Multi-Agent Research Assistant

A production-style, full-stack AI research workspace that coordinates a four-stage multi-agent pipeline — from live web search to final critic review — with a polished dark UI and real-time progress streaming.

🔗 Live Demo: multi-agent-research-assistance.netlify.app

Enter a research topic and the pipeline kicks off four sequential agents:

1. Search Agent — queries the web via Tavily Search and retrieves relevant results
2. Reader Agent — scrapes and extracts content from the returned URLs using BeautifulSoup
3. Writer Chain — synthesizes the scraped content into a structured report using Mistral AI via LangChain
4. Critic Chain — reviews the generated report and provides structured feedback

The frontend streams each stage's progress in real time via a Next.js API proxy, so the browser never talks to FastAPI directly.

• Next.js 15 (App Router)
• TypeScript
• Tailwind CSS
• Framer Motion for animations
• shadcn/ui-style components

• FastAPI (Python)
• LangChain orchestration
• Mistral AI for LLM inference
• Tavily Search for web retrieval
• BeautifulSoup for HTML scraping

• Dark glassmorphism AI workspace layout
• Live streaming pipeline progress via newline-delimited JSON
• Collapsible execution cards per agent stage
• Typing-style animated report reveal
• Critic feedback panel rendered separately from the report
• Agent status sidebar with compact stage indicators
• Responsive design, mobile-friendly
• Next.js API proxy — no direct browser-to-FastAPI calls
• CORS and environment variable support throughout

.
├── backend/
│   ├── __init__.py
│   ├── .env.example
│   ├── agents.py          # Agent definitions
│   ├── main.py            # FastAPI app
│   ├── pipeline.py        # Orchestration logic
│   ├── schemas.py         # Pydantic models
│   ├── settings.py        # Config / env loading
│   └── tools.py           # Search & scraping tools
├── frontend/
│   ├── app/
│   │   ├── actions.ts
│   │   ├── api/research/
│   │   │   ├── route.ts          # Standard endpoint proxy
│   │   │   └── stream/route.ts   # Streaming endpoint proxy
│   │   ├── globals.css
│   │   ├── layout.tsx
│   │   └── page.tsx
│   ├── components/
│   ├── lib/
│   ├── services/
│   ├── next.config.ts
│   ├── package.json
│   └── tsconfig.json
├── agents.py              # Root-level copies (legacy)
├── pipeline.py
├── tools.py
├── requirements.txt
└── .env.example

Returns service status. Useful for uptime checks and deployment health probes.

Runs the full pipeline and returns a structured JSON response:

{
  "search_results": "...",
  "scraped_content": "...",
  "report": "...",
  "feedback": "...",
  "steps": []
}

Streams pipeline progress as newline-delimited JSON events, then ends with the same report payload. This is the endpoint the frontend uses for live stage updates.

Copy backend/.env.example to backend/.env and fill in:

MISTRAL_API_KEY=your_mistral_key
TAVILY_API_KEY=your_tavily_key
MISTRAL_MODEL=mistral-medium   # or your preferred model
CORS_ORIGINS=http://localhost:3000

Copy frontend/.env.example to frontend/.env.local and set:

BACKEND_API_URL=http://127.0.0.1:8000

python -m venv .venv

# Windows
.venv\Scripts\activate

# macOS / Linux
source .venv/bin/activate

pip install -r requirements.txt

cd frontend
npm install

uvicorn backend.main:app --reload --host 0.0.0.0 --port 8000

cd frontend
npm run dev

Open http://localhost:3000 in your browser.

• Valid Mistral and Tavily API keys are required before any research run will succeed.
• The root-level agents.py, pipeline.py, and tools.py are legacy files left in place for reference; the canonical versions live inside backend/.
• The streaming UI is powered by a Next.js route handler acting as a proxy, keeping all FastAPI calls server-side.