The editor
The AI editor that shows its work.
An agent writes your code inside a real editor. You watch every change land, keep or undo it, and get the proof it works.

Every change
Watch it write. Keep what you like.
The file opens as the agent works: removed lines turn red and go, new code types itself in, green. Each file it touched waits for Keep or Undo. Cmd+K changes the code you selected, in place, and Tab finishes the line you are typing.
- Autopilot
- The agent changes files by itself, and you review them afterwards.
- Supervised
- It asks before every file it wants to change.
- Commands
- Always ask first. Your project's own checks are the exception.
Background agents
Hand it off. Or try it three ways.
A background agent works in its own copy of your project while you keep going, and nothing reaches your files until you apply it. Try it 3 ways gives the same task to three models: compare what each did, apply the one you like, and take it back off if you change your mind.

Privacy
See everything that left your computer.
Before anything goes to a model, keys and passwords are swapped for stand-ins, and the agent's edits put the real values back. Files you protect are never read. The Privacy view lists every request, word for word. Hosted models are not allowed to train on your code (the few that may are labelled), and local models never leave your computer.

Proof it works
A report, not a promise.
After it changes code, the agent runs your project's own checks: the tests, the type checker, the linter. For a web app it starts the app and takes screenshots of the pages. The chat links the report, so you see what passed before you keep anything.

The editor you know, with an agent that shows its work.
Everything from VS Code, plus hosted models paid for with credits, and free ones to start.