Embeddings API

POST /v1/embeddings converts text into numeric vectors, using the same request and response shape as the OpenAI embeddings API. Use it for semantic search, clustering, deduplication, recommendation and retrieval pipelines.

#Quickstart

bash

curl https://api.radienthq.com/v1/embeddings \
  -H "Authorization: Bearer $RADIENT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "auto",
    "input": "Embeddings power semantic search, clustering and retrieval."
  }'

json

{
  "object": "list",
  "data": [
    {
      "object": "embedding",
      "embedding": [ -0.022395002, 0.0020443022, "… 1,536 values total …" ],
      "index": 0
    }
  ],
  "model": "…",
  "usage": { "prompt_tokens": 10, "completion_tokens": 0, "total_tokens": 10, "cost": 0.000001 }
}

model and the vector length are trimmed above: the response reports which embedding model served the request, and that model's vectors. With model: "auto" today the vectors are 1,536 dimensions.

#Request

Field	Type	Required	Notes
`model`	string	yes in practice	`"auto"` selects the default embedding model. Any other id is sent to the serving endpoint unchanged; an id the endpoint does not serve fails the request (see Errors).
`input`	string or array of strings	yes in practice	The text to embed. An array returns one vector per element, in order (`index` matches the position).
`encoding_format`	string	no	Accepted for OpenAI compatibility; returned vectors are floats.
`user`	string	no	Accepted for compatibility.

There is no server-side range checking: a missing model or input is forwarded as-is and the request fails at the serving step (a 500, below).

#Response

Field	Meaning
`object`	`"list"`
`data[]`	One entry per input: `{object: "embedding", embedding: [floats], index}`
`model`	The embedding model that served the request
`usage`	`prompt_tokens`, `completion_tokens`, `total_tokens` (the token count billed) and `cost` in USD

usage.cost is the amount deducted from your balance for the call. When a cost is absent, no figure was available; it does not mean the call was free.

#Batch example

bash

curl https://api.radienthq.com/v1/embeddings \
  -H "Authorization: Bearer $RADIENT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "auto",
    "input": ["first document", "second document", "third document"]
  }'

The response has three data entries, index 0–2. Keep batches to a size the serving model accepts; extremely large batches are refused by the endpoint.

#Errors

Status	Body	Cause
`400`	`{"error":"Invalid request format"}`	The body is not valid JSON
`401`	`{"error":"missing Authorization header"}` and the other auth bodies	No or invalid credential (Authentication)
`402`	`{"error":"insufficient credits"}`	The credit gate refused the call; see the credit gate
`500`	`{"error":"Failed to process embeddings request"}`	Every serving failure — unknown model, missing `input`, endpoint errors — is flattened into this one response. The upstream's own message is logged, not returned.

#Cost

Embeddings are billed per input token at the embedding model's rate, with no markup — the same figure is deducted from your balance that the model charges. A failed call records a zero-usage entry against your account, so failures do not consume credits.