Pricing

Three lines, and one of them is not ours

Gibreen is priced the way an implementation is priced, because that is what it is. There are no credits, no seat minimums, and no margin taken on the model usage — that bill goes to you, from the model vendor, at their price.

One · once

Implementation

fixed fee, scoped after a first call

The eight-week engagement
Your instance, provisioned and isolated
Connectors configured against your systems
Two workflows live, one built with your own people
A baseline in week one and a re-measurement in week eight

Two · annual

Instance subscription

per instance, per year

Hosting, patching, backups and restore drills
Platform updates and connector maintenance
The gate, the record and the console
Support with a named person, not a queue
Priced per instance — not per user, not per agent, not per run

Three · yours

Model usage

billed to you, by the model vendor

Runs on your own Claude account
You see every cost at source
We take no margin and never see the invoice
You can cap it, audit it, or move it
If you leave, that line simply stops

Why there are no numbers on this page

Because we do not have one honest number to put here yet. A fixed fee against an unknown scope is either a guess that protects us or a guess that traps you, and both are worse than a scoping call. You will get a number in writing before any work starts, and it will not move afterwards.

If you want a rough sense before that call: we anchor the implementation against what an ERP implementation of similar scope costs, and the subscription is the smaller, recurring half.

What we deliberately do not sell

✗ Not on offer

Credits. The major vendors have re-cut their credit meters repeatedly and at least one withdrew its published counts entirely. A unit of value nobody can forecast is not a price, it is a liability you inherit.
Per-seat licensing. The point of an agent team is that it is not people. Charging as though it were rewards us for the wrong thing.
Per-run or per-action billing. It makes you ration the thing you bought, and it makes us prefer chatty agents.
A free self-serve tier. Every failure would happen in front of a stranger with nobody to answer the gate.
Markup on tokens. Ever. It is the one line where our interests could quietly diverge from yours.

✓ What that buys you

A cost you can model: one fee, one subscription, one vendor bill you already understand.
No incentive for us to make the agents talk more than the work needs.
The ability to cap and audit the volatile line yourself, at source.
A clean exit: stop the subscription, keep the record, keep the account.
A price conversation that happens once, with a person, in week zero.

Questions people ask before the call

Why is there no trial or signup button?

Because the product's reliability depends on a gate, and a gate depends on a person answering it. A stranger who signs up at midnight has no such person, so the first thing they would experience is the failure mode. We would rather have fewer customers who succeed than a funnel that produces a bad first week.

This is a deliberate trade and it costs us volume. If self-serve is what you want, say so on the call and we will name the products that do it well.

What does the model usage actually cost to run?

It depends almost entirely on how much context each run needs and how often the agent runs — which is why we will not quote it blind. What we can tell you is that it is measurable from the first day of your engagement, because every run is recorded in a usage ledger on your own instance, and the week-eight review includes a real monthly figure rather than a projection.

Can we run it on our own infrastructure?

Yes, and some procurement processes require it. It changes what the subscription covers — we can no longer patch and restore for you — so it is scoped separately and is generally a slightly larger implementation and a smaller subscription.

What happens if we stop paying?

The instance stops. You keep the repository — the charters, the jobs, the runs, the decisions, the refusals — and it opens in any text editor. Your Claude account is yours and is unaffected. We do not hold your operational history hostage, because the format was chosen specifically so that we could not.

Do you charge for additional agents?

No. An agent is a file; charging per file would be absurd. Additional workflows — which mean connector work, testing and gate design — are scoped as work. By then you will have built one yourself and will know roughly what it should cost.

Is there a minimum term?

The implementation is eight weeks and you may stop at the end of any week, keeping what has been built. The subscription is annual per instance, because hosting, patching and restore drills are annual commitments on our side.