# Keep what’s read with the repo

> So others don’t read it again. Every page of kelve's docs is listed at https://kelve.ai/llms.txt.

What your agent reads of a project can be kept in the project itself. Then no one who works on it has their agent read the same file twice.

## Turn it on or off

**Settings**, **Reading**, **Keep what's read in the repo**. The setting is on to start.

- **On:** what's read is kept in the project's `.kelve/`, to commit with the work. Someone else who opens the project in kelve starts with it.
- **Off:** it's kept by kelve alone.

What's read is kept in the project only in a git repository. Outside one, kelve keeps it alone either way.

## What's in it

For each file read, kelve keeps each part's ideas, with a line on how the work uses each. It also keeps what the reading looked for. There's one file for each version of a file, named by git's own hash of what's in it. So a file that changes is read again, and only its changed parts go to your agent.

What's read is kept with what it looked for. So a teammate who asks kelve to mark less reads nothing new. A teammate who asks for more reads only that. That way, a team's reading only grows.

The way in is kept too. That's the order to read the project in, and a line on what each folder is for.

A file read plainly, and a folder's page, are kept by kelve alone, never in the project.

## The rest of `.kelve/`

| What | What's in it |
|---|---|
| `.kelve/read/` | What's read, when this is on |
| `.kelve/way.json` | The way in, and each folder's line, when this is on |
| `.kelve/notes/` | Notes kept with the work, and what your agent found out for you |

kelve never counts `.kelve/` as part of a change. None of it goes to the shared graph. What's in `.kelve/` tells anyone who can read the repo which ideas its files rest on, as the code does.
