An API key for a model, on the bill you already have.

Add AI to your application without opening an account with another company, handing over another card, or discovering at the end of the month what it cost. Your own key, your own limits, your own invoice.

Talk to us about AI endpoints

What you get

A key and an endpoint

Point your application at it and call it. Nothing to install, nothing to keep running, and no GPU to rent by the hour and forget about.

A monthly cap you set

The number you are not willing to go past. Past it the endpoint refuses the call rather than carrying on and billing you for it.

A rate limit that is yours

Your limit is your limit. Another customer having a busy day does not spend it.

It runs on our machines

The same estate as your hosting, in the same place, under the same agreement — rather than your customers' text being sent to a third party you have not vetted.

On your existing invoice

One supplier, one renewal date, one bill. If you stop paying, it suspends alongside everything else instead of becoming a separate problem.

Usage you can check

Every call is recorded individually, so the figure on your invoice is a sum of rows you can look at rather than a total you have to take on trust.

The bits developers ask about first

Short answers, because these are the questions that decide whether this is usable and they should not need a sales call.

Over your rate limit is a 429 With a Retry-After header, because coming back in a moment is the right thing for your client to do.
Over your monthly cap is a 403 Deliberately not a 429 — retrying will not help, and a client that keeps trying would hammer us for the rest of the month.
Your key is stored as a hash Only a SHA-256 of it is kept, and it is compared in constant time. The key itself is not in our configuration, not on the machine serving you and not in the logs.
A call we cannot measure is not billed If usage cannot be read back from the engine it is recorded as unmetered rather than estimated. We would rather under-count than put an invented number on your invoice.

Managing it

What you do	What happens
Rotate the key	A new key works immediately and the old one stops. Useful when someone leaves, or a key ends up somewhere public.
Raise or lower the cap	Takes effect straight away. Lower it before a risky deploy if you want to.
Pause it	The key stops working and nothing is charged. Turn it back on when you want it.
Cancel it	The endpoint goes and nothing of yours stays behind on the machine.

Frequently asked questions

Is my data used to train anything?

No. These are models serving requests, not a training pipeline. Your calls are metered and answered; they are not collected as a dataset.

Which models can I call?

It depends what the machines in your location are running, and it changes as we add capacity. Ask and we will tell you what is available where you are — rather than publish a list that is wrong in a month.

What does it cost?

You are charged on usage, against the cap you set. Tell us roughly what you are calling today and how much, and we will give you a real figure rather than a range.

Can I try it before committing?

Yes. Set a small cap and point something at it. The cap is the whole safety net — there is no way for it to cost more than the number you chose.

What happens if I hit the cap mid-month?

Calls are refused with a 403 until you raise it or the month rolls over. Nothing silently continues and appears on the invoice.

It sits with the rest of your hosting

Same panel and same bill as your websites, email, DNS, databases and machines.

See all hosting

Tell us what you are building.

What you are calling today, roughly how much, and what you want it to cost. We will tell you plainly whether this fits.

Talk to us about AI endpoints