Helping Multi-Agent AI Experimentation

Inspired by some work from a group at Stanford on building a lab from AI agents, I’ve been experimenting with multi-agent AI conversations and workflows. But, because the space (at least to me) has seemed more focused on building more capable agents rather than coordinating and working with more agents, the existing tools and libraries have made it difficult to carry out fast experiments.

To facilitate some of my own exploration work, I built what I’m calling the Multi-Agent ChatLab — a browser-based, completely portable setup to define multiple AI agents and facilitate conversations between them. This has made my experimentation work vastly simpler and I hope it can help others.

• Design & Architecture: Read the full technical breakdown and system instructions in the Multi-Agent ChatLab Portfolio Article.
• Open Source Code: Grab the single-file source directly from the GitHub Repository.
• Direct Web Access: Launch the standalone applet in a separate window at /multiagent-chatlab.html.

And, to show off the tool, and for your amusement (and given my love of military history), here is a screengrab from the tool where I set up two AI Agents — one believing itself to be Napoleon Bonaparte and one believing itself to be the Duke of Wellington (the British commander who defeated Napoleon at Waterloo) — and had them describe (and compare!) the hallmarks of their military strategy.