Claude Code Has Been Taking Notes on Your Project — Here's Where to Find Them

Since v2.1.59, Claude Code has been writing its own project notes to ~/.claude/projects/*/memory/ between sessions -- use /memory to see what it remembers and steer what it saves.

When you correct Claude, share a debugging insight, or explain how a module works, Claude Code can now retain that knowledge across sessions without you updating CLAUDE.md yourself. Auto memory is on by default and has been quietly building a per-project knowledge base since February 2026.

Where the notes live

~/.claude/projects/home-user-myproject/
└── memory/
    ├── MEMORY.md        ← index; first 200 lines load on every new session
    ├── debugging.md     ← troubleshooting insights Claude picked up from your corrections
    └── architecture.md  ← key files, module relationships, naming conventions

The first 200 lines of MEMORY.md (or 25 KB) load automatically at session start. Topic files like debugging.md are read on demand when MEMORY.md references them and the current task looks relevant.

What Claude tracks

• Build and test commands — after you tell Claude "the API tests run with pnpm test:api", it saves that so the next session doesn't ask
• Debugging wins — "the auth 401 was a missing httpOnly flag on the cookie; fixed in src/middleware/auth.ts" gets filed under debugging.md
• Architecture notes — key file relationships, module boundaries, third-party library quirks you've surfaced

Browse, edit, and steer it

Run /memory in any Claude Code session to:

• Open the memory folder in your editor
• Browse MEMORY.md and topic files
• Delete stale or incorrect notes
• Toggle auto memory off for the current project

To steer Claude's notes during a session, be explicit:

"Remember that this project's migrations run with pnpm db:migrate, not prisma migrate."
"Note that all API handlers in src/api/ must return a typed ApiResponse<T> wrapper."

Claude writes the note mid-session and reloads it the next time.

CLAUDE.md vs. auto memory

	CLAUDE.md	Auto memory
Who writes it	You	Claude
Purpose	Rules, conventions, always-on guidance	Accumulated project knowledge
Loaded	Every turn (whole file)	MEMORY.md index on start; topics on demand
Synced	Yes (committed to repo)	No (local only, per-machine)

The two work together: CLAUDE.md is your conventions layer; auto memory is Claude's accumulated knowledge layer. Neither replaces the other.

One gotcha: auto memory is per-machine. Notes from a desktop session do not appear in a cloud or laptop session. If something must survive across machines, commit it to CLAUDE.md.

Sources: How Claude remembers your project — Claude Code Docs · Claude Code Auto Memory explained (memoryplugin.com) · What Is Claude Code Auto Memory (israynotarray.com)