Spec Studio Available now

Why it exists

The gap is where the money goes.

Every requirement nobody wrote down is still a decision. Somebody makes it. Increasingly that somebody is a model, at machine speed, without telling you.

That distance between what you said and what had to be assumed is the Inference Gap. It is not a documentation problem — documentation is written after the fact and nobody reads it. It is a specification problem, and specification is a discipline with a hundred years of precedent in every other branch of engineering.

Spec Studio exists because that discipline needs a place to live. Requirements scattered across tickets, decks, and a Slack thread from March are not a specification, no matter how many of them there are.

Elicit

Get it out of people's heads.

Structured, AI-facilitated elicitation sessions that ask the questions an expert requirements analyst would ask — if every requirements analyst knew enough about each stakeholder's area of expertise to ask them.

Sessions run from guides that encode a real facilitation method rather than a blank prompt box: what to ask, in what order, what to do when an answer is vague, and when to stop and escalate instead of accepting something ambiguous. What comes out is captured as structured intent, not a transcript.

It works with stakeholders who are not in the same room, on the same continent, or awake at the same time — which is the normal condition of any organisation large enough to need this.

Review & Instruct

Somebody has to sign it.

A specification nobody approved is a draft. Review and approval are first-class here, with a named human accountable for the result.

The Alignment Authority reviews the Build Package, adjudicates what is contested, and approves it. That approval is a real gate, not a checkbox — it is the point where an organisation commits to what it actually means, before a generator turns the ambiguity into ten thousand lines of confident code.

The same surface produces the instruction documents your coding agents read. If your teams use Claude Code, Copilot, Cursor, or Codex, the specification and the agent instructions come out of one place instead of drifting apart in two.

Library

Stop solving it from scratch.

The Golden Record, plus the reusable scaffolding your organisation builds up over time.

Approved specifications, templates, chassis, and session guides live in one place and get reused. The second project in a domain should be substantially cheaper than the first, and it is not, in most organisations, because nothing from the first one was kept in a form anybody could reuse.

This is also where an organisation's standards actually become standards — not a policy document nobody opens, but the template people start from because starting there is easier than not.

How to get it

Free on our cloud. Licensed on yours.

Pick the one your security review can live with. Where there is a license, it is a nominal fee that covers the cost of supporting the software.

Bring your own model

Your models, your credits, no markup from us.

Both downloadable images are bring-your-own-model. You point them at your own LLM provider and pay that provider directly.

Use your existing Azure — or any other provider's — credits for model usage. We add no upcharge, no markup, and no per-token add-on. What your provider charges you is what it costs.

This matters more than it sounds. Reselling model capacity at a margin is the default business model in this category, and it quietly turns your AI spend into somebody else's revenue line. We would rather sell you software once than tax every prompt you run through it forever.