★ FEATUREDBuild your BI dashboard from scratch — faster than taming Power BIRead →
scope · from meeting to spec

The meeting became a spec. Not a recording nobody plays again.

Most meetings end as an audio file and four pages of notes, and both sit untouched a week later. scope takes the meeting all the way: it records, every speaker gets their name on what they said, and it writes a scope of work from the decisions that were actually made — what is in, and what is not. You approve it, the client approves it, and only then does an AI agent start building on it. The audio is processed on our own infrastructure in the EU, and there is no per-seat licence.

Record·Who said what·Scope of work·Client sign-off·Agent builds
The scope

Minutes say what was said. A scope says what gets built.

That is where most meeting tools stop: they hand you a tidy summary, and a human still has to sit down and turn it into a piece of work. scope pulls the decisions out and lays them down as something you can build from — what is in scope, what is out, what still has no answer, and who owes that answer. The unresolved parts stand as open questions with a name attached, instead of disappearing into an audio file. The name is meant literally: a scope is the line that decides what the job is, and the periscope is what looks into the meeting to find it.

In scope·Out of scope·Open questions·Who owes the answer
Who said what

A spec without names is a conversation nobody can act on.

«We agreed to wait» means something very different from «the client asked us to wait». scope knows the voices from a short sample participants record themselves before the meeting, and puts a name on every passage. Anyone who would rather not give a voice sample is skipped — they simply get a name assigned by hand. It is voluntary, and what the sample is used for is written down.

30-second voice sample·Voluntary·Name on every passage·Manual naming as an alternative
Danish, not translated English

Danish speech understood by a model trained on Danish.

Ordinary speech-to-text models are trained on English and guess at Danish. We use open models fine-tuned on Danish, and the difference is measurable: under half the errors a general model makes on the same Danish benchmark. Domain terms, product names and abbreviations go into a dictionary up front, so «MCP» does not come back as «em see pee».

Danish-trained model·Under half the error rate·Own dictionary·Jargon and abbreviations
Where the audio goes

The recording never leaves the house. There is no vendor to ask.

Processing happens on our own infrastructure with open models. No audio is sent to a US service, and there is no per-seat licence running whether or not we hold meetings. When a session needs more compute, it runs on an EU cloud GPU paid per use — not a US cloud with a European region. Participants are told what is recorded, why, how long it is kept, and how to withdraw consent.

Own infrastructure·Open models·No per-seat licence·EU cloud GPU·Versioned consent
The sign-off

Nobody builds on a spec the client has not seen.

The scope goes to the client in their own portal login. They approve it, or they send back a correction in their own words — and a new version is written. Only once it is approved is it locked and handed on to an agent that produces the plan, the mockup or the first version. That is where the time is saved: not by writing the minutes faster, but by the next step starting on its own.

Client approves·Corrections with reasons·Versions preserved·Locked once approved
Hi — I'm Aidan