Frequently asked questions

Do I need a special SDK?

No. The Gateway is compatible with the OpenAI, Anthropic, and Google GenAI SDKs — point them at the Gateway base URL and use a project API key. See Integrations.

What is the base URL?

https://api.atptoken.ai for the Anthropic and Gemini formats, and https://api.atptoken.ai/v1 for the OpenAI format. Each SDK page shows the exact value.

Which models can my key call?

Only the models on its project's allowed list. GET /v1/models lists what the platform offers; access is enforced per project at request time, so the list is the menu — not your key's permissions.

Can I stream responses?

Yes. Set stream: true and the Gateway streams SSE in the format matching your SDK. See OpenAI SSE and Anthropic SSE.

What happens if a provider is down?

Each model is served by a pool of providers and the Gateway fails over automatically. If every provider for a model is unavailable you get a 503 with Retry-After. See Provider routing.

How am I billed, and are credits refundable?

Usage is metered on input + output tokens and paid in credits (1 credit = USD 0.01). Top-ups and credits are non-refundable. See How credits work and Top up & wallet.

How do I see what I spent?

The Usage page summarizes credits and tokens by model and by key; request logs give a per-request audit trail. See Usage & logs and Tracking spend.

Why did a reasoning model return 200 with an empty message?

max_tokens was too low. Reasoning (extended-thinking) models spend their thinking inside that budget; when it runs out before any visible output, you get an empty 200 — typically with zero usage and no credits deducted. Raise max_tokens to cover thinking plus your expected output. See Common responses.

Does ATP Token only serve text models?

No. The catalog also includes image, video, audio (TTS) and embedding models. See Media models for what each modality does and how it is billed.