The fixed price for AI work.

AI products sell work at fixed prices, and buy the inference behind it at floating ones. We fix the price of each unit of work.

X-rayed to date
110B
tokens
21,000
runs

Airlines sell seats at fixed prices and buy fuel at floating ones, so they hedge their fuel.

AI products sell work at fixed prices and buy inference at floating ones. Merlinfield fixes that cost.

Why now

The cost under a fixed price moves every month

Agents

Agents make the cost of a single task unpredictable, from one run to the next.

Providers

Providers change their prices and models every few months, so cost moves even when your product doesn't.

Boards

AI gross margin is now a standing question in board meetings.

The problem

Your price is fixed.
Your cost per unit isn't.

In the traces we've studied, the same agent on the same task cost up to 30 times more from one run to the next. You charge the same for both. The gap comes out of your margin.

Your price per unit30×UP TO
Run cost above priceRun cost below priceSame agent, same task, run by run · Illustration, log scale

Your X-ray shows your own spread, by day and by month.

The product

Fix as much of the cost as you want

Start with the X-ray. Each step hands us more of the swing, up to one price per unit for the year.

Free
X-ray
Your real cost per unit, how much it moves, and a quote.
At cost
Pass-through
We buy the inference and bill it per unit, every call itemized.
Bounded
Price collar
Your price per unit moves only between a floor and a ceiling.
Fixed
Full guarantee
One price per unit for 12 months. The swings are ours.
Who it's for

Built for AI products sold at a fixed price

You charge per outcome, per seat, or a flat fee, and you pay per token.

AI support
The unit is
a resolved ticket

Its cost moves with
turns·tool calls·retries
Document AI
The unit is
a processed document

Its cost moves with
scan quality·second passes·retries
Where we sit

Others fix the price of tokens.
We fix the price of the work.

Reserved capacity

Reserved capacity and committed-use discounts fix what a token or a GPU hour costs.

Gateways

Gateways and cost dashboards route the tokens and show where they went.

Merlinfield

We fix what a resolved ticket or a processed document costs, for the whole year.

For the CFO and the CTO

Finance

A cost line you can forecast

One cost per unit for the year. Your margin moves with your price, not your agents.

For finance
Engineering

Nothing in your request path until you sign

Then a small change in each SDK, and your own keys stay as the fallback.

For engineering

Built so you don't have to take us on trust

Billed in arrears

You never pay for a unit we haven't served.

A reserve in trust

Held for you at a US bank, with an independent trustee.

Your own keys

If our endpoint fails, your calls fall back to your providers on your keys.

Find out what a unit of work costs you

Two files. Five business days. Free.