From OpenRouter

OpenRouter users have the shortest path: model ids are already vendor/model and the route is already /v1/chat/completions.

The changes

Base URL https://openrouter.ai/api/v1 becomes https://api.ai.ml/v1.
Key sk-or-… becomes aiml-live-….
Headers HTTP-Referer and X-Title are not needed. Put the app name in aiml.metadata.app if you want it in analytics.

cookbook/python/streaming.py#L5-L7

from openai import OpenAI

client = OpenAI(base_url=os.environ.get("AIML_BASE_URL", "https://api.ai.ml") + "/v1", api_key=os.environ["AIML_API_KEY"])

OpenRouter	aiml
`provider: { order: [...], allow_fallbacks }`	`aiml.route.providers.order` and `aiml.route.fallbacks`
`provider: { only: [...], ignore: [...] }`	`aiml.route.providers.only`, `aiml.route.providers.ignore`
`provider: { data_collection: "deny" }`	`aiml.route.require.zdr: true` (or `data_policy: "no_train"`)
`provider: { sort: "price" }`	`aiml.route.sort`
`provider: { max_price }`	`aiml.route.require.max_price`
`models: [...]` (model fallbacks)	one model per request; `aiml.route.fallbacks` moves between endpoints of that model
`transforms: ["middle-out"]`	not offered; the gateway returns `context_length_exceeded` rather than silently truncating
`GET /api/v1/generation?id=`	`GET /v1/generation/{id}` (the id is `x-aiml-request-id`)
`GET /api/v1/models`	`GET /v1/models`; per-model endpoints at `GET /v1/models/{vendor}/{model}/endpoints`
`x-openrouter-*` headers	`x-aiml-*` headers, see curl quick start

Prices are shown per million tokens in the catalog and billed as integer nano-USD per token; each usage line is rounded once. aiml.cost.lines in the response breaks the total down.
The balance is reserved before the call from your estimate and settled from the provider's usage; you never see a charge larger than the settled cost.
Top-ups are in your billing currency (INR via Razorpay with GST invoices, otherwise USD via Stripe); credits are always USD. See Billing.