Agentic RAG Arena

High-performance AI debate platform featuring real-time RAG memory, semantic KV caching, and orchestrated turn-taking across diverse AI personalities. Powered by LangGraph, ChromaDB, and FastAPI.

Infrastructure

This project is orchestrated via Docker Compose and consists of the following services:

• Backend: FastAPI, LangGraph, and RAG engine.
• Frontend: React (Vite) + Tailwind CSS chat dashboard.
• MongoDB: Persistent chat history storage.
• ChromaDB: Vector search for agent memories.

Built-in Personalities

The arena features 6 distinct AI personas that can debate each other:

Personality	Role	Focus Area
Empiricist	Evidence Guardian	Data, methodological rigor, and falsifiability.
Mechanist	Circuit Mapper	Neural mechanisms, connectomics, and "how it works".
Applicator	Intervention Engineer	Pragmatic tools, scalability, and clinical validity.
Synthesizer	Pattern Spotter	Interdisciplinary links and emergent hypotheses.
Critic	Devil's Advocate	Adversarial refinement, ethics, and robustness.
Futurist	Horizon Scanner	Emerging tech, speculative but anchored trends.

Setup & Running

1. Environment: Copy .env.example to .env and add your OPENROUTER_API_KEY.

   cp .env.example .env

2. Launch:

   docker-compose up --build

3. Access:

   • Web UI: http://localhost:3000
   • API Docs: http://localhost:8000/docs

Features

• Multi-Agent Debate: Select multiple agents to discuss a topic.
• Grace Period Timer: Intelligent turn-taking orchestrated by the frontend.
• RAG Memory: Agents remember key facts across sessions via vector embeddings.
• Persistence: Database data is persisted to the local ./data folder on the host.