Build on models,
not on vendors.

Every provider ships its own SDK, its own auth, its own quirks. That is fine for them and expensive for you. One API key and one endpoint should reach every model, so switching is a string change, not a migration.

You should not have to learn a new wire format to use a new model. Call any model through OpenAI Chat Completions, OpenAI Responses, or Anthropic Messages, and we handle the translation in between. Your existing clients keep working.

No plans, no seats, no minimums, no annual commitments. Top up once and every request is billed per token at the listed rate. If you stop calling, you stop paying.

Every rate is published before you make a single call: input, output, cache, and context window. When a model is discounted, we show the normal price next to it. When a price changes with the time of day, we label it. You should never have to reverse-engineer an invoice.

We route requests, we do not mine them. By default we keep the metadata needed to bill and debug, such as timestamps, token counts, and latency, and not the content of your prompts or responses. See our Privacy Policy for the details.

Upstream providers go down. Pretending otherwise helps nobody. We track the health of every model and publish its recent uptime, so you can decide with real numbers instead of marketing claims.

Clear errors over vague ones. Docs you can copy from. A playground to try a model before you write a line of code. We ship what we would want to use ourselves, and we cut what we would not.