On this page

• Endpoints
• Not supported
• Behaviour that differs

router.africa works with the OpenAI SDK and the Anthropic SDK when you change the base URL and the key. Request fields that this site doesn’t mention are passed through as the model’s API defines them. This page lists where we differ.

Endpoints

You can call:

• POST /v1/chat/completions, with and without streaming
• POST /v1/messages, in the Anthropic format (see Messages)
• POST /v1/responses, in the OpenAI Responses format (see Responses)
• GET /v1/models
• GET /v1/key
• GET /v1/generation (see Look up a request)

Other endpoints are not part of the public API, even if a path seems to answer. This includes the OpenAI batch API, embeddings, images, audio and rerank.

Not supported

If you are coming from OpenRouter, these features don’t exist here:

Feature	Status
:free, :nitro and :floor model variants	Not supported.
~alias model names	Not supported. Use the full model id.
Routers such as openrouter/auto, pareto-router and fusion	Not supported.
provider preferences, sort, plugins and route request fields	Not supported.
models[] and fallbacks arrays, and session_id sticky routing	Not supported.
Presets	Not supported.
Bring your own provider keys	Not supported.
Workload identity tokens (/oauth/token)	Not supported.
Attribution headers	Not used.
Batch or async endpoint	Not available.
Management APIs for keys, workspaces and guardrails	Use the dashboard. They are not part of the API.
Regional hostnames and data-region routing	Not available.
Reading stored prompts and responses through the API	Not available. Admins read them in the dashboard.

Behaviour that differs

• Credits, not currency. You pay with credits. A request that can’t be paid for returns 402, not 429. OpenAI uses 429 for this, which the SDKs retry. A 402 isn’t retried. See Credits and billing.
• Error body. Errors use OpenRouter’s envelope: error.code is a number, and error.metadata.reason gives the specific cause. The SDKs work unchanged. On POST /v1/messages errors use Anthropic’s format instead. See Errors.
• Rate-limit headers. Only Retry-After is sent. There are no x-ratelimit-* headers.
• Anthropic SDK. Use the default api_key (apiKey in TypeScript). POST /v1/messages accepts x-api-key as well as Authorization: Bearer. No other endpoint accepts x-api-key.
• Generation lookup. GET /v1/generation returns metadata only. OpenRouter’s /generation/content and provider_responses[] don’t exist here.
• Headers we ignore. OpenAI-Organization, OpenAI-Project and OpenAI-Beta are accepted and ignored.
• Idempotency. Duplicate requests are caught automatically, and Idempotency-Key is an extension. See Idempotency and retries.
• max_tokens. A value above the model’s limit is lowered to the limit, on max_output_tokens too. It isn’t an error.
• Models. A model must be switched on, allowed for your key and have a published rate. Otherwise you get one 404 for all three cases.
• Data. Prompts and responses are not saved unless a workspace turns saving on. See Data and privacy.