Skip to content
Get packs

Open Deep Research

A configurable deep research agent that writes research reports, built on LangGraph and usable with many model

GitHub Download source
Stars
12.7k
Forks
1.9k
Licence
MIT
Language
Python
Last update
3 months ago
Type
AI model / agent

This repository is archived and no longer maintained.

What it is

Open Deep Research is an open source agent that takes a research question, searches the web and other sources, and writes a report. It is built with LangGraph and LangChain's model interface, so you can switch among many LLM providers and search tools. The README says its performance is comparable to other deep research agents on the Deep Research Bench leaderboard. The repository is archived, so it is no longer being updated.

  • Ask a research question in LangGraph Studio and receive a generated report
  • Change the summarization, research, compression, or final report models in the configuration
  • Switch the default Tavily search to another supported search tool or add MCP servers
  • Run the included evaluation script against the Deep Research Bench dataset

Great for

  • Developers who want a readable, editable reference implementation of a deep research agent
  • Teams experimenting with swapping LLM providers, search APIs, or MCP tools
  • Researchers who want to run the Deep Research Bench evaluation on their own model choices
  • Students learning how multi-step LLM agents are structured with LangGraph

Not ideal for

  • Non-technical users who need a ready-made app, since setup requires Python, uv, and API keys
  • Anyone needing ongoing maintenance or support, because the repository is archived
  • Users on a tight budget, since the README warns a full 100-example evaluation can cost roughly $20-$100 depending on the model

How to get it

  1. 1

    Clone and build from source

    You'll need: Python 3.11 · uv · API keys for your chosen LLM and search provider

    Clone the repository

    git clone https://github.com/langchain-ai/open_deep_research.git

    Install dependencies

    uv sync

    Create your environment file

    cp .env.example .env

    Start the LangGraph server locally

    uvx --refresh --from "langgraph-cli[inmem]" --with-editable . --python 3.11 langgraph dev --allow-blocking
  2. 2

    Run it

    • Clone the repository and create a virtual environment with uv venv
    • Run uv sync to install dependencies
    • Copy .env.example to .env and add your API keys
    • Start the LangGraph server and open LangGraph Studio at the local URL
    • Enter a question in the messages field and click Submit
  3. 3

    Get the most out of it

    • Change model choices per task (summarization, research, compression, final report) in the configuration fields
    • Add MCP servers through the mcp_config field to extend the agent's tools
    • Use the native web search option for Anthropic or OpenAI models, as the README describes
    • Use Ollama or OpenRouter for local or alternative models by following the linked setup guides
    • Run the Deep Research Bench evaluation scripts to compare configurations

Downloads

Source code (.zip)main branch · straight from GitHub

Strengths

  • Fully open source under the MIT license
  • Works with many model providers through init_chat_model()
  • Configurable through a clear set of model and search settings

Trade-offs

  • The repository is archived, so it is unlikely to receive fixes or new models
  • Selected models must support structured outputs and tool calling, which limits some choices
  • Default setup relies on OpenAI models and Tavily search, which need separate API keys

How it compares

ProjectStarsTypeLevelUpdatedBest for
open_deep_research (this)12.7kAI model / agentintermediate3 months agoDevelopers who want a readable, editable reference implementation of a deep research agent
storm31.8kAI model / agentintermediate1.1 years agoDevelopers who want to build a customizable research or report-writing pipeline

Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.