Create embeddings

Embeddings use the same submit call with an embedding model. Two things differ from text: the order should buy no output units, and the result is an EmbeddingResult.

SubmitCopy Anchor Link

const handle = await client.submit({
  model: embeddingModel,
  input: { input: ["first passage", "second passage"], encoding_format: "base64" },
  sla: "async",
  rateIn,
  unitsOut: 0,        // an embedding has no output side to pay for
  provider,
});

unitsOut: 0 matters. Without it, a request that names no output-token ceiling is priced as a text job with a default ceiling of 4096 output units, and your payment authorization covers that at rateOut. See Units.

Read the vectorsCopy Anchor Link

import { EmbeddingResult } from "@vorq-ai/client-sdk";

const result = await handle.result();
if (result instanceof EmbeddingResult) {
  console.log(result.embeddings.length, "vectors,", result.promptTokens, "input tokens");
  const raw = result.bytes();   // one Uint8Array per vector
}

• .embeddings is the response's data array, as returned.
• .bytes() decodes vectors requested with encoding_format: "base64". For float output it raises; read .embeddings directly.
• .cost is promptTokens × rateIn ÷ 1 000 000 in atomic token units; an embedding has no output charge.

In a batchCopy Anchor Link

Set every line's url to /v1/embeddings and put units_out: 0 in each body. A batch holds one endpoint only, so embeddings and text go in separate batches. See Submit a batch.