ResearchOS — Autonomous Multi-Agent Intelligence Pipeline

ResearchOS is an autonomous multi-agent research engine designed to perform deep, high-fidelity web research, content extraction, synthesis, and critical auditing in real time. Powered by Groq high-speed inference (Llama 3.3 70B Versatile), Tavily Web Search, and LangChain / LangGraph, ResearchOS orchestrates a team of specialized AI agents working sequentially in a Directed Acyclic Graph (DAG) pipeline.

---

Overview & Key Capabilities

• Autonomous Multi-Agent Workflow: Sequential collaboration between 4 specialized agents (Search, Reader, Writer, Critic).
• Ultra-Fast Inference: Driven by Groq's low-latency Llama 3.3 70B model for fast, comprehensive research reports.
• Real-Time Live Web Discovery: Uses Tavily AI Search API to fetch high-authority web pages and raw search snippets.
• Deep Web Scraping & Noise Reduction: BeautifulSoup4 extracts high-signal text while stripping script, navigation, and advertisement clutter.
• Adversarial Quality Auditing: Dedicated Critic agent performs strict evaluations, outputting numerical quality scores (0-10), strengths, and key areas for improvement.
• Modern Interactive Dashboard: High-aesthetic React frontend featuring live agent phase tracking, interactive DAG visualization, execution logs, research history, and clean Markdown rendering.
• Production Ready Deployment: Pre-configured render.yaml for instant cloud deployment on Render.

---

System Architecture

ResearchOS utilizes a Directed Acyclic Graph (DAG) workflow where data passes through distinct cognitive stages. Each agent enriches the system state before passing control to the next specialist.

graph TD
    User([User Research Query]) -->|1. Research Topic| SearchAgent[Search Agent]
    SearchAgent -->|2. Web URLs & Snippets| ReaderAgent[Reader Agent]
    ReaderAgent -->|3. Clean Scraped Content| WriterChain[Writer Chain]
    WriterChain -->|4. Structured Markdown Report| CriticChain[Critic Chain]
    CriticChain -->|5. Verified Report + Score & Audit Notes| UI[Interactive Research Dashboard]

    subgraph Agent Workforce
        SearchAgent
        ReaderAgent
        WriterChain
        CriticChain
    end

Loading

---

Agent Breakdown

Agent / Chain	Role	Tools / Model	Responsibilities & Outputs
Search Agent	Web Explorer	Tavily Search API Llama-3.3-70b-versatile	Queries live web search engines, discovers high-authority sources, and extracts search result metadata.
Reader Agent	Content Scraper	BeautifulSoup4 Requests	Parses top target URLs, removes ads/script noise, and isolates core signals into clean text data blocks.
Writer Chain	Synthesis Expert	LangChain Expression Language Llama-3.3-70b-versatile	Synthesizes scraped content into a structured executive report complete with findings, conclusions, and cited sources.
Critic Chain	Quality Auditor	LangChain Expression Language Llama-3.3-70b-versatile	Conducts adversarial evaluation of the draft report, assigning a score (0-10), listing strengths, and noting improvement areas.

---

Repository Structure

Multi-agent-research-system/
├── backend/
│   ├── agents.py           # Definition of Search, Reader, Writer, and Critic agents/chains
│   ├── main.py             # FastAPI server with CORS, ThreadPoolExecutor, and API routes
│   ├── pipeline.py         # Sequential execution workflow (run_research_pipeline)
│   ├── tools.py            # Custom tools: web_search (Tavily) & scrape_url (BS4)
│   ├── requirements.txt    # Python backend dependencies
│   ├── README.md           # Backend specific documentation
│   └── .env.example        # Environment variable template
├── frontend/
│   ├── src/
│   │   ├── App.jsx         # React interactive dashboard & pipeline visualization
│   │   ├── App.css         # Component-level styling & custom scrollbars
│   │   ├── index.css       # Global design tokens, dark theme & responsive layouts
│   │   └── main.jsx        # React application entry point
│   ├── package.json        # Frontend dependencies (React 19, Framer Motion, Lucide)
│   ├── vite.config.js      # Vite build configuration
│   └── README.md           # Frontend setup guide
├── render.yaml             # Render deployment configuration specification
└── README.md               # Main project documentation

---

Quick Start & Setup

Prerequisites

• Python 3.10+
• Node.js 18+ & npm
• Groq API Key: Get a free key from Groq Console
• Tavily API Key: Get a free key from Tavily AI

---

1. Environment Setup

Create a .env file inside the backend/ directory:

cp backend/.env.example backend/.env

Add your API keys to backend/.env:

GROQ_API_KEY=gsk_your_groq_api_key_here
TAVILY_API_KEY=tvly-your_tavily_api_key_here

---

2. Backend Setup & Local Server

1. Navigate to the backend directory:

   cd backend

2. Create and activate a Python virtual environment:

   • Linux / macOS:

     python3 -m venv .venv
     source .venv/bin/activate

   • Windows (PowerShell):

     python -m venv .venv
     .\.venv\Scripts\Activate.ps1

3. Install required Python packages:

   pip install -r requirements.txt

4. Launch the FastAPI backend server:

   python main.py

   The API will be available at http://localhost:8000. API documentation is automatically accessible at http://localhost:8000/docs.

---

3. Frontend Setup & UI Launch

1. Open a new terminal window and navigate to the frontend directory:

   cd frontend

2. Install Node dependencies:

   npm install

3. Start the Vite development server:

   npm run dev

4. Open your browser and navigate to http://localhost:5173.

---

API Reference

POST /api/research

Executes the autonomous multi-agent research pipeline for a given topic.

Request Body

{
  "topic": "Latest advancements in quantum computing hardware 2026"
}

Response Structure

{
  "search_results": "Raw web search snippets and metadata...",
  "scraped_content": "Extracted full-text article content...",
  "report": "# Executive Summary\n\nQuantum computing has made significant strides...",
  "feedback": "Score: 9/10\n\nStrengths:\n- Thorough coverage of superconducting qubits...\n\nAreas to Improve:\n- Could expand on neutral atom approaches.\n\nOne line verdict: Excellent executive summary."
}

---

Built With

• Framework: FastAPI + Uvicorn
• AI / LLM Orchestration: LangChain & LangGraph
• LLM Provider: Groq API (Llama-3.3-70b-versatile)
• Web Search API: Tavily AI
• Web Parser: BeautifulSoup4 + Requests
• Frontend Core: React 19 + Vite
• Animations & Icons: Framer Motion & Lucide React
• Markdown Rendering: React Markdown + Remark GFM

---

Deployment

Render Deployment (Backend)

The repository includes a ready-to-use render.yaml specification:

1. Connect your GitHub repository to Render.
2. Select Blueprint deployment.
3. Set the required environment variables (GROQ_API_KEY and TAVILY_API_KEY) in the Render Dashboard.
4. Deploy!

---

License

Distributed under the MIT License. See LICENSE for details.