Embeddings

The Embeddings endpoint turns text into vectors you can store and compare, for search, clustering or recommendations. It is OpenAI-compatible: same request, same response.

Create embeddings

import Glexir from "@glexir/sdk";

const glexir = new Glexir({ apiKey: process.env.GLEXIR_API_KEY });

const res = await glexir.embeddings.create({
  model: "text-embedding-3-small",
  input: ["The food was delicious.", "The service was slow."],
});

const vector: number[] = res.data[0]?.embedding ?? [];
console.log(vector.length, res.usage.total_tokens);

Response:

{
  "object": "list",
  "data": [
    { "object": "embedding", "index": 0, "embedding": [0.0023, -0.0093, 0.0158] },
    { "object": "embedding", "index": 1, "embedding": [-0.0117, 0.0041, 0.0212] }
  ],
  "model": "text-embedding-3-small",
  "usage": { "prompt_tokens": 11, "total_tokens": 11 }
}

Vectors are shortened here. data[i].index matches the position of the input.

Models

The price book currently sells text-embedding-3-small and text-embedding-3-large. Call models.list() for the live list. A model that is not priced returns 403 with code model_not_priced. A model that is not an embedding model returns 400 with code model_not_found.

Use dimensions to get shorter vectors from the text-embedding-3 models:

import Glexir from "@glexir/sdk";

const glexir = new Glexir();

const res = await glexir.embeddings.create({
  model: "text-embedding-3-large",
  input: "Glexir",
  dimensions: 256,
});
console.log(res.data[0]?.embedding.length); // 256

Encoding format

When you leave out encoding_format, the SDK asks for compact base64 on the wire and decodes it into number[] for you. OpenAI's SDK does the same. Base64 responses are smaller than JSON floats.

If you set encoding_format, the SDK sends it unchanged and does not decode:

• "float": embedding is a number[].
• "base64": embedding is a base64 string of little-endian float32 values. The return type is CreateEmbeddingResponse<string>.

import Glexir from "@glexir/sdk";

const glexir = new Glexir();

const raw = await glexir.embeddings.create({
  model: "text-embedding-3-small",
  input: "hello",
  encoding_format: "base64",
});
const encoded: string | undefined = raw.data[0]?.embedding;

Over plain HTTP (curl), the API returns floats unless you send "encoding_format": "base64". The gateway passes either format through untouched.

Method

glexir.embeddings.create()

import type { APIPromise, CreateEmbeddingResponse, EmbeddingCreateParams, RequestOptions } from "@glexir/sdk";

// The method on glexir.embeddings, with its two overloads
interface Embeddings {
  create(body: EmbeddingCreateParams & { encoding_format: "base64" }, options?: RequestOptions): APIPromise<CreateEmbeddingResponse<string>>;
  create(body: EmbeddingCreateParams, options?: RequestOptions): APIPromise<CreateEmbeddingResponse>;
}

POST /embeddings. options are the per-request options described in Client configuration.

EmbeddingCreateParams also accepts other keys, which are sent as-is.

Response

Billing

Embeddings are billed per input token at the model's embedding rate, from usage.prompt_tokens. There is no output charge. Before the call, Glexir reserves an estimate of the input cost as a hold and settles it against the real usage when the response arrives. See Credits & billing and Pricing.

Through Batches (/v1/embeddings), embeddings cost half the standard rate.

Errors

See HTTP error codes for the full list.