about
About Stet
Stet is a regression gate for AI coding agents. It is a free, open-source command-line tool and MCP server that you point at an existing repository, and it does one job: it stops an AI agent from quietly breaking a feature that already worked.
Modern agents write good code fast, and still delete a validation rule, rename a field, or rewrite a component three tickets away from the one you asked about. Stet gives each working feature a short charter, records which files it owns and which of your tests prove it, and puts that context in front of the agent before it starts. After the change, stet check runs those tests and fails when a locked behaviour moved. Every PR gets a check that says kept or broken, in plain English, next to the diff.
It is built for small, fast teams: founders, CTOs and engineers of one to twenty people who are shipping with Claude Code, Cursor, Codex or Copilot and who cannot afford to re-read every diff. If you already have tests and a review habit, Stet widens the net; if you are a solo builder, it is the second pair of eyes you do not have.
Stet is built by Shamaz Khan, a founder and engineer who kept watching the same regression land twice in a week of heavy agent use. It is developed in the open against real repositories — nothing in it exists because a demo looked good.
how we build
local first
Stet is a command-line tool that runs in your repo. No account, no telemetry, and your code never leaves your machine.
uses your tests
It does not invent its own checks. It records which of your tests prove a feature works, then runs them after every AI change.
open source
The whole thing is MIT licensed and on GitHub. Read it, fork it, or run it in CI before you trust it.
no lock-in
You keep your editor and your AI tool. Stet sits beside them, not on top of them.