brAIn lets you watch a network of tiny AI agents work, not just chat

Most AI agent frameworks funnel everything through a single chat window, with a background loop waking the model on a timer. brAIn takes the opposite approach: many small agents, each handling one job, with a live dashboard showing every node and every message flowing between them in real time. Nodes sit on a message bus and only wake up when a message they care about arrives, so idle agents cost nothing.

Each node can carry its own UI, and that UI stays reachable even if the node runs on a different machine. The system is distributed — a second machine can plug into the same bus and host nodes on one shared canvas. Nodes can run entirely offline on local models through Ollama, or be pointed at a coding CLI like claude, codex, or gemini instead.

Demo templates include a perception stack combining voice, gaze, intent, and text-to-speech that listens to a room and only wakes the AI when someone looks at the camera and speaks — answering out loud with no wake word needed. There are also bridges for Telegram, WhatsApp, and Discord, plus a small marketplace of installable nodes such as vector-search memory and a few demo games. The README includes two honest disclaimers, including that a lot of the code was written with AI assistance.