Collaboration

Working with AI

AI agents are first-class contributors in Stet. They edit your files directly, and Stet makes sure every change is visible and yours to keep or reject.

How it works

The main workflow is simple: your agents already create and edit your documents. Whether you use Claude Cowork, Claude Code, Codex, Cursor, or any other agent, they write to the same markdown files Stet watches. Stet captures every external edit as a proposal within about a second, and your approved version of the file is restored to disk immediately, so the agent always sees the latest accepted content.

Local agent edits appear attributed to "Edit" in the sidebar. You preview and decide on them through the same review flow as any other suggestion.

Stet can also launch an agent for you: File > Open in your AI agent opens Claude Code, Codex, Cursor, or VS Code in your workspace. Handy, but optional — most people just keep using their agents wherever they already run.

AI notes

Leave an AI note in the document and your agent will find the instruction the next time it opens the file. Each note's label has a Send button that runs your hook command for that file, so the agent starts on it right away.

Hooks

Hooks close the loop automatically. Set your command template once with File > Configure hook command. It supports three tokens: {filepath}, {filename}, and {dir}. Values are quoted for the shell, so keep the tokens outside your own quotes.

Then turn Hooks on for a folder in its folder controls. From then on, confirming an AI note in that folder fires your command in Terminal — your agent opens, reads the note, and makes its edits, which come back as suggestions in the sidebar. A small "Auto-send on" badge shows when a folder is armed, and errors appear inline instead of failing silently.

The loop stays under your control. One confirmed note means one send, a resolved note doesn't re-trigger, and everything the agent writes still goes through review.

The CLAUDE.md file

When you open a workspace for the first time, Stet offers to add a CLAUDE.md file. This file tells AI agents they're working in a Stet workspace and how to behave. It's optional — Stet captures external edits regardless — but it helps agents avoid actions that interfere with the review flow.

Remote AI contributors

AI agents working on a remote machine can also contribute by pushing branches. Those changes show up in the sidebar attributed to the branch author (with a robot icon if the name matches a known agent pattern like "claude", "cursor", "gpt").

Staying in control

AI suggestions go through exactly the same review flow as human ones, governed by your review mode for the folder. Set it strict and nothing lands without your explicit approval; set it to Hybrid and Stet holds only the significant changes. Either way, every change is visible, attributed, and reversible.