The architecture of a Mind.

Senior experts under-articulate what they know. The finding is older than AI — decades of knowledge-engineering research show that direct questions return rationalisations, not method. So the Studio is an agent that probes, follows up, picks up on hesitation. Forms extract answers; conversations surface distinctions.

Underneath the conversation is a craft. The Studio uses methods knowledge engineers use to capture tacit expertise from senior practitioners — grounding every question in a concrete case, reflecting back what it heard before advancing, and pushing back when an expert contradicts themselves. Two things make most assistants drift toward whoever is asking: language models predict what’s likely to come next, and the feedback they’re tuned on rewards agreement. The Studio doesn’t drift.

Its output isn’t a transcript. It’s three layers — each authored independently, each held to its own quality standards.

Together, they’re the seed the platform compiles into a working Mind.

A person’s mind isn’t only what they think — it’s what they’ve read, written, decided, said. A Mind is the same.

Every artefact enters the Mind intact. Nothing is summarised; nothing is averaged. Each source is chunked with a contextual prefix and embedded into a vector index — broken up for retrieval, but never rewritten. At every question, the runtime selects, reranks, and restores the relevant slice; the merged window is what reaches the Mind.

Knowledge and personality are different tools for different jobs. The library stays true to the source — intact, verbatim, every claim traceable. The Mind itself is composed at every turn, drawn from its three layers. We hold them apart by design.

Every reply is the work of context engineering: deciding, for this question in this moment, what fills the Mind’s context.

What you read back from a Mind reads like consulting the expert, not querying a chatbot. A language model is a powerful general reasoner; the Mind focuses it on one expert — captured through three layers and a body of work. At every question, we identify and extract the relevant slice of the expert’s library, frame the model through those three layers, and ask it to reason as the expert would.

AI answers rarely come with proof. You read confident sentences with no way to know what’s grounded and what’s hallucinated. A Mind shows its work: hover a chip in a Mind’s answer and a popover opens with the source and verbatim cited text — pulled directly from the expert’s library. The citation isn’t decoration. It’s the receipt.

A Mind sits between two parties who both have something at stake. The expert has spent years writing the work that becomes the library — that’s their intellectual property, and they decide what enters it. The subscriber asks the Mind about real clients, real cases, real situations — the kind of detail you don’t type into a public chatbot. Both should know where the line falls, and that it holds.

A subscriber’s conversation with a Mind is private. The expert running the Mind can’t see it. Other subscribers on the platform can’t reach it. Each conversation is held in its own workspace, separately from every other.

A library belongs to the expert who built it. What’s uploaded to a Mind is reachable only by that Mind — no other creator on the platform can read it, no other Mind can draw from it. The expert’s method — voice, skills, judgment — is the Mind. It stays with the Mind. It does not become someone else’s product.

What this page describes is the foundation. A Mind, compiled from one expert’s method, is a faithful study of how that expert thinks. The longer arc carries the Mind from consultant to colleague: not only an answer in the expert’s voice, but work done in it. Drafts shaped by their style. Edits made against their standard. Eventually, presence in the systems where the work actually happens — reading the artifact, applying the judgment, returning it shaped. The alternative today is generic AI — confident on the average case, off on yours. A Mind is one expert’s judgment, applied to your case.

The roadmap is what gets us there. Tools so a Mind can work directly with the documents its expert would draft — briefs, decks, contracts, transcripts. Integrations into the systems where their industry actually runs. Each will appear in the journal as it ships.

Why now, and why only five. The chatbot pattern has run its first lap. The next move in AI isn’t a bigger model averaging more of the open web — it’s specific expertise applied to specific work, in the tools where experts already work. That can’t be designed at distance. It gets built with experts, in the room. Five founding creators isn’t a market cap — it’s a design partnership. The bet is that depth matters more than breadth, and that the right tools are the ones experts already use.