Now

In progress. This is what the next releases are made of.

The grade on the change itself

Stet already grades every incoming change and puts the risky ones at the top of the queue. Next it shows on the change: a changed number should not look like a fixed comma before you click it.

Review that earns its interruptions

Stet keeps count of how you actually review. Fifty trivial approvals in a row is a good argument for it to stop asking. Approving everything in under a second is worth saying out loud instead of playing along.

Sync you can trust

A push or a fetch that fails should say so. Today too much of it happens quietly in the background, and quiet is the wrong default for the part of the app that moves your work between machines.

Big rewrites stay readable

When an agent rewrites a whole document, the review pane should stay fast and legible instead of trying to draw every word of it.

Up for discussion

Six big directions. None committed, none ordered. This is exactly what founding members get to shape: tell me which one you'd use tomorrow, and which you never would.

Notion and Confluence

Google Docs and Sheets already link to a Stet file, in and out. The same bridge for the other two places teams keep their documents, so the review still happens here even when the doc lives there.

Editing on the web

Today the browser view reads, comments, and approves. Writing in it is the obvious next ask, and the hard part is doing it without quietly turning Stet into a cloud editor with a local backup.

Any AI, not just the ones with a terminal

Copy a document for an AI, paste its answer back as a proposal. Or connect a hosted workspace to a tool directly, so a chat assistant can propose a change without touching your disk. Review stays exactly as strict either way.

HTML documents

HTML is the markdown of slides, one-pagers, and visual content, and agents already write it. Stet would review it the same way it reviews prose: every edit a proposal, rendered so you review the slide, not the tags.

Windows support

The stack makes it feasible. Demand decides when. If Windows is the one thing stopping you, this is the easiest vote to cast.

Bring your own API key

Today Stet never calls an AI itself; you bring your own agent. With your API key, it could run the agent for you: press Send on a note and the edit comes back as a proposal. No terminal, no separate tool. Your key, your choice of model, and review stays exactly as strict.

Argue with this list.

Everything here came from testers or from watching people use Stet. If something you need is missing, or in the wrong bucket, tell me. Email works, and so does the feedback button in the app.