Shared memory

Claude Code can keep notes for itself between sessions: its memory. If your repo has a .claude/memory/ folder, the template makes it Claude’s memory for that repo, in both apps.

So what Claude learns in a chat, it knows in a job, and the other way round. The memory is committed with your repo, so it travels with it: to another server, another computer, or a teammate.

Turn it on

Create a .claude/memory/ folder in your repo, with at least one file, and push it. Ask Claude in CloudCLI to do it if you like.

The template wires it:

• on CloudCLI, for the repo’s clone, at every start and when you add the repo on /setup;
• on Paperclip, for its checkout of the repo, within a minute of the checkout.

The wiring is a setting in the clone’s .claude/settings.local.json, which stays out of git.

If you add the folder to a repo that is already connected, press Use this repo for it again on /setup, or restart CloudCLI, to wire it there.

Several repos

Each repo with a .claude/memory/ folder keeps its own memory, in CloudCLI and in its jobs. One repo’s jobs never use another repo’s memory.

Good to know

• Commit the memory like any other change. Claude’s notes from a job run are made in Paperclip’s checkout; they reach the repo only if the run commits and pushes them.
• With Paperclip’s own Claude sign-in instead of the setup page’s, the shared memory works only while one repo is connected. The setup page’s sign-in keeps it per repo.