MoA Chat Documentation

What is MoA Chat?

MoA (Model of Agents) is a method that lets multiple AI agents (different LLMs) collaborate to generate a higher-quality response than a single model could.

MoA Chat implements a simple version of this architecture by:

• Querying several different models (LLM-A, LLM-B, LLM-C) at once.
• Combining their answers using another model (LLM-D, the aggregator).
• Delivering a smart, single, structured reply to the user.

How MoA Works (Visual)

Source: Together MoA Architecture Concept

How to Use

1. Click ⚙️ to open the configuration panel.
2. Select your preferred models for LLM-A, LLM-B, LLM-C, and Aggregator (LLM-D).
3. Type your message in the input box.
4. Press Send.
5. Watch multiple models collaborate for the best response!

Features

• Parallel querying of multiple free or premium models (via OpenRouter, Together, Grok, etc.).
• Structured prompts for each model to encourage quality responses.
• Aggregator model intelligently fuses outputs into one reply.
• Dynamic Light/Dark mode (Gruvbox Material Theme).
• Minimal, fast, and secure frontend with no API keys exposed.

Deployment

MoA Chat is optimized to run on Hugging Face Spaces or any platform that supports Python 3.11+, Flask, and Docker-based containers.

Requires setting your API keys via Hugging Face's Secrets system (never expose them to the frontend).

File Structure

• app.py — Flask backend server.
• llm/agents.py — Query and aggregation logic for MoA system.
• llm/model_config.json — Define available models and providers.
• templates/ — Contains index.html and docs.html.
• static/ — Contains style.css and script.js.

Credits

Made with ❤️ in Panamá by Until Dot. Inspired by Together's MoA architecture.