1 · Initialize once
Set up the repository
cd <your-repository>
ait initAIT sets up its own state and writes the workflow instructions into AGENTS.md.
Docs
A clear place to start with agent-written changes.
Initialize AIT once in a repository, then say what you want in plain language. Your coding agent follows the repository's instructions, and every change stays checkable from request to result.
Three steps to start
2 · Just ask
Say what you want
Add an accessible password-visibility control.
Keep the keyboard behavior and add tests.Say what you want and what to avoid. Your agent handles the code and the workflow.
3 · Check
Read the current state
ait status --jsonCheck where things stand whenever you want, in a form you can parse.
What AIT holds together
What you asked for
The agreed result and its limits travel with the work, so it can't drift halfway through.
The code
The agent works in a workspace owned by the Task, and nothing touches the target line until it's ready.
The evidence
Check results and the Snapshot are kept, so finished work stays inspectable.
See more of the workflow
Interactive Demo
Watch a feature request or a regression fix run start to finish.
Learn
What you decide, what the agent does, and what AIT records.
Remote Infrastructure
How the server and runner split the work when you need a remote setup.
Want the full detail?
The versioned technical documentation covers 1.1.1 concepts, workflows, commands, integrations, operations, and release boundaries, and every page links back to the exact public source it came from.