Embeddings

An embedding is a list of numbers that captures the meaning of a piece of text. Texts that mean similar things land near each other in vector space, which is what makes semantic search, retrieval-augmented generation (RAG), deduplication, clustering, and classification work.

Embed some text

Pass a single string or an array of strings as input. You get back one vector per input, in the same order. Embed your documents once and store the vectors; at query time, embed the query and find the nearest stored vectors.

from pendra import Pendra

client = Pendra()
result = client.embeddings.create(
    model="nomic-embed-text",
    input=["the quick brown fox", "jumps over the lazy dog"],
)
for item in result.data:
    print(item.embedding)

import Pendra from 'pendra';

const client = new Pendra();
const result = await client.embeddings.create({
  model: 'nomic-embed-text',
  input: ['the quick brown fox', 'jumps over the lazy dog'],
});
for (const item of result.data) {
  console.log(item.embedding);
}

[ 0.0131, -0.0442,  0.0921, … ]   # "the quick brown fox"
[ 0.0204, -0.0118,  0.0815, … ]   # "jumps over the lazy dog"

One vector per input — two strings in, two vectors out, in the same order. Each is truncated here (…); a full vector has hundreds or thousands of dimensions, depending on the model.

Choosing a model

Pendra ships several embedding models — Qwen3-Embedding (0.6B/4B/8B), EmbeddingGemma, BGE-M3, and nomic-embed-text among them. Dimensionality and similarity behaviour differ per model, so pick one and embed your whole corpus with it consistently. Browse the options at /models?type=embedding.

Batching

Embedding many strings at once is far faster than one request each. Send a few hundred per request rather than one huge batch — a single request should complete quickly (it can run up to ~30 minutes before timing out). For big wire payloads, set encoding_format: "base64" to shrink the response.