Brain · Auto-memory

Auto-memory

Claude Code's own native auto-memory for each repo: the ~/.claude/projects/<project>/memory/ that Claude Code writes itself, with a 200-line / 25KB session-load budget meter (only that much of MEMORY.md loads at session start) plus the on-demand topic files. The counterpart to your authored .claude/brain.

Start here if you want the lessons RepoOps proposes from your own sessions, beside the memory Claude Code keeps on its own: the Brain reflection guide. Both open from Memory.

See it in motion

Where to find it

  • Localhost: /auto-memory.html?repo=<id>
  • API: GET /api/auto-memory?repo=<id>
  • Keyboard: ⌘ K then auto memory
  • Navigation: Memory in the sidebar, then Auto-memory under All tools, in the Lessons, wiki and patterns group

What it does for you

See the memory Claude Code keeps for you, and its load budget.The tab reads the memory/ folder beside the same project directory the telemetry reader resolves for transcripts, and reports how many lines and bytes of MEMORY.md load: the session load truncates at whichever budget hits first, 200 lines or 25KB. Past either budget the meter says so and suggests trimming the index or moving detail into topic files. Read-only, no LLM call; a repo with no auto-memory reads empty rather than failing.

Read more

Last updated