Paste your secret key.
Don't have an account? Get an invite
Your typed message goes to Anthropic as-is. Don't paste raw HETUs, bank accounts, or card numbers — upload the document instead.
Rapu remembers everything you feed it — forever. It's not like ChatGPT or Claude, where every conversation starts blank. Three things to know before you start:
An empty brain can't help you. Drop a file, forward an email, or connect the desktop agent. Then the answers get good.
Point Gmail or Outlook at your brain email. Every receipt, contract, and conversation becomes searchable — automatically, in the background.
Say "Remember my son's name is Otso" and it sticks. Every future answer knows. This is what ChatGPT can't do.
Already have an LLM you love? Plug your brain into Claude Code, Claude Desktop, or any MCP client — get the setup snippet →
Wondering what protects your data? Run a live security audit →
Curious what Rapu actually is (and isn't)? Read the honest version →
Live audit, generated just now. Every check is a real query against your brain — no marketing, no guesses. Refresh
Your typed message goes to Anthropic as-is — the server can't strip PII from your own words. For sensitive data, upload the document or forward the email instead and Rapu will detect and encrypt PII on ingest.
Scale, shape, and health — the apparatus's view of what you've built.
Every source that went in: emails forwarded to your brain's address, and files you've uploaded. This is the raw material underneath the memories, entities, and claims.
Every LLM or app you've granted access to your brain, plus any API keys you've created. Each got its own token — revoke any one without affecting the others.
Each brain is a separate space with its own memory, documents, and email address. Create one per context — work, home, a project — and share it with people you trust.
People, organizations, places, and projects your brain has extracted, drawn by how they connect. Bigger dots are mentioned more often; drag any dot to rearrange.
You committed to these — the review date has passed. Tell your brain what actually happened. Scoring back-propagates calibration to the atoms that fed each prediction.
Each section gathers claims shaken loose by the same event — a prediction missing, a memory being retracted. For each: hit Update if the claim should say something new (you'll get an inline editor), or Looks fine if it still holds. Both clear it from your list; your audit history is preserved either way.
Your brain noticed these might be goals you're pursuing. They're guesses, not commitments, and nothing acts on them until you say so. Endorse turns a guess into a tracked intent; Dismiss sets it aside. You can nudge how important or urgent it feels before endorsing.
Prove you control your Estonian (or Latvian / Lithuanian) digital identity. The verification takes ~30 seconds — you'll see a 4-digit code, then confirm it in your Smart-ID app.
Open your Smart-ID app. You should see this code:
The codes must match — that's how you prove a man-in-the-middle didn't swap the hash being signed.
Waiting for confirmation…
You are about to approve a high-risk action with your Estonian e-ID. Read the summary below carefully — the 4-digit code on your phone must match the one shown here.
Loading action…
You will approve
—
payload_hash: —
Expires: —
Open your Smart-ID app. You should see this code:
The code is derived from the exact payload above — that's how you prove you are approving this action, not a different one swapped in by a man-in-the-middle.
Waiting for confirmation…
The action has been step-up confirmed. You can now complete it from the original tool with critical_action_id.
Forward Gmail or Outlook here. Every email becomes searchable.
Gmail: Settings → Forwarding → add this address. Then ask "what was in the last email from X?"
Each brain is a separate space with its own memory and email. Create one per context, rename it, and share access with people you trust.
🗂️ Manage brains →Forward your Gmail here. Everything becomes searchable.
Always-on rules you set in chat (e.g. "Always summarise in three bullets"). Applied to every future answer until you remove them.
Add every email you forward from so your brain skips spam classification.
Your brain is an MCP service — Claude Code, Claude Desktop, Cursor, custom agents can all use it. Get the setup snippet for whichever client you use:
⚙ Set up LLM connection →MCP endpoint: Copy URL
Pasting this URL into a chat won't connect anything — it goes into your client's MCP config (or the Connectors UI) along with your API key. Use the setup guide above for the exact steps.
Prove your identity with Smart-ID so critical actions (granting brain access, signing decisions, exporting fund records) can be confirmed with the same trust model Estonian banks use.
Critical actions you've scoped but haven't yet executed. Each needs a Smart-ID confirmation before its underlying tool will run.
xattr -cr /Applications/Platform\ Local\ Agent.app