OctaRouter

Models & protocols

Models & protocols

OctaRouter is built so you never need separate accounts, billing, or API keys at each LLM vendor. One OctaRouter account, one prepaid balance, one API key, and one set of base URLs — then point any AI client, agent, SaaS backend, or workflow at OctaRouter.

One key, three protocols

OctaRouter exposes three HTTP protocols. They are not three vendors — they are three ways to call the same gateway with the same API key:

ProtocolBase URL (typical)When to use
OpenAI{GATEWAY_URL}/v1Default for every model. Use this when your tool expects OpenAI Chat Completions or you are unsure.
Anthropic Messages{GATEWAY_URL} (root)Recommended for Claude models when you want native Messages API features (streaming, tool use, etc.).
Gemini generateContent{GATEWAY_URL} (root)Recommended for Gemini models when you want Google's native REST shape.

Replace {GATEWAY_URL} with the proxy root shown on the home page or in your deployment env (GATEWAY_URL).

Which protocol for which model?

  • All models — callable via the OpenAI protocol (POST /v1/chat/completions with the model ID in the model field).
  • Claude / Anthropic models — also callable via Anthropic Messages (POST /v1/messages). Prefer this when your SDK or agent already speaks Anthropic.
  • Gemini / Google models — also callable via Gemini generateContent (POST /v1beta/models/{model}:generateContent). Prefer this when your client is Gemini-native.

For other vendors (OpenAI, DeepSeek, Mistral, etc.), stick with the OpenAI protocol — no need to configure vendor-specific base URLs.

Live model catalog

The Models page lists models enabled on your Gateway deployment (active routes), refreshed about every 60 seconds (configurable via GATEWAY_CATALOG_CACHE_SECONDS).

The portal server calls GET {GATEWAY_URL}/catalog/models — a public, unauthenticated discovery API that aggregates supported protocols from active model_routes. No user API key or master key is required for the catalog.

Each row shows:

  • Model ID — pass this in the model field (OpenAI) or path (Gemini).
  • Vendor — who provides the model.
  • Protocols — derived from runtime routes; the recommended badge marks the best native fit when Claude or Gemini applies.
  • Recommended base URL — copy this into your client when using the suggested protocol.

Portal-only copy (descriptions, featured badges, extra tags) lives in lib/models/catalog/_overrides.json and is merged on top of Gateway data.

If the list is empty, verify GATEWAY_URL on the portal server and that Gateway exposes GET /catalog/models with at least one active model route.

Optional env:

  • GATEWAY_CATALOG_CACHE_SECONDS — cache TTL (default 60).
  • GATEWAY_CATALOG_ROUTE_GROUPS — CSV filter passed to Gateway (default: all active route groups).

Quick integration checklist

  1. Sign up and top up credits (Quickstart).
  2. Create an API key in Dashboard → API keys.
  3. Pick a protocol and base URL from the table above (or from Models for a specific model).
  4. Set Authorization: Bearer YOUR_API_KEY (or the header your SDK expects).
  5. Change only the model name when switching LLMs — keep the same OctaRouter base URL and key.

See also: