Did You Know Claude Code Remembers Between Conversations?

Claude Code has a feature called auto memory: while you work, Claude writes notes about your project to a local directory and reads them at the start of every new session. It's not native model memory — it's a file-based system that automatically injects context.

The effect is surprisingly useful: Claude "remembers" your project's specific build commands, debugging patterns it discovered, architecture decisions you explained to it — from last week or three weeks ago.

TL;DR The ~/.claude/projects/<project>/memory/ folder is Claude's persistent brain. Every session, it reads the first 200 lines of MEMORY.md. What it writes today, it knows next time.

Example ~/.claude/projects/<project>/memory/MEMORY.md:

# Memory

## Build commands
- `npm run build:staging` auto-loads staging env vars — no .env.staging needed
- Deploy takes ~4 min because it regenerates responsive images

## Debugging patterns
- CORS errors in local are always the Vite proxy, never the backend
- Payment test fails intermittently — known race condition, ticket #892

## Architecture decisions
- SQLite in production — single server, no connection pooling needed
- Translations live in `/lang/`, not `/locales/` — legacy from the original project

Managing memory

1. See what Claude remembers

/memory

Lists all files loaded into the session: CLAUDE.md files, rules, and auto memory files. Lets you open any file directly in your editor.

2. Edit the notes

Memory files are plain markdown. Correct wrong information, add context Claude didn't capture, or delete stale entries:

open ~/.claude/projects/$(basename $(pwd))/memory/MEMORY.md

3. Disable it if you don't want it

{
  "autoMemoryEnabled": false
}

Also available as an environment variable: CLAUDE_CODE_DISABLE_AUTO_MEMORY=1

Reference

Official docs: How Claude remembers your project