​
1. Install

​
2. Sign in

Opens Google sign-in in your browser and provisions your workspace, your first twin, and an API key. You never paste a key.

​
3. Get real context in

Pick whichever is closest to where your work already lives. Start with --dry-run so you can see what would come in before it does.
Ingest something you’d actually want to look up in three months. A twin fed on test data gives test-data answers, and you’ll conclude it doesn’t work.

​
4. Ask it something

Pulled from is the part worth reading. It shows which memory layers grounded the answer — the fastest way to tell whether what you ingested actually landed.

​
5. Work the queue

Extraction doesn’t silently become truth. Review what it pulled out:
Approving a memory is you standing behind it. That’s what makes the answers yours rather than a summary of whatever it happened to read.

​
Next steps

Organize with stores

Compartments inside a twin, and auto-routing content into them.

Wire up your AI agent

Let Claude Code, Codex, or Cursor read and write your twin.

Connect Slack

Where your team actually asks the twin things.

See why an answer happened

bestmate trace — what was retrieved, ranked, and returned.