Context before agents. Here's what that actually means.
The operating memory we build, why it has to come before the AI, and exactly how the engagement works.
One memory your team and your AI both run on
Not a doc nobody updates. A living source of how the business actually runs, structured in three layers so people and agents can both pull from it.
What's true now
The current state of how the business actually runs. One trusted answer, not ten stale docs disagreeing with each other. This is what your team and your AI both act from.
Why it changed
Every decision, the dead-ends, the paths you rejected and the reason you killed them. So the thinking survives, not just the outcome.
The evidence underneath
The calls, the notes, the raw detail the rest is built from. Every answer traces back to where it came from.
operating-memory/
├─ artifact what's true now
├─ ledger why it changed
└─ source the evidence
One source. Your team and your agents both read it.
An agent on top of nothing learns nothing.
Point AI at knowledge that only lives in people's heads and it doesn't fill the gap, it inherits it. It guesses, confidently, and scales the guess across everything it touches.
Give it the operating memory first and the same AI stops inventing and starts answering from how you actually run. The order is the whole point. Context first. Then agents.
We build it hands-on, in two steps.
- Retainer rateoverview says £3,500 · now £4,800HIGH
- Primary contactoverview says Tom · now PriyaHIGH
- Plumbing IFC exportblocks Estimating Assistant 22/25MED
Drift surfaced against the decision log.
Find where the knowledge is trapped
We map where your operating knowledge actually lives, where it's costing you, and exactly what to build first. About 2 to 3 hours of your time, and you come out with a clear picture of the gaps and a plan for the first layer, yours to keep either way.
If we don't surface at least 3 high-value gaps, you don't pay.
From $1.5k · credited in full to the build
Build the first layer, and the assistant on it
We build that first layer: your operating memory, plus an assistant that answers real, company-specific questions from it. We agree the exact questions it has to answer before we start.
If it can't answer them, we keep working until it can.
- 04-20 retainer £3,500 → £4,800 (scope grew).
- 05-30 Inv #8 (£4,800) issued · outstanding.
- 06-09 Estimating Assistant acceptance 22/25.
Every answer reconciles from here, not a stale summary.
How we get it out of your head, without homework.
Writing it all down is the thing every founder means to do and never does. So we don't ask you to. You talk, we do the writing.
Talk, don't write
We record short interviews and turn them into clean, structured notes. You'll talk for an hour. Nobody writes for one.
Start with what exists
First we pull in what's already scattered across Slack, email, docs and past calls, so we're not asking you for things you've already said.
Watch the real work
For the things that only live in how someone does them, a quick screen recording, captured and written up for you.
You confirm it's right
Nothing becomes the source of truth until you've checked it. It's your knowledge, verified by you, not us guessing at your business.
How it stays current. The honest version.
Every wiki you've abandoned died the same way: keeping it updated became a second job nobody did. So we'll be straight with you. Nothing stays current by magic, and anyone selling you self-healing software is selling something that doesn't exist yet.
What we build is a small, deliberate routine where updates happen as part of the work that's already changing, not as admin piled on top of your team. The difference between a habit and a chore. We'd rather under-promise this and have it hold.
One source of truth your whole company can use. People and AI.
Your operating memory, live and maintained, plus an assistant that answers real questions from it. New hires ramp faster, the knowledge stops living only in a few heads, and your AI finally knows how the business actually works.