KPTR.0.1.4FF40175.ADAPTER.V1

The Enterprise Brain. Every Word Kept.

Knowledge from your conversations, held as a graph your organisation can question. Every line placed, every claim traceable to the exact words someone said.

Try the demo Why it exists

495,822 claims, each receipted 1,612 meetings placed 1,271 labels, grown from speech 8B parameters, one GPU

Trained on Isambard AI UK national compute, under an AIRR award. The model that does the placing runs inside your own tenant, so nothing said leaves the organisation.

Invite it to the call. The brain builds as people speak.

Keptar joins Teams, Zoom, Google Meet or Webex as a participant. Each line is placed on the brain as it is said. In the chat it posts what it captured, asks before adding something new, challenges when a line differs from what the organisation already holds, and answers a direct question with the claim and the words behind it.

Transcripts, documents and channel messages go in the same way. Meetings are simply where most of it is said first.

Not summaries. Claims you can check to the word.

The unit is a claim. Each one records what kind of thing was said, a decision, a commitment, a directive, a proposal, a question or a statement, what it was about, and a receipt: the exact words, the speaker, the meeting, the moment. No claim exists without its receipt, so a wrong placement is visible and an invented one is impossible.

They are organised by what the organisation talks about, not by meeting. Suppliers, systems, sites, clients, programmes. Each becomes a node the first time it is named, so a question is answered from everything ever said about it, across every room. Ask it anything and it answers in quotes. When it does not know, it says not in the brain and names who does.

Built for the places where the record matters.

Keptar is horizontal infrastructure, deployed first in sectors that have to be able to show what was decided, by whom, on what basis, and what was known at the time.

A 70B teacher model was trained on Isambard AI, the UK national AI research resource, under an AIRR award, and distilled into a small model that runs inside the tenant's own environment. No confidential data leaves the organisation and no third party frontier API is involved. That is what makes it deployable where the data cannot travel.

The record becomes the evidence.

Governance fails on absence. A summary cannot tell you what it left out, so nobody can rely on it when it matters. Keptar keeps the whole record and evidences every claim, so what was decided, who committed to what, and on what basis is answerable at any point.

It works while the decision is being made, not after. When something said contradicts what the organisation already holds, it says so in the room with both quotes, before the decision lands. And it is bi temporal, so it answers what we knew then, not only what we know now.

The record has to be whole. A summary is an opinion about a record.

A / 01

Keep everything.

Every line, in order. What the model cannot place stays in the record where it was said, marked unplaced. Importance is the reader's call, not ours.

A / 02

Show the receipt.

Every claim carries the words it came from, who said them and when. You never have to trust the model. You check the source, in one click, in seconds.

A / 03

Live, not afterwards.

It says so in the chat while everyone is still in the room. Governance that arrives a week later in a document nobody opens is archiving.

One transcript. Your own brain, built in front of you.

Register and we send you a link. Paste or drop in one transcript, up to a few hundred lines, and watch it placed line by line, every claim with its receipt. Browsing the demo organisations needs no registration at all.

The Enterprise Brain. Every Word Kept.

Invite it to the call. The brain builds as people speak.

Not summaries. Claims you can check to the word.

Built for the places where the record matters.

Health and care

Government

Financial services

Legal