Get
started

Two ways to get started. Neither commits you to anything.

A proof of concept is worth doing when it answers a question you already have, on your own data, in a fixed number of weeks, with the goals written down before it starts. That is the whole offer. The fee is credited at 10% of your recurring charges each month for the first year, up to the fee itself if you proceed, and the metered usage accrues on the console from day one, so you leave with the forecast for production.

Start a paid proof of concept Get a free fixed price first

The two options

One is free and takes an email. The other is four weeks and ends in a decision.

Option 1 — free

The fixed price analysis

Send us the last three months of invoices from the vendors currently holding, searching and answering questions about your data.

You receive a fixed monthly price for the same workload on Data Oil, held for twelve months, with the full working shown — every meter, every quantity, and what we assumed where an invoice did not say.

It costs nothing and it commits you to nothing.

Option 2 — paid

The proof of concept

A proof of concept is a decision instrument, not a demo. Four weeks. One workload you define. Goals agreed in writing in week one.

A fixed fee, credited back at 10% of your recurring monthly charges for the first year, up to the value of the fee itself. A written report either way, whether you proceed or not.

You leave with a decision, not a meeting summary.

Most customers do the first, then the second. Nothing stops you doing only the first.

Week one, before anything else

The goals, in writing

Nothing is measured until this document exists and both sides have signed it. It is short, and it is the thing you keep whether or not you proceed.

The week-one document

  • The workload. One, named: a multilingual search, a set of documents that answers with citations, a reporting store replacing two systems, a migration off a database-plus-search-engine pair.
  • The twenty questions the proof has to answer, written by the person on your side who knows what the answers should be.
  • The measures. Relevance on the twenty questions; the queries the reporting has to run; the languages search has to work in; the load that has to complete; the restore that has to be proved.
  • The date. Four weeks from the day the data lands.
  • What proceeding means, and what it costs: the fee credited at 10% of your recurring charges each month for the first year, up to the fee itself, and the usage that has been accruing on the console as the forecast.

The four weeks

Every week leaves you holding evidence

01

Week one

It deploys where your data has to be, with sign-in through your own directory — or a tenancy on the hosted platform is provisioned. Your data lands. The security demonstrations run.

You keep: a deployment you signed into, and the goals document.

02

Week two

Your real data in your real shapes, indexed four ways, the schema described in your words, and the ingest held to your budget at every stage as it ran.

You keep: a schema export, a catalogue, and the bill itemised by meter.

03

Week three

The query you cannot run today, measured against the query you run now; your model and your key, and exactly what would leave.

You keep: a measured comparison and a reviewed /query/prepare output.

04

Week four

A backup drill you watched, a self-service restore into a new name, permissions down to the column, an audit export — and the written answer.

You keep: a restore that happened, an audit file, and a report you own.

The last one is a written answer either way.

For your security team

Seven demonstrations, live, in week one

A security architecture review is where a small vendor usually loses weeks. Here it is where the platform is strongest, so it runs first, in detail, and live rather than in slides.

The seven, in the order reviewers ask for themOpen the seven

A column fenced by a role

A role granting two columns of one collection to one named person; SELECT * returning the record without the third; the same caller naming that column and being refused; and the audit line recording the narrowing — the question a compliance audit actually asks.

The same fence over the PostgreSQL port

Both query surfaces enforce it. Connect psql and repeat the refusal.

A directory group at sign-in

A group your identity provider asserts on the token reaches a role as directory:<group>; a group with no mapping is refused with a message naming the groups the token carried; a mapping cannot grant the platform administrator.

A residency refusal

Declare a region, write to a database placed outside it, show that nothing was stored.

What the model is sent

POST /query/prepare: the whole submission that would go to the model provider, handed back with nothing having left. Let them read the prompt.

A key rotation with no outage

Then a revocation, then the introspection route answering active:false identically for unknown, expired and revoked.

The console's own audit

Open the audit page, take an action, and show the line in your own log with the actor, the action and the outcome; and that nothing on the customer surface can remove it.

Why it is paid

A paid proof has a date, a scope and a name on it

It happens

The data arrives, the questions are written down, and the weekly session has an agenda. A free pilot is the one that is still open in six months.

The fee is credited at 10% of your recurring charges each month for the first year, up to the fee itself — against storage, compute, statements, searches and the assistant, and never against ingestion and translation. You know the fee when you sign, so you know the cap.

The usage is real

The meters are the actual meters on your actual data. You leave with an invoice that is the forecast for production, itemised, with the corroboration figures behind it — not a quote.

And a buyer who does not proceed paid for a report they can act on.

Write down the workload and the three things that hurt.

Reply to the mail this button opens. We answer with whether a proof of concept would prove anything — and if it would not, we say so.

Start a paid proof of concept