Quickstart

What you need

Two things: an account with some credit on it, and an API keyAPI keyA long password your software uses to prove a request is yours. Anyone who has it can spend your credit, so treat it like a card number.Shown once, at creation. We keep only a hash, which means we genuinely cannot recover it for you — a lost key has to be replaced.. Both take about a minute from signup. A key is shown to you exactly once, at the moment it is created — we store a SHA-256 hash of it and nothing else, so there is no screen that can show it to you again.

The call

The base URL is https://api.multigrid.ai/v1. Everything is JSON over HTTPS and the key travels as a bearer token.

curl https://api.multigrid.ai/v1/chat/completions \
  -H "Authorization: Bearer $MULTIGRID_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-5-nano",
    "messages": [{"role": "user", "content": "Explain what a gateway does, in two sentences."}]
  }'

That is the whole integration. The request body is OpenAI’s, so any client already written against /chat/completions works once you change two strings — see SDKs and clients.

What comes back

An ordinary OpenAI completion, with one extra object bolted on: multigrid, which says what actually happened to your request.

{
  "id": "req_…",
  "object": "chat.completion",
  "created": 1785000000,
  "model": "openai/gpt-5-nano",
  "choices": [
    {
      "index": 0,
      "message": { "role": "assistant", "content": "A gateway sits …" },
      "finish_reason": "stop",
      "logprobs": null
    }
  ],
  "usage": {
    "prompt_tokens": 21,
    "completion_tokens": 48,
    "total_tokens": 69,
    "prompt_tokens_details": { "cached_tokens": 0 }
  },
  "multigrid": {
    "model": "openai/gpt-5-nano",
    "provider": "openai",
    "attempts": 1,
    "cost_usd": 0.0000202,
    "cost_micros": 20,
    "byok": false,
    "request_id": "req_…"
  }
}

cost_micros is the number that gets billed — an integer count of millionths of a dollar. cost_usd is the same figure as a decimal, for humans. The same facts also ride on the response headers, which is usually easier to log:

Choosing a model

Model ids look like openai/gpt-5-nano or anthropic/claude-sonnet-5 — vendor, then model. The full list, with prices, is public and needs no key:

curl https://api.multigrid.ai/v1/models

If you would rather not choose, pass multigrid/auto and we pick the cheapest model that can actually handle the request. The response tells you which one it picked and why.

Next steps

• Set a spending limit before you put a key anywhere it might leak.
• Add a fallback model so one provider having a bad hour is not your outage.
• Send an Idempotency-Key on anything your client might retry.