sharedbrain

One memory for every Claude surface. Claude Code, Claude Desktop (Chat + Cowork), claude.ai — same brain, same memories.

Tell Claude something in Desktop Chat. Ask about it in Code. It just knows. No "remember this" needed — Claude saves everything important automatically.

Install

npx sharedbrain install

That's it. Wires into Claude Code and Claude Desktop automatically. Restart Desktop to pick it up.

What it does

Claude gets five memory tools it calls on its own — no prompting needed:

| Tool | What it does | |---|---| | remember | Saves facts, preferences, decisions, context | | recall | Pulls up a specific memory or lists recent ones | | search | Full-text search across everything it's remembered | | forget | Deletes a memory | | brain_stats | Shows what's in the brain |

As you talk, Claude automatically saves things — opinions, personal details, project decisions, preferences. Next conversation, different surface, it loads what it knows and picks up where you left off.

All memories live in one local SQLite file: ~/.sharedbrain/brain.db. No cloud, no API keys, no network required.

Extend to claude.ai web + mobile

Claude Code and Desktop work out of the box. To also connect claude.ai web or mobile:

sharedbrain token create my-token
sharedbrain serve-http --port 3000
cloudflared tunnel --url http://localhost:3000

Add the tunnel URL + token as a Connector in your claude.ai settings.

CLI

sharedbrain install                  # set up Claude Code + Desktop
sharedbrain search "<query>"         # search memories from terminal
sharedbrain stats                    # see memory counts
sharedbrain token create|list|revoke # manage HTTP auth tokens

Security

Local mode has no network and no auth — it runs as your OS user. HTTP mode uses bearer tokens hashed with SHA-256 and stored in the same brain.db. Tokens are revocable and shown only once at creation.

brain.db holds everything Claude remembers about you. Back it up. Don't commit it.

License

MIT