Turn text into vectors
Turns text into vectors for search, retrieval, clustering and deduplication in the OpenAI embeddings shape, served by openrelay/bge-m3 today. openrelay/bge-m3 is multilingual and returns 1024-dimension dense vectors; check a model's supported_apis on GET /v1/models for embeddings. Requires a funded balance. No output rate: billed only on usage.prompt_tokens. stream is dropped. Limits: 8192 tokens per input item, 1024 items per request, 10 MiB request body. Also runs through the Batch API at endpoint /v1/embeddings, at the batch rate.
Required permission
Requires the inference.invoke permission. See Permissions for the roles and API key permissions that include it.
Command line
The orl CLI wraps this endpoint:
orl embeddings create --input <input> --model <model>apiKeyAuthorizationBearer <token>OpenRelay API key. Send it as Authorization: Bearer or_… (legacy vl_… keys still work).
application/json- body
An OpenAI-compatible embeddings request. stream is dropped; a response is always a single JSON body.
model*stringAn OpenRelay public model id, e.g. openrelay/bge-m3.
input*|array<string>The text to embed. An array embeds each item independently and returns one vector per item, in order.
[key: string]?anyOK
application/json- response
An OpenAI embeddings list object (opaque; additionalProperties carries the rest of the dialect).
object?"list""list"data?array<>model?stringusage?openrelay?What the request is billed, on every online inference response.
[key: string]?anycurl -X POST 'https://inference.openrelay.inc/v1/embeddings' \ -H 'Content-Type: application/json' \ -d '{ "model": "string", "input": "string"}'{ "object": "list", "data": [ {} ], "model": "string", "usage": {}, "openrelay": { "cost_nano_usd": 0, "priced_at": "string" }}Overview
Turn text into vectors for search, retrieval, clustering and deduplication, OpenAI-compatible, from the multilingual openrelay/bge-m3. Served by the inference endpoint (inference.openrelay.inc).
Overview
Answer typed questions about an input (a choice, a score or a yes/no) with a classifier instead of a generator, online only and billed per input token. Served by the inference endpoint (inference.openrelay.inc).