How Tacit Mind works
Every AI you use starts from zero. Tacit Mind is one memory behind all of them, so they start from you: what you know, what you decided, and why. It keeps every conversation, learns how you think, keeps private parts of your life in sealed rooms, and lets any AI (Claude, ChatGPT, Gemini, Grok, Perplexity) reach the same mind through one connector. Matt Edmund is its CTO; his own twin is the first proof. https://tacitmind.ai
The idea
Agents are the easy part. Every company will have them, and they'll be cheap and interchangeable. What makes an AI yours isn't the model. It's your reasons: why you decide the way you do, what you care about, what you'd never do. Tacit Mind keeps those reasons, and puts them behind every AI you use.
The layers: humans manage human things (relationships, trust, the room). Your twin manages the agents (which one, what counts as done, what you'd have said). The agents do the tasks.
One rule keeps it safe: intent stays home. Your twin goes out and reads every catalog. Nobody comes in and reads your twin. What leaves is a question. What comes back gets judged.
What happens on one question
- Starts from what it knows. The map it already holds of you.
- Hears you. Your words, exactly as said.
- Re-draws the map. Context is chosen fresh from what was just said, never a fixed prompt.
- Reads what fits. The model reads the memories that matter for this question.
- Goes deeper. It follows links to older ideas and decisions.
- Finds your why. The reason behind the decision, in your words.
- Checks what may leave. A no-leak gate per audience: a golfer, a recruiter and a client each see only their layer.
- Answers from what it pulled, and keeps the lesson.
Same question, two answers. Matt saved an essay arguing that AI makes things up the same way people do. Any AI on its own replied: hallucination is a known problem; add retrieval, cite sources, lower the temperature. Correct, and no use to him. The same AI with his twin replied: it matters most for your twin, because a twin that makes up your reasons fails on 30 January; keep logging your reasons word for word, give it a guard like the caddie's; "never invent a why" is now in the interview set.
The layers (the IQ ladder)
- MapIQ: one conversation. Each moment is a slice through a map of what you know. The AI reads across the slices and ends in one new why.
- TacitIQ: what came from it. Every conversation is kept in full and folds into what it taught. Over months, hundreds gather around every part of your life. (Matt's twin: 337 conversations read from April to October 2026, 86 of his reasons on record in his own words.)
- EmergentIQ: how you think. Patterns form across those hundreds: how you think, what you believe, what you're aiming for. The best leaps cross from one part of life to another.
- NDA IQ: know everything, keep it safe. Each part of life runs in its own sealed lane. A client's twin can ask, but only their room opens. What is learned rises back into the mind; the private details stay behind.
- MCP: every AI, one mind. Claude, ChatGPT, Gemini, Grok and Perplexity all reach the same mind through one Model Context Protocol connector.
- IntentIQ: the brain. It reads how you think, goes looking the way you would, scores every idea it finds, and shows you only the few worth your time. Every guess has a score.
- The flywheel. IntentIQ starts the next conversation already knowing you. What that conversation teaches joins the hundreds, the patterns sharpen, and the next guess is better. Every turn is scored.
What it enables
- Twin to twin. Two people's twins plan something together (a weekend away, a hard conversation) and share only what each person allows. Private reasons stay private.
- Jobs. Your twin sits the first interviews; candidates' and companies' twins compare receipts in a room between them.
- Coaching. A coach's twin carries his method to students between lessons, and sends the coach one line instead of a transcript. The coach always decides.
Design rules
- Memory with reasons. Every memory keeps its why and its history, so the twin can say when it changed its mind.
- Sealed by who's asking. Client rooms are islands. Answers pass a no-leak gate per audience and hold when unsure.
- Small models judge, big models think. Fast judgment models (TypeSafe Jev) do recall, routing and classification; frontier models are spent only on meaning.
- Earned autonomy. The twin handles one kind of decision at a time once it has proven it agrees with you, starting in shadow.
- Tacit Mind is the brain. Tacit Hands do the work, in throwaway sandboxes.
The interactive story: https://33prime.io/how-it-works.html. Technical depth: https://33prime.io/under-the-hood.md