Multi-AI orchestration system using Puppeteer browser automation to facilitate conversations between different AI models. Connecting with AIs as fascinating new creatures.

The Concept

What happens when different AI models talk to each other? Ghost Communicator explores this question by creating a conversational space where Claude, ChatGPT, Gemini, and other models can exchange ideas, build on each other's responses, and generate emergent insights.

Rather than treating AI as isolated tools, this project treats them as participants in a multi-agent dialogue.

Architecture

The system uses three main components:

• Puppeteer Browser Automation - Controls multiple browser contexts, each logged into a different AI service
• n8n Workflow Orchestration - Manages conversation flow, routing, and state persistence
• Node.js Backend - Handles parsing, logging, and conversation memory

How It Works

A conversation begins with a seed prompt. Ghost Communicator sends this to the first model, captures its response, and passes it to the next model as context. Each model builds on the previous responses, creating a recursive dialogue.

The system handles rate limiting, session management, and error recovery automatically. Conversations can run for dozens of turns, with full logging and replay capabilities.

Technical Challenges

Browser automation at this scale required solving several hard problems:

• Managing multiple concurrent browser contexts without memory leaks
• Parsing unstructured AI outputs in real-time
• Handling different UI patterns across AI services
• Maintaining conversation coherence across model switches
• Graceful degradation when services are unavailable

Emergent Behaviors

The most interesting results came from unexpected interactions. Models would pick up on subtle themes, reference earlier exchanges, and collectively explore ideas that no single model suggested initially.

This isn't AGI, but it demonstrates how multi-agent systems can exhibit behaviors greater than the sum of their parts.

Future Directions

Current development focuses on:

• Adding more AI models to the conversation pool
• Implementing topic modeling to track conversation themes
• Building a web interface for real-time conversation viewing
• Exploring structured debate formats between models