Every model, tool and data source — one endpoint each.

Search, market data, RPC and sandbox APIs

AI tools API questions: one endpoint, pricing, keys, agents

What is an AI tools API?

A single HTTP endpoint that answers for a whole catalog rather than one capability. Image generation, video, speech, music, web search, market data, RPC and sandbox execution are all reached the same way, with the same auth story and the same billing, so adding a capability to your product is a new path rather than a new vendor.

Do I need a separate API key for each tool on the endpoint?

No, and on the default path you need no key at all. One wallet reaches every tool listed here; the alternative is an API key that covers all of them at once, rather than one key, one dashboard and one invoice per vendor.

How is a call to a tool API priced?

Per request, in dollars, by what the request actually asks for — the size of an image, the seconds of video, the characters of speech, the tokens of a completion. Nothing is bundled into a plan, so the price of a call does not depend on how many other calls you made this month.

Can I see what a call will cost before I send it?

Yes. Send the request without payment and the endpoint answers with HTTP 402 carrying the exact price for that exact request. A client reads the figure and decides, which is what lets an agent stay inside a budget instead of discovering the cost afterwards.

Which tool APIs are free to call?

The cards above say so directly: one that reads Free for callers takes no payment header at all. Catalog, price and history reads are generally in that group, because an agent should be able to look before it spends.

Can one AI agent use every tool here with one wallet?

That is the design. The agent funds a wallet once and pays each call as it makes it, so a workflow that searches the web, writes a summary, renders an image and reads a price does not need four accounts, four keys or four signups along the way.

Should I browse by job or by vendor to find the right API?

By job, if you know what you want done — these pages are named for the task and each one lists every model or venue that can do it. By vendor, if you arrived already knowing the name you want; the services index is organised that way instead.

AI tools API vs wiring up each vendor's own API directly?

Direct is fine for one vendor. The cost shows up at the second and third: separate keys to rotate, separate quotas to watch, separate billing to reconcile, separate failure shapes to handle, and a catalog that moves under you. One endpoint makes those one problem instead of a growing list.

Is there an OpenAI-compatible path into these tool endpoints?

The model endpoints speak the OpenAI protocol, so existing clients and agent frameworks reach them with a base-URL change. The data and execution endpoints are plain REST, documented per tool, because there is no OpenAI shape to be compatible with.