Create an embedding vector

Compatible with OpenAI POST /v1/embeddings. input may be a single string or an array of strings; when masking is on, personal fields detected in the text are replaced with placeholders before the vector is computed. The returned vector is therefore the vector of the masked text.

Two consequences follow, and both should be known upfront:

The same document always yields the same vector. Placeholders on this endpoint are deterministic (<PERSON_1>), so re-embedding a document keeps your index consistent.
Masked entities cannot be told apart. Two documents that differ only in a person’s name produce the same vector (measured: similarity 1.000), and a query naming a person scores both equally. Search does not fail; it returns the wrong person’s document just as readily. Topic search is unaffected. Indexing and querying must go through the same path.

Pre-tokenized input (an array of token ids) is not text and cannot be masked, so it is rejected fail-closed with 415.

POST/v1/embeddings

Request body

requiredapplication/json

modelstringrequired

Embedding model id (defined in the panel).

inputstring | string[]required

Text or array of texts to embed. Token id arrays are not accepted (they cannot be masked and are rejected with 415).

Show properties

One of:

string

string

string[]

Array of string

string

dimensionsinteger

Output vector size, when the model supports it.

min 1

encoding_formatstring

default: "float"

Allowed:floatbase64

Responses

200List of vectors

objectstring

modelstring

dataobject[]

Show properties

Array of object

objectstring

indexinteger

embeddingnumber[] | string<byte>

A number array when `encoding_format` is the default (`float`); a base64-encoded STRING when `base64` is requested. Your client must handle both shapes.

Show properties

One of:

number[]

Array of number

number

string<byte>

string<byte>

usageobject

Show properties

prompt_tokensinteger

total_tokensinteger

401Missing or invalid virtual key

errorobject

Show properties

messagestring

typestring

413Input exceeds the masking limit. The request is not forwarded; split the text and retry.

errorobject

Show properties

messagestring

typestring

415Input is pre-tokenized (an array of token ids). It is not text and cannot be masked, so the request is **fail-closed** and never reaches the provider. Send text instead.

errorobject

Show properties

messagestring

typestring

429Quota or rate limit exceeded

errorobject

Show properties

messagestring

typestring

502The masking step could not run. The request is **fail-closed** and never reaches the provider; there is no "send it unmasked" option.

errorobject

Show properties

messagestring

typestring

Request

import requests

response = requests.post(
    "https://gw-tr.gurubase.io/v1/embeddings",
    headers={
        "Content-Type": "application/json"
    },
    json={
  "model": "text-embedding-3-small",
  "input": "string",
  "dimensions": 0,
  "encoding_format": "float"
},
)

curl -X POST "https://gw-tr.gurubase.io/v1/embeddings" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "text-embedding-3-small",
  "input": "string",
  "dimensions": 0,
  "encoding_format": "float"
}'

const response = await fetch("https://gw-tr.gurubase.io/v1/embeddings", {
  method: "POST",
  headers: {
    "Content-Type": "application/json"
  },
  body: JSON.stringify({
  "model": "text-embedding-3-small",
  "input": "string",
  "dimensions": 0,
  "encoding_format": "float"
})
});

Response

{
  "object": "list",
  "model": "text-embedding-3-small",
  "data": [
    {
      "object": "embedding",
      "index": 0,
      "embedding": [
        0
      ]
    }
  ],
  "usage": {
    "prompt_tokens": 0,
    "total_tokens": 0
  }
}