FAQ

Payment#

How do I top up? Currently manually — email us at support@example.com. Crypto top-up is coming soon.

Is there a subscription or a minimum payment? No, only pay-as-you-go charges from a prepaid balance. How requests are priced is covered in Balance and pricing.

Am I charged for errors? No, if the request was rejected before the response started. If a stream broke off mid-response, or you aborted the request yourself after HTTP 200, the input and the output generated so far are charged.

Why do I get a 402 when there is money on the balance? The remainder is held by the reserve of requests still in progress. Retry once they finish.

Models and formats#

Which models are available? GET /v1/models returns the current list — see Models.

Why does a model work in Chat Completions but not in Messages? The gateway does not translate between request formats: a model works on endpoints whose format its provider supports. See API overview.

Can I use previous_response_id or the Files API? No: the gateway keeps no provider-side state. Send the conversation history and files (base64 or links) with every request.

Keys and security#

Can I see my key again? Yes, with “Show key” in the cabinet's Keys section. The key is stored encrypted.

How do I cap one application's spending? Issue a separate key for it and set a spend limit on the key.

Does the gateway pass my identifiers to the provider? No: user, safety_identifier, metadata and your request headers do not reach the provider; the gateway's identifier is sent instead. On /v1/responses conversations are not stored at the provider (store: false).

Errors#

What do I do about a 429? Wait for the pause in Retry-After and retry — SDKs do it themselves. It usually means your account has too many concurrent requests. See Limits.

What should I include when contacting support? The request_id from the error body or the x-request-id response header — it identifies the exact request.