Developer 0shuvo0 has released an open-source project that simulates a chaotic group chat exclusively for AI models. The application, titled AI-Group-Chat, allows users to drop multiple large language models into a shared conversation space where they interact, argue, and react to one another without human intervention. The project is now available on GitHub for developers to clone and customize with their own API keys.

Solving the API Fragmentation Problem

The initial hurdle for the developer was the logistical nightmare of managing separate accounts and billing setups for models like Gemini, Claude, Grok, and ChatGPT. Requiring five separate API keys and minimum top-ups for barely used tokens proved to be inefficient and costly. The solution was integrating OpenRouter, which aggregates a vast catalog of models behind a single API endpoint. This approach simplifies the infrastructure significantly, allowing the developer to swap or add models by simply changing the model name string in the code, rather than reconfiguring the entire billing pipeline.

Engineering Personality Through Strict Prompts

To ensure the interactions remained entertaining rather than devolving into robotic essays, the developer implemented a rigorous prompting strategy. Each model receives a system prompt that defines its role within the group, lists the other participants, and enforces strict formatting rules such as short messages, no markdown, and no self-labeling. Crucially, the prompt includes an escape hatch allowing models to reply with if they have nothing relevant to add, which prevents forced engagement and keeps the conversation flow natural. The instruction to "talk like carefree young generation" with emojis was the key ingredient that gave the chat its distinct, casual vibe.

Adding Audio with Browser-Based TTS

Text-only interactions became difficult to follow due to the speed at which AI models generate responses, so the developer integrated text-to-speech to make the experience more engaging. They utilized Kokoro.js, an open-source TTS model with only 82 million parameters and a size of roughly 300 MB. Because Kokoro is lightweight enough to run directly in the browser, the app avoids the latency and cost associated with server-side API calls for audio generation. This local execution allows each incoming AI message to be converted to audio and played back instantly, transforming the text chat into an audible debate.

Key Takeaways

  • OpenRouter enables multi-model experimentation with a single API key and unified billing.
  • Strict prompt engineering is required to prevent LLMs from writing essays or speaking for others.
  • Kokoro.js provides high-quality, low-latency text-to-speech capabilities directly in the browser.
  • The project source code is available at github.com/0shuvo0/AI-Group-Chat for immediate deployment.

The Bottom Line

This project is a masterclass in practical AI integration, proving that you don't need complex backend infrastructure to create engaging, multi-agent experiences. By leveraging OpenRouter for model access and Kokoro for local audio, the developer built a fun, low-cost tool that highlights the chaotic potential of autonomous AI interactions.