Docs

Embeddings

A vector per input, OpenAI-compatible. No streaming, and charged on input only.

A vector per input, OpenAI-compatible. No streaming — one request, one answer. The same ir- key and the same wallet.

curl https://iranrouter.com/v1/embeddings \
  -H "Authorization: Bearer ir-..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/text-embedding-3-small",
    "input": ["سلام دنیا", "hello world"]
  }'

A few things to know

• input may be a string, an array of strings, or an array of tokens. You are charged on input only; there is no completion.
• The response is the provider's own, plus usage.cost_rial and an x-request-id header — the same id you will see under Usage.
• Not every catalogue model is an embedding model. A model that cannot serve this path answers 503 with no_capacity.
• With Idempotency-Key a retry is never billed twice. If the original response was too large to store for replay — a large embeddings batch usually is — the retry gets a 409 with code idempotency_completed_unreplayable, meaning the original completed — you choose, with a fresh key, whether a re-run is worth paying for.