RS InferUsage
Interactive reference for every HTTP endpoint, generated from the server's OpenAPI 3.1 schema.
These pages are generated from docs-site/openapi.yaml in the repository with
Fumadocs OpenAPI.
Each operation page includes the request/response JSON Schema, sample payloads,
and an interactive playground that hits the configured server URL
(default: http://localhost:8080).
Endpoint families
Embeddings
POST /v1/embeddings (OpenAI/vLLM shape) and POST /embed (TEI shape).
Rerankers
POST /v1/rerank, POST /rerank, POST /v1/score.
PII
POST /pii/detect and POST /pii/redact (with /v1/* aliases).
Classification
POST /classify/zero-shot and POST /classify/true-false.
Operations
GET /health, GET /v1/models, GET /metrics.
Error format
Errors return a JSON body shaped like:
{
"error": {
"message": "model not found: my-model",
"type": "model_not_found",
"code": 404
}
}| Status | When |
|---|---|
400 | Invalid body / parameters (input empty, wrong kind for endpoint, …). |
404 | Requested model not loaded, or a different kind than the endpoint. |
408 | server.request_timeout_ms exceeded. |
413 | Body larger than server.max_body_mb. |
429 | Too many waiters for the model (server.max_queue reached). |
503 | Waited longer than server.queue_timeout_ms for a session slot. |
Benchmarking
Load-test any endpoint with the HTTP benchmark client and read the results.
/v1/embeddings POST
Create embeddings (OpenAI/vLLM compatible). Accepts either a string, an array of strings, an array of token ids, or an array of token-id arrays. Returns one embedding per input in `data`, ordered by the `index` field. `dimensions` performs Matryoshka truncation and re-normalizes. `encoding_format: "base64"` returns each vector as a little-endian f32 blob.