Skip to main content
Bringing your own key means giving Uniblock a credential you already hold for a provider we carry, so a request to that provider runs on your account there instead of ours. The provider bills you directly for it, the way it would if you called that provider yourself.

What “free” means, precisely

A request served on your own key costs you nothing on your Uniblock balance. Nothing is charged and nothing is held against your credit for it, not even briefly. That is not the same as invisible. The request is still fully metered: it gets a usage record with real prompt and completion token counts, it appears in Usage and in Activity exactly like any other request, and it is counted apart from the requests we serve on our own keys rather than folded into them. Read “free” as we charge you nothing for this one, not as this one leaves no trace. See Usage and metering for the full field set a request produces. The provider itself is not free. Running on your own key means the provider invoices your account for the tokens, which is the entire point of bringing one: the cost moves from your Uniblock balance to your account at the provider.

Your key, and only your key, for that provider

Once your workspace holds a key for a provider, every direct request to that provider runs on it. We do not fall back to our own key for that provider, because that would mean charging you for a request you brought your own credential to cover. That cuts the other way when the key stops working. If the provider will not accept it, the request is refused rather than quietly served on our key and billed as usual. See If the provider refuses your key below.

It does not add anything to your catalog

Bringing a key changes which credential answers your calls to a provider we already carry. It does not make a new provider or a new model callable. What you can call is still exactly what Uniblock carries. Your own key never adds to that list, and a model that is not already callable through us stays not callable no matter whose key you hold for it. See Models for how the callable set is decided and discovered.

Which providers you can bring a key for

Not every provider we carry takes a key from you. Only a provider whose authentication involves a credential at all can accept one, so a provider that needs nothing from you to reach has nothing here for a key to replace. The screen where you add a key offers exactly the providers this applies to, and only among the providers we currently carry.

If the provider refuses your key

If the provider turns down your key on a live request, you get a plain refusal rather than a silent switch to our own key or a pass-through of whatever the provider sent back internally: the message names the provider, says nothing was charged, and tells you to fix the key. The message calls that place the Provider keys screen; the dashboard’s own sidebar calls it BYOK, and they are the same screen. See Add a provider key for how to replace it, and Errors and refusals for how a response like this is shaped.

Who can manage keys

Managing provider keys, including seeing which ones a workspace already holds, takes the same authority as managing routing or inviting somebody: owner or admin. A developer or a viewer cannot see or change this workspace’s provider keys. See Workspaces and members for the full role table.

Next

Add a provider key

Finding the screen, adding a key, and what proving it before saving means.

Models

How the callable set is decided, and why your own key cannot extend it.

What this page does not cover yet

The routing-profile case. A request that can try more than one model in one chain is out of scope here. This page describes a direct request to a single provider you hold your own key for. A walkthrough against a deployed environment. The rules above are what the service enforces. The screen itself has not been walked end to end against a deployed environment for this documentation. Where the screen and this page disagree, trust the screen and tell us.