OpenDeepResearcher – AI Autonomous Research Assistant

An AI-powered autonomous research platform that automatically plans, searches, analyzes, and generates comprehensive academic research papers using a local LLM, eliminating cloud AI API costs.

---

Overview

OpenDeepResearcher is a full-stack AI research assistant that mimics the workflow of a human researcher. Given any research topic, it automatically breaks the problem into sub-questions, searches the web, synthesizes information, evaluates research completeness, and produces a professional academic research paper.

The application is powered by a local Large Language Model (LLM) running through LM Studio, allowing users to perform advanced AI-powered research without paying for cloud AI APIs. The backend is built with FastAPI and LangGraph, while the frontend is developed using React and Vite, providing a fast and interactive user experience.

---

Features

• 🤖 Autonomous Multi-Agent Research Pipeline
• 🧠 Local LLM Integration using LM Studio
• 🌐 Real-time DuckDuckGo Web Search
• 🔄 LangGraph-based Agent Workflow
• 📡 Live Progress Updates using Server-Sent Events (SSE)
• 📄 Automatic Academic Research Report Generation
• 📑 Export Reports as PDF
• 📊 Export Reports as PowerPoint (PPTX)
• 💬 Follow-up Question Answering
• 🎤 Voice Input Support
• 📚 Retrieval-Augmented Generation (RAG)
• 📝 Document Upload and Search
• 🌙 Dark & Light Theme
• 🕒 Session History
• 📱 Responsive Modern Interface

---

Architecture

                 User
                  │
                  ▼
          React Frontend
             (Vite + React)
                  │
          HTTP / SSE Requests
                  │
                  ▼
          FastAPI Backend
             (server.py)
                  │
                  ▼
    ┌───────────────────────────┐
    │     LangGraph Workflow    │
    │                           │
    │   Planner                 │
    │      │                    │
    │      ▼                    │
    │   Searcher                │
    │      │                    │
    │      ▼                    │
    │    Writer                 │
    │      │                    │
    │      ▼                    │
    │  Reflection ─────────┐    │
    │      │               │    │
    │      └── Continue? ──┘    │
    │             │             │
    │             ▼             │
    │         Reporter          │
    └───────────────────────────┘
                  │
      ┌───────────┼─────────────┐
      │           │             │
      ▼           ▼             ▼
 PDF Export   PPT Export   Follow-up Q&A
                  │
                  ▼
          Final Research Report


  External Services Used
  ┌────────────────────────────┐
  │ LM Studio (Local LLM)      │
  │ DuckDuckGo Search (DDGS)   │
  │ RAG Document Store         │
  └────────────────────────────┘

---

Tech Stack

Frontend

• React 18
• Vite
• React Markdown
• CSS Modules

Backend

• Python 3.12
• FastAPI
• Uvicorn
• LangGraph
• LangChain
• LangChain OpenAI

AI

• LM Studio
• Qwen2.5-3B-Instruct
• OpenAI Compatible API

Search

• DuckDuckGo Search (DDGS)

Export

• ReportLab
• python-pptx

Other Libraries

• PyPDF2
• python-dotenv

---

Multi-Agent Workflow

📝 Planner

Breaks the research topic into focused sub-questions to create a structured research plan.

🔍 Searcher

Retrieves real-time information using DuckDuckGo and searches uploaded RAG documents when available.

✍️ Writer

Summarizes search results into coherent research notes while continuously improving the report.

🤔 Reflection

Evaluates whether sufficient information has been collected or another research iteration is required.

📄 Reporter

Generates a polished academic research paper containing:

• Title
• Abstract
• Introduction
• Key Findings
• Applications
• Challenges
• Future Outlook
• Conclusion
• References

---

Project Structure

OpenDeepResearcher/
│
├── agents/
│   ├── planner.py
│   ├── searcher.py
│   ├── writer.py
│   ├── reflection.py
│   └── reporter.py
│
├── frontend/
│
├── utils/
│   ├── llm_client.py
│   ├── pdf_export.py
│   ├── pptx_export.py
│   ├── rag.py
│   └── state.py
│
├── data/
│
├── graph.py
├── server.py
├── config.py
├── requirements.txt
└── .env

---

API Endpoints

Endpoint	Method	Description
/research	POST	Generate complete research report
/research/stream	POST	Stream research progress
/followup	POST	Ask questions about the generated report
/export/pdf	POST	Download report as PDF
/export/pptx	POST	Download report as PowerPoint
/rag/upload	POST	Upload documents for RAG
/rag/search	POST	Search uploaded documents
/health	GET	Server health check

---

Installation

Clone Repository

git clone https://github.com/yourusername/OpenDeepResearcher.git
cd OpenDeepResearcher

Install Backend

pip install -r requirements.txt

Install Frontend

cd frontend
npm install

---

Environment Variables

Create a .env file:

LLM_BASE_URL=http://localhost:1234/v1
LLM_API_KEY=lm-studio
LLM_MODEL_NAME=qwen2.5-3b-instruct

MAX_SEARCH_RESULTS=5
MAX_RESEARCH_LOOPS=2

APP_HOST=0.0.0.0
APP_PORT=8000

---

Running the Application

Step 1: Start LM Studio

• Load your preferred LLM
• Start the Local Server

Step 2: Start Backend

python server.py

Step 3: Start Frontend

cd frontend
npm run dev

Open your browser:

http://localhost:5173


---

# Future Improvements

- Multiple LLM Support
- Google Scholar Integration
- Citation Quality Evaluation
- Research Mind Maps
- Docker Deployment
- Cloud Deployment
- Collaborative Research Sessions
- Advanced Semantic Search

---

# Why OpenDeepResearcher?

- Runs entirely on a local LLM
- No AI API costs
- Autonomous multi-agent workflow
- Real-time research progress
- Retrieval-Augmented Generation (RAG)
- Modern responsive UI
- Professional PDF and PPT exports
- Interactive follow-up Q&A

---

# Author

Shreeya Bhalwatkar
⭐ If you found this project useful, consider giving it a Star!