fasttextembed (Node / JS)

Fast, dependency-free text embeddings for BAAI/bge-small-en-v1.5 — the pure-C engine compiled to WebAssembly (runs in Node and the browser). The model (~64 MB) downloads and caches on first use.

npm install fasttextembed

const { TextEmbedding } = require("fasttextembed");

const model = await TextEmbedding.create();      // downloads + caches the model on first call
const vectors = model.embed(["hello world", "fast"]); // array of 384-float vectors
const one = model.embedOne("hello world");

• embed(texts) — batch; embedOne(text) — single.

• WASM backend is single-threaded and portable; an optional native N-API addon is used automatically if present (multi-threaded, SIMD, ~2× faster — Node-only). Build it with:

  npm run build:native   # compiles native/ via node-gyp (needs a C compiler + Python)

  Once built, require("fasttextembed") loads the native addon and transparently falls back to WASM if it isn't available.

• Env: FTE_MODEL_DIR (local model.fte/vocab.tsv), FTE_MODEL_URL, FTE_CACHE.

In the browser, pass model bytes directly: TextEmbedding.create({ fte: Uint8Array, vocab: Uint8Array }).