The source of truth — for you and your agent
How Ctrl AI works, the full CLI, the contract your agent follows, and an honest account of what the lock guarantees today. Every page is generated from the repo, so it stays in sync with the code.
How Ctrl AI works
The shared mental model: capability / part / adapter / seam, the lock, the spec-loop, the /ctrl panel — and an honest account of what's guaranteed today vs roadmap. Read this first.
Read →Quickstart
One command to a running, locked backend on a zero-config DB → add a capability → provision a real database → push (CI) → deploy.
Read →Agent guide
The boundary contract for an AI agent in a locked repo: what's read-only, how it's enforced, the sanctioned verbs, and what to do at the wall.
Read →CLI reference
Every ctrlai command, grouped by lifecycle, grounded in the real CLI.
Read →Trust & security
The honest threat model: what the lock catches today (drift, mechanically), what it doesn't yet (tamper-proofing — the cross-repo tier), and exactly what “verified” means.
Read →Troubleshooting
CI went red — which signal (BOUNDARY / IMPORTS / VENDORS / HAND_ROLLED / …), which command, which line. Keyed by what you see.
Read →Wiring up an AI agent? llms.txt is a single machine-readable cold-load manifest of the boundary rules and the capability catalog.