.

Configuring Your Tavily API Key#

  1. Create a Tavily Account:
  2. Visit Tavily

  3. Click Sign Up and create an account

  4. Get Your API Key:

  5. Navigate to Tavily API Dashboard

  6. Generate an API key under API Keys

  7. Set the TAVILY_API_KEY Environment Variable: 

    export TAVILY_API_KEY="your_api_key_here"

Package Installation#

Install AG2 with the tavily extra (and openai for the example below):

pip install -U "ag2[openai,tavily]"

Note: autogen and ag2 are aliases for the same PyPI package: 

pip install -U "autogen[openai,tavily]"

Implementation#

Imports#

import os
from autogen import AssistantAgent, UserProxyAgent, LLMConfig
from autogen.tools.experimental import TavilySearchTool

Agent Configuration#

llm_config = LLMConfig({"api_type": "openai", "model": "gpt-4o"})

assistant = AssistantAgent(
    name="assistant",
    llm_config=llm_config,
)

user_proxy = UserProxyAgent(
    name="user_proxy",
    human_input_mode="NEVER",
)

Tool Setup#

tavily_search_tool = TavilySearchTool(tavily_api_key=os.getenv("TAVILY_API_KEY"))

# Register the tool for LLM recommendation and execution.
tavily_search_tool.register_for_llm(assistant)
tavily_search_tool.register_for_execution(user_proxy)

Usage Example#

response = user_proxy.initiate_chat(
    recipient=assistant,
    message="What are the latest developments in multi-agent AI systems?",
    max_turns=2,
)
print(f"Final Answer: {response.summary}")

Parameters#

TavilySearchTool accepts the following search parameters at call time:

Parameter Type Default Description
query str required The search query string
search_depth str "basic" Search depth — "basic" for fast results or "advanced" for deeper, more comprehensive search
include_answer str "basic" Whether to include an AI-generated answer — "basic" or "advanced"
include_raw_content bool False Whether to include the raw page content in results
include_domains list[str] [] Restrict search to specific domains (e.g., ["arxiv.org", "github.com"])
num_results int 5 Maximum number of results to return

Output#

Each search returns a list of dictionaries with:

  • title — the result title
  • link — the result URL
  • snippet — a relevant content excerpt

Deep Research with Tavily#

Tavily powers the web search layer in both of AG2's deep research approaches:

  • DeepResearchTool — AG2's built-in tool for multi-step research tasks. Uses Tavily under the hood for web retrieval, tightly integrated into your agent pipeline.
  • GPT Researcher — a standalone multi-agent research system that uses Tavily for recursive querying and produces structured reports (Markdown, PDF, Docx).

Both approaches require a TAVILY_API_KEY. See the Deep Research guide for setup and configuration.

See Also#