Overview

aiml is an LLM gateway. You send requests in the wire format you already use (OpenAI Chat Completions or Anthropic Messages), name a model as vendor/model, and the gateway picks a provider endpoint, streams the reply back, meters it and bills your prepaid balance.

What you get

• Two dialects, one key. POST /v1/chat/completions speaks OpenAI; POST /v1/messages speaks Anthropic. The official SDKs work unchanged; only the base URL and the key differ.
• Every model behind one catalog. Model ids are vendor/model (openai/gpt-5-mini, anthropic/claude-sonnet-4.6). GET /v1/models lists them with prices and capabilities.
• Routing you can see. The response tells you which provider served it (x-aiml-provider, x-aiml-region) and how many attempts it took (x-aiml-attempts). Send x-aiml-debug: 1 and x-aiml-route-trace explains the choice.
• Prepaid, integer billing. Costs are micro-USD integers, reserved before the call and settled after it; the settled cost travels in x-aiml-cost-micro and in the response's aiml.cost object. Nothing is rounded twice.
• A request id on everything. x-aiml-request-id (also the id of the response) works for GET /v1/generation/{id}, the console and support.

Base URL and keys

Keys carry scopes: inference for calls, read:usage, read:credits, manage:credits, manage:keys, manage:policies for management routes. Create them in the console or with POST /v1/keys.

Where next

Every snippet in these docs is an include from the cookbook, which runs against a staging cell every night. If a snippet stops working, the build of this site fails before you see it.