Skip to content

Home

Jump to bottom
yohn1985 edited this page Jun 15, 2026 · 2 revisions

Jarvis Wiki

A standalone, self-improving autonomous agent that lands on a machine, sets itself up, learns its world, and works for you.

⚠️ Early beta. Jarvis reads, reasons, and can act on a real machine. Run it on a box you trust and keep it in the default shadow / propose-only mode until you've watched it.

Start here

  • Getting Started — install, log in, connect a brain, say hello.
  • Architecture — the wake→perceive→decide→act→reflect loop and how the pieces fit.
  • Configuration — config.yaml, secrets, model routing, what it may do on its own.
  • Skills — using the shipped skills and writing your own.
  • Safety — propose-only defaults, fail-closed verification, the self-mod seatbelt, the access token.
  • Contributing — dev setup, the fitness gate, how to add adapters/skills.

What makes it different

Jarvis is not a chatbot wrapper. It's a bounded loop with a real mind:

  • It rebuilds its own context every wake from durable memory.
  • It is skeptical by design — it doesn't accept a conclusion until that conclusion has survived an adversarial red-team, and it routes what it can't verify to you instead of guessing.
  • It bootstraps itself (pure Python, no AI needed for setup), autodiscovers its environment, and keeps a knowledge base.
  • Its most dangerous power — editing its own code — runs behind a tamper-proof fitness gate + independent red-team + version archive it can roll back to.

Links

Clone this wiki locally