🎓 Auto-Researcher: Multi-Agent Collaborative System for Academic Reviews

Auto-Researcher is an autonomous, multi-agent system designed to revolutionize how academic literature reviews are conducted. By orchestrating a team of specialized AI agents, it performs deep research, analyzes complex academic papers, and synthesizes comprehensive reviews with verified citations.

Now featuring a Hybrid Engine: Run 100% locally for privacy using Ollama, or switch to OpenRouter to leverage state-of-the-art cloud models like Grok, GPT-4, or Claude.

---

🚀 Why Auto-Researcher?

Academic research is time-consuming. Finding relevant papers, reading them, synthesizing findings, and ensuring accuracy can take weeks. Auto-Researcher compresses this workflow into minutes.

🌟 Key Differentiators

Feature	Auto-Researcher	Standard Chatbots (ChatGPT/Claude)	Traditional Search
Architecture	Multi-Agent Loop (Researcher → Analyst → Critic)	Single-shot generation	Manual keyword search
Visibility	Real-time Streaming (Watch agents think & work)	Static loading state	N/A
Flexibility	Hybrid Engine (Local Ollama OR Cloud OpenRouter)	Vendor Lock-in	N/A
Source Truth	Grounded in PDFs (Arxiv, Edu, ResearchGate)	Training data (often outdated/hallucinated)	Links only (no synthesis)
Privacy	Local Option Available (Ollama)	Data sent to cloud	N/A
Depth	Iterative Refinement (Self-Correction)	First draft only	N/A
Customization	Adjustable Depth, Source Count & Strictness	Fixed parameters	N/A

---

🧠 How It Works

The system utilizes a Graph-based Multi-Agent Architecture (built with LangGraph) with Real-time Streaming:

1. 🕵️‍♂️ The Researcher:

   • Parallel Search: Simultaneously queries Tavily and DuckDuckGo to maximize coverage.
   • Academic Filtering: Specifically targets arxiv.org, .edu, .ac.uk, and researchgate.net.
   • Smart Parsing: Downloads PDFs and extracts high-density text snippets (up to 1500 chars per section) while ignoring references/bibliographies to save context window.

2. ✍️ The Analyst:

   • High-Density Synthesis: Drafts comprehensive reports (2400+ words for deep searches).
   • Thematic Grouping: Automatically groups findings into logical themes.
   • Structured Output: Generates Markdown with Executive Summary, Key Findings, Methodological Notes, and Implications.

3. ⚖️ The Critic:

   • Fact-Checking: Reviews the draft for hallucinations and vague generalizations.
   • Quantitative Enforcement: Rejects drafts that lack specific numbers/data.
   • Feedback Loop: Triggers revisions if the quality score drops below the threshold (configurable via Strictness setting).

Real-time Updates: The backend streams events (SSE) to the frontend, allowing you to watch the agents transition between states (Researching → Drafting → Critiquing) live via a high-tech dashboard.

---

🛠️ Tech Stack

• Backend: Python, FastAPI, LangGraph, LangChain
• Communication: Server-Sent Events (SSE) for real-time streaming
• Frontend: React 19, Vite, Tailwind CSS v4, Framer Motion
• Visualization: React Force Graph (Interactive Knowledge Graph)
• LLM Engine:
  • Local: Ollama (Llama 3, Mistral, etc.)
  • Cloud: OpenRouter (Grok, GPT-4, Claude, etc.)
• Search: Tavily API (Primary) + DuckDuckGo (Fallback/Supplement)
• PDF Processing: PyMuPDF (Fitz) with intelligent text extraction

---

⚡ Getting Started

Prerequisites

• Python 3.10+
• Node.js 16+
• Ollama installed and running (if using local mode)

Installation

1. Clone the repository

   git clone https://github.com/royxlead/auto-researcher-python.git
   cd auto-researcher-python

2. Setup Backend

   python -m venv .venv
   source .venv/bin/activate  # Windows: .venv\Scripts\activate
   pip install -r requirements.txt

3. Configure Environment Create a .env file in the root directory:

   TAVILY_API_KEY=your_key_here  # Recommended for better search results
   
   # Local Mode Settings
   OLLAMA_BASE_URL=http://localhost:11434
   OLLAMA_MODEL=llama3:8b
   
   # Cloud Mode Settings (Optional)
   OPENROUTER_API_KEY=sk-or-...
   OPENROUTER_MODEL=x-ai/grok-4.1-fast

4. Start the System

   • Backend: python run.py
   • Frontend: cd frontend && npm install && npm run dev

Usage

1. Open http://localhost:5173 in your browser.
2. Enter your research topic (e.g., "Impact of solid-state batteries on EV range").
3. Configure Research:
   • Depth: Controls the length and detail of the report (Fast / Balanced / Deep).
   • Papers: Select how many sources to analyze (5 - 50).
   • Strictness: Adjust the Critic's threshold for accepting drafts (Lenient / Balanced / Strict).
   • Provider: Choose Ollama (Local) or OpenRouter (Cloud).
   • Model Name: Specify the model (e.g., llama3 or x-ai/grok-4.1-fast).
4. Click the arrow to start. The system will visualize the research steps in real-time.
5. Interact:
   • Read Aloud: Use the Text-to-Speech feature to listen to the report.
   • Visualize: Explore the connections between papers in the Knowledge Graph.
   • Export: Download the report as a Markdown file.

---

🔮 Future Roadmap

• Cloud Mode: Optional switch to OpenRouter for heavier workloads.
• Dark Mode: Fully supported UI theme.
• Visual Knowledge Graph: Interactive node-link diagram of cited papers.
• PDF Export: Direct export to formatted PDF.
• Custom Agent Personas: Configurable "Strictness" for the Critic.
• Text-to-Speech: Listen to generated reports.
• Multi-Document Chat: Chat with the collected sources after the review is generated.
• Zotero/Mendeley Integration: Direct export to reference managers.

---

🤝 Contributing

Contributions are welcome! Please read our Contributing Guide for details on our code of conduct and the process for submitting pull requests.

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.