@anvia/transformers

Generate text embeddings locally with Transformers.js. Build semantic search and retrieval workflows without sending document text to a hosted embedding service.

Install

pnpm add @anvia/transformers @anvia/core

Quickstart

The first load may download model files. Subsequent loads can reuse the local cache.

import { embedDocuments } from "@anvia/core/embeddings";
import { InMemoryVectorStore, retrieveDocuments } from "@anvia/core/vector-store";
import { loadTransformersEmbeddingModel } from "@anvia/transformers";

const model = await loadTransformersEmbeddingModel({
  modelId: "Xenova/all-MiniLM-L6-v2",
});

try {
  const { documents } = await embedDocuments({
    model,
    documents: [{ id: "password-reset", text: "Reset links expire after 30 minutes." }],
    id: (document) => document.id,
    content: (document) => document.text,
  });
  const store = InMemoryVectorStore.fromDocuments({ documents });

  const results = await retrieveDocuments({
    store,
    model,
    query: "How long does a reset link last?",
    topK: 3,
  });
  console.log(results);
} finally {
  await model.close();
}

What it supports

• Configurable pooling, normalization, batching, device, and dtype.
• Model caching and local-files-only loading.
• Existing Transformers pipelines through adaptTransformersEmbeddingModel().
• Explicit cleanup with close() or await using for loaded models.

Reuse the model across requests. Adapted pipelines remain caller-owned; loaded models wait for active inference before disposing their runtime.

Learn more

• Usage guide
• Anvia overview
• Contributing