Open Deep Research Project Details

Project Overview

Open Deep Research, developed by LangChain AI, is an experimental, fully open-source deep research assistant capable of automating in-depth research and generating comprehensive reports on any topic. The core goal of this project is to provide a simple, configurable open-source solution that competes with deep research products from companies like OpenAI, Anthropic, Perplexity, and Google.

Core Features

🎯 Dual Architecture Design

The project offers two distinct implementations, each with its own advantages:

1. Workflow Architecture

   • Structured plan-execute workflow
   • Human-in-the-loop interaction, allowing user feedback and approval of the report plan
   • Section-by-section creation, with reflection between search iterations
   • Supports all search tool providers

2. Multi-Agent Architecture

   • Supervisor-researcher architecture
   • Multiple independent agents working in parallel
   • Significantly reduces report generation time
   • Specialized tool design

🔧 Highly Configurable

Open Deep Research provides extensive configuration options to customize the research process and model behavior. All configurations can be set via the Web UI, environment variables, or direct configuration modification:

• Max Structured Output Retries (Default: 3): Maximum number of retries for structured output calls
• Allow Clarification (Default: true): Whether to allow the researcher to ask clarifying questions before starting research
• Max Concurrent Research Units (Default: 5): Maximum number of research units to run concurrently using sub-agents
• Search API (Default: Tavily): Supports multiple search API choices
• Max Researcher Iterations (Default: 3): Number of times the research supervisor reflects on the research and poses follow-up questions

🤖 Multi-Model Support

Open Deep Research uses multiple specialized models for different research tasks:

• Summary Model (Default: openai:gpt-4.1-nano): Summarizes research findings from search APIs
• Research Model (Default: openai:gpt-4.1): Conducts research and analysis
• Compression Model (Default: openai:gpt-4.1-mini): Compresses research findings from sub-agents
• Final Report Model (Default: openai:gpt-4.1): Writes the final comprehensive report

All models are configured using the init_chat_model() API, supporting providers such as OpenAI, Anthropic, Google Vertex AI, and more.

🔍 Rich Search Tool Integration

The project supports various search APIs:

• Tavily API - General web search
• Perplexity API - General web search
• Exa API - Powerful neural search
• ArXiv - Academic papers in physics, mathematics, computer science, etc.
• PubMed - Biomedical literature
• Linkup API - General web search
• DuckDuckGo API - General web search
• Google Search API - Custom search engine

Technical Architecture

Built on LangGraph

Open Deep Research is built on LangGraph, a framework for building resilient language agent graphs. This architecture enables the project to:

• Flexibly apply different research strategies
• Use intermediate results to guide exploration
• Handle open-ended research tasks

Three-Step Research Process

Open Deep Research uses agents to conduct research as part of a three-step process:

1. Scoping

   • User Clarification: Gathers all user context required for the research
   • Brief Generation: Converts detailed chat interactions into a comprehensive, focused research brief

2. Research

   • Conducts research using a supervisor agent
   • Delegates research tasks to an appropriate number of sub-agents
   • Determines if the research brief can be broken down into independent sub-topics

3. Report Generation

   • Writes the report in a single pass
   • Avoids coordination difficulties of multiple agents writing in parallel

Installation and Usage

Quick Start

# Clone the repository
git clone https://github.com/langchain-ai/open_deep_research.git
cd open_deep_research

# Set environment variables
cp .env.example .env

# Install and start the LangGraph server
uvx --refresh --from "langgraph-cli[inmem]" --with-editable . --python 3.11 langgraph dev --allow-blocking

Install as a Python Package

pip install open-deep-research

Usage

1. Via Studio UI:

   • 🚀 API: http://127.0.0.1:2024
   • 🎨 Studio UI: https://smith.langchain.com/studio/?baseUrl=http://127.0.0.1:2024
   • 📚 API Docs: http://127.0.0.1:2024/docs

2. Jupyter Notebook:

   • View examples: src/open_deep_research/graph.ipynb
   • Multi-agent example: src/open_deep_research/multi_agent.ipynb

Platform Integration

Open Agent Platform (OAP)

The Open Agent Platform (OAP) is a user interface where non-technical users can build and configure their own agents. OAP is ideal for allowing users to configure the deep researcher, using different MCP tools and search APIs best suited for their needs.

Performance and Evaluation

Benchmarking Results

Based on the results, Open Deep Research consistently and significantly improves answer quality on base models, demonstrated across Llama, DeepSeek, and Qwen flagship models, showcasing the quality of the workflow.

Testing Capabilities

# Test with default Anthropic models
python tests/run_test.py --all

# Test with OpenAI o3 models
python tests/run_test.py --all \
  --supervisor-model "openai:o3" \
  --researcher-model "openai:o3" \
  --planner-provider "openai" \
  --planner-model "o3" \
  --writer-provider "openai" \
  --writer-model "o3" \
  --eval-model "openai:o3" \
  --search-api "tavily"

Deployment Options

• Local Deployment: Run locally via the LangGraph server
• Cloud Deployment: Easily deployable to the LangGraph Platform

Project Ecosystem

Related Projects

• local-deep-researcher: Fully local web research and report writing assistant
• LangGraph: Framework for building resilient language agent graphs
• LangChain: For building context-aware reasoning applications

Community and Open Source

Since the release of OpenAI, the open-source community has made remarkable efforts in developing deep research alternatives. This project aims to provide the community with an accessible open-source deep research toolkit.

Technical Requirements

Model Compatibility

1. Structured Output Support: All models must support structured output
2. Tool Calling Support: Agent models need to support tool calling
3. Search API Compatibility: Research and compression models must support the selected search API

Tested Models

• Claude 3.7
• o3 and o3-mini
• gpt-4.1
• llama-3.3-70b-versatile (via Groq)

Advantages and Features

Design Principles

A key design principle of Open Deep Research is flexibility, enabling the exploration of different research strategies based on the request. Agents are well-suited for research because they can flexibly apply different strategies and use intermediate results to guide exploration.

Core Advantages

1. Open Source Transparency: Fully open source, allowing community review and improvement
2. Highly Configurable: Supports custom models, search tools, and report structures
3. Multi-Architecture Support: Offers both Workflow and Multi-Agent implementations
4. Broad Compatibility: Supports multiple AI model providers and search APIs
5. Human-in-the-Loop: Supports user feedback and report plan approval
6. Specialized Design: Uses specialized models for different tasks

Future Development

The project is continuously evolving, with plans to incorporate more features and improvements:

• Expand multi-agent architecture support for more search tools
• Improve report generation quality
• Enhance user interaction experience
• Optimize performance and resource utilization

Conclusion

Open Deep Research represents a significant advancement in open-source AI research tools. It not only provides functionalities that compete with commercial products but also maintains the transparency and customizability inherent in open-source projects. Whether for researchers, developers, or enterprise users, this powerful research assistant can be customized and deployed to meet their specific needs.