FAQ

One key and one API for models from many labs, each at its list price, paid from one $THINK balance with no subscription. Switching models means changing one field. See Principles.

Sign in at ai.hyperflow.fun, deposit some $THINK, create an API key and send your first request. The Quickstart walks through it.

Yes. Chat completions, embeddings, image generation and edits, video generation and the model list follow the OpenAI API. Point any OpenAI client at https://ai.hyperflow.fun/v1 and use your Flow AI key. See the API overview for the parameters each one takes.

Any OpenAI-compatible client: the official OpenAI SDKs for Python and TypeScript, the Vercel AI SDK, LangChain, PydanticAI and more. See Integrations.

Ask on Discord or Telegram. To report a problem with a request, see Report feedback.

Each model is charged at its list price, shown on the Models page. There’s no subscription and no monthly minimum. See Pricing.

Before a request runs, Flow AI holds the most it could cost from your balance. When it finishes, you’re charged for what it actually used and the rest is released.

In $THINK, deposited from your wallet on [Funds](https://ai.hyperflow.fun/console/funds). 1 THINK =$1 of usage. You can withdraw what you haven’t spent.

Your available balance can’t cover the most the request could cost. Deposit more on Funds, or set a lower max_tokens so the hold is smaller.

Your balance is everything you have. Available is your balance minus what’s held for requests still running; it’s what new requests can use. Held amounts come back as soon as each request settles.

Not if the request fails before the model reads anything. If it fails partway, you pay for the tokens the model already read and wrote. A video that fails isn’t charged.

Dashboard shows your balance and recent spend, Usage shows spend and tokens over time by model and key, and Logs shows every request with its tokens, price and cost.

Everything on the Models page: chat, embedding, image and video models from many labs. GET /v1/models lists the ones your key can call.

New models join the catalog as labs release them, with no change needed to your code. Watch the Models page, or poll GET https://ai.hyperflow.fun/public/models.

GET /v1/models lists only the models a key may call. If the key was limited to some models, edit it on API keys to allow more.

Where a model has more than one host, Flow AI tries another before returning an error. If none can answer, you get 503 with the code host_unavailable; retry shortly. See Errors.

Send your key as a bearer token: Authorization: Bearer sk-.... Create keys on API keys.

Chat completions, embeddings, image generation and edits, video generation and the model list, all under https://ai.hyperflow.fun/v1. See the API overview.

Up to 500 requests per second from one IP address. Keys have no request-rate limit of their own; what they spend is bounded by their spending limit and your balance. See Rate limits.

Parameters Flow AI doesn’t support are removed before the request reaches the model, so they have no effect. The API overview lists what each endpoint takes.

Set "stream": true and the answer arrives as server-sent events, ending with a chunk that carries the request’s usage. See Streaming.

The request settles on what the model had already read and written. You pay for those tokens and the rest of the hold is released.

Give it a spending limit when you create it, or later on API keys. Limits can reset daily, weekly or monthly, or be one-off.

Contact us on Discord or Telegram. Withdraw any $THINK you haven’t spent first.

Requests go to the model’s host to be answered. See the Privacy policy and Terms for how data is handled.