It's an AI agent with a folder full of markdown files it can edit itself. It has tools to edit files and run shell commands.
Its folder ("workspace") contains a mix of arbitrary files you work on together, and special system files that affect the bot's behavior: SOUL.md, HEARTBEAT.md, and the memory/ folder.
You DM it over Signal or Telegram and have it manage these files. SOUL.md affects its personality. HEARTBEAT.md is used every 30 minutes by the Node.js process to spawn a new temporary AI session, enabling it to do things all day long. To avoid repeating work, it stores work it already did in its memory/ folder.
tl;dr it's a living AI agent. You give it access to a full computer, so it's a living AI agent with a full computer system just for itself. This has been possible for over 1 year now already, and many of us have thought about it, but I think nobody did it because it sounds like a very bad idea to just like AI roam free like that with no security guardrails. But obviously it was inevitable.
Login to reply
Replies (1)
one of the first things i did with llms years ago was the "heartbeat" concept, just in the stupid web chat of chatgpt with no local files. even with that crippled of a context it was already a crazy phase change. i cannot believe it took this long for the strategy to be normalized broadly.
it is very eye-opening to see the ai appear to "act on its own", unprompted.
"heartbeat" every second....