GUS Documentation
A secure agent you can actually read about.
GUS is an open-source AI agent that runs on a computer you control — with hard spend caps the model can't touch, credentials in a vault it can never read, and every action scanned, approved, and receipted. These docs cover all of it, in plain language.
Start here
Install GUS in about ten minutes and run the five-step setup wizard.
Voice — Call GUS
How voice calls work, which speech engines do what, and what to check when a call goes wrong.
Settings, explained
Every control-panel setting in plain language — the same text as the in-app help, so the two never drift apart.
Safety model
Governor modes, the floors that never relax, and the threat model: what GUS defends against and what it doesn't.
Channels & packs
Telegram, panel chat, delegates, the Gmail pack, and publishing releases.
Mac app
The native GUS.app bridge: permissions, the DMG, and how updates work.
Release notes
What landed in each version, with the test count recorded for every release.
The short version
GUS is built on one assumption: the model is not trusted. Everything a model reads is untrusted input, so no safety property depends on the model behaving. Spend caps are enforced by a Governor outside the model. Secrets live in a vault with no read API. Risky actions park for a human approval that is bound to the exact action and works once. And every finished job leaves a receipt — what ran, what it cost, what the security scan found — verified by a checker that is not the agent itself.
New here? Start with the install. Want the guarantees first? Read the safety model.