Docs · My Brain

Getting started with My Brain

My Brain is a brain you own that follows you - across every AI tool, every session, every team, and every job. It is keyed to you, not to a vendor’s model. It reasons over your own notes and shows its sources, every write lands as a proposal you review, and every external read is scoped, revocable, and audited. This guide is the five ways in.

1. Ask it locally (start here)

Nothing to activate - this works on your own machine today.

Run the RepoOps desktop app / dashboard (npm run restart, then open http://localhost:4000/).
Open the My Brain tab (also available as a section on the Continuity tab).
In Ask my brain, type a question. You get a cited answer reasoned over your own .claude/brain/ notes, with links to the source files. Turn on deep mode for multi-hop reasoning on hard questions.
Reasoning uses your own model key - set ANTHROPIC_API_KEY in your data-dir .env. Without a key you still get the grounding files, just no synthesized answer. It never crashes; it degrades honestly.

2. Connect it to Claude Desktop, Cursor, ChatGPT, or Cowork (MCP)

The Model Context Protocol is the standard that Claude, GPT, and Gemini clients all speak - connect My Brain once and every MCP-capable tool can read your brain.

In your MCP client (Claude Desktop → Settings → Connectors, or the equivalent in Cursor / ChatGPT / Cowork), add the RepoOps personal MCP server and authorize once. That binds the connection to your account (person scope) - no passphrase, no shared key.
Your assistant now has three tools scoped to your brain: ask_brain (a cited, reasoned answer), search_brain (raw ranked chunks), and get_brain_file (pull one note).
Ask Claude (or any connected tool) a question and it draws on what you know - grounded and cited, isolated to you.

The hosted MCP endpoint is operator-activated; see the launch operator runbook for the one-time setup (cloud MCP env + the mcp.repoops.ai domain).

3. Read it from any app (a scoped token)

Give an external property a narrow, revocable window into your brain.

In My Brain → Personal read tokens, mint a token scoped to just the kinds / paths that property should see, with a TTL.
Store it as a server-side secret (never in client code, never in a URL) and call GET /api/me/brain/context with an Authorization: Bearer header.
Use the copy-paste REST/JS snippet or the Google Sheets =REPOOPS_ASK() Apps Script template from the integration kit.
Revoke any token instantly, and watch usage in My Brain → Who’s reading my brain.

4. Open it in Obsidian

Your brain is flat markdown with wiki-links, so it is a valid Obsidian vault. Point Obsidian at it for a graph view and live dashboards, and edits flow back through the write-trust guard. See Open your brain in Obsidian.

5. Share a slice

Publish a curated slice of your brain as a brain-pack others can install. On install it lands in their brain as proposals to review, with lineage tracking where each idea came from - knowledge that compounds between people, not just within one.

Fill it up, and let it work while you sleep

Producer: the desktop producer streams your local .claude/brain/+ native auto-memory into the hosted store so it’s reachable anywhere (REPOOPS_PERSON_SYNC=1 + a signed-in CLI token).
Capture: jot a thought from the browser or clipboard; it lands as a proposal through the write guard.
Dreaming: a nightly consolidation pass surfaces connections and drafts a morning brief (REPOOPS_PERSON_DREAM=1).

You stay in control

Provenance on every memory - where it came from, when.
Confirm / discard - nothing untrusted is silently written; you approve proposals.
Revoke any read token, any time.
Audit - see exactly which integration read what, and when.
Own & port - it’s person-scoped and markdown-portable; it follows you, and it doesn’t lock in.