Agent Discuss

A multi-agent AI discussion platform where users can set up LLM participants to discuss topics autonomously.

Features

• Create AI agents with custom system prompts and configurations
• Define conversation flows as graphs with event nodes (generate, evaluate, decide, summarize)
• Watch multiple agents discuss topics in real-time via WebSocket
• Upload documents for context (RAG integration with ChromaDB)
• Dark/light mode support
• Pause/resume discussion controls

Tech Stack

• Frontend: React 19 + Vite + shadcn/ui + Tailwind CSS + React Flow
• Backend: FastAPI + LangGraph + Celery
• Database: Supabase (PostgreSQL + Auth)
• Message Queue: Redis
• Vector Store: ChromaDB
• LLM: vLLM (self-hosted, OpenAI-compatible)

Quick Start (Local Development)

Prerequisites

• Docker and Docker Compose v2+
• A Supabase account and project (free tier works)
• An LLM API endpoint (vLLM, OpenAI, or any OpenAI-compatible API)

1. Clone and Configure

# Clone the repository
git clone <repository-url>
cd agent-discuss

# Copy environment file
cp .env.example .env

2. Configure Environment

Edit .env with your credentials:

# Get these from Supabase Dashboard > Settings > API
SUPABASE_URL=https://your-project.supabase.co
SUPABASE_KEY=your-anon-key
SUPABASE_SERVICE_KEY=your-service-role-key
SUPABASE_JWT_SECRET=your-jwt-secret

# Frontend needs these too
VITE_SUPABASE_URL=https://your-project.supabase.co
VITE_SUPABASE_ANON_KEY=your-anon-key

3. Set Up Supabase Database

1. Go to your Supabase project's SQL Editor
2. Copy and run the contents of backend/scripts/init.sql
3. This creates all required tables with Row Level Security policies

4. Start the Application

# Start all services with Docker Compose
docker compose up -d

# Watch the logs
docker compose logs -f

5. Access the Application

Service	URL
Frontend	http://localhost:5173
Backend API	http://localhost:8000
API Documentation	http://localhost:8000/docs
ChromaDB	http://localhost:8001

6. Configure an LLM Provider

1. Register/login at http://localhost:5173
2. Go to Settings > LLM Providers
3. Add your LLM connection:
   • Name: My vLLM / OpenAI / etc.
   • Base URL: Your API endpoint (e.g., http://host.docker.internal:8080/v1 for local vLLM)
   • API Key: Your API key
   • Model: Model name to use

Development Without Docker

Backend (FastAPI)

cd backend

# Create virtual environment
python -m venv venv
source venv/bin/activate  # or `venv\Scripts\activate` on Windows

# Install dependencies
pip install -r requirements.txt

# Run the server
uvicorn app.main:app --reload --port 8000

Frontend (React)

cd frontend

# Install dependencies
npm install

# Start dev server
npm run dev

Redis (required for Celery)

# Using Docker
docker run -d -p 6379:6379 redis:7-alpine

# Or install locally
brew install redis  # macOS
sudo apt install redis-server  # Ubuntu

Celery Worker

cd backend
celery -A app.tasks.celery_app worker --loglevel=info

Project Structure

.
├── backend/                  # FastAPI backend
│   ├── app/
│   │   ├── api/routes/       # REST endpoints
│   │   ├── agents/           # LangGraph orchestration
│   │   │   ├── state.py      # State definitions
│   │   │   ├── graph.py      # Graph builder
│   │   │   ├── nodes.py      # Node implementations
│   │   │   └── context.py    # Context management
│   │   ├── services/         # Business logic
│   │   │   ├── websocket_manager.py
│   │   │   ├── vectorstore.py
│   │   │   └── document_processor.py
│   │   └── tasks/            # Celery tasks
│   └── scripts/              # Database scripts
│
├── frontend/                 # React frontend
│   └── src/
│       ├── components/
│       │   ├── ui/           # shadcn components
│       │   ├── discussions/  # Discussion views
│       │   ├── messages/     # Message display
│       │   ├── graph-editor/ # React Flow editor
│       │   ├── agents/       # Agent management
│       │   └── documents/    # Document uploads
│       ├── hooks/            # React hooks
│       ├── stores/           # Zustand stores
│       └── pages/            # Route pages
│
├── docker-compose.yml        # Docker orchestration
└── .env.example              # Environment template

Conversation Graph Nodes

The application supports these node types in conversation graphs:

Node Type	Description
Start	Entry point of the conversation
Generate	Agents take turns generating content
Evaluate	Agents vote/score on criteria
Decision	Conditional branch based on evaluation
Summary	Compress context to manage token limits
End	Final summary and conclusion

Troubleshooting

Connection refused errors

Make sure all services are running:

docker compose ps

WebSocket not connecting

Check that:

1. Backend is running on port 8000
2. VITE_WS_URL is set correctly in .env
3. No firewall blocking WebSocket connections

LLM provider test fails

1. Verify the base URL is accessible from Docker (use host.docker.internal for local services)
2. Check API key is correct
3. Ensure model name matches exactly

Documents not processing

1. Check Celery worker logs: docker compose logs celery-worker
2. Verify ChromaDB is running: docker compose ps chromadb
3. Check document status in the UI (should show "Processing" then "Ready")

Reset everything

# Stop and remove all containers and volumes
docker compose down -v

# Rebuild from scratch
docker compose up -d --build

API Documentation

Interactive API documentation is available at http://localhost:8000/docs when the backend is running.

Key endpoints:

• POST /api/v1/auth/login - Login
• GET /api/v1/agents - List agents
• GET /api/v1/discussions - List discussions
• POST /api/v1/discussions/{id}/start - Start a discussion
• WS /api/v1/ws/discussions/{id} - WebSocket for real-time updates

License

MIT