Ternlight

An embedding model in 7 MB.

Runs on your CPU. No API.

Drop it in, embed text in milliseconds, and never call a server.

Runs in your browser

no API call

Engine + weights · 7 MB

mini variant · 5 MB

Fast embeddings · ~5 ms

CPU only · no GPU

USE IT

Three lines to semantic search

Ships as a single npm package. No model download step, no server.

npm ↗GitHub ↗

$npm install @ternlight/base⧉

example.js⧉

import { embed, similar } from '@ternlight/base';

similar('easy weeknight dinner ideas', recipes, { topK: 3 });
// → ranked matches · ~5 ms · zero network

EXAMPLE

React docs search

Search React's docs in your browser. Type a question — powered by @ternlight/mini, the 5 MB tier.

Embedding…

CHUNKS

24 emb/sec

ON YOUR CPU

3.1 s

ELAPSED

›—

ASK share state across components skip a re-render run code after render avoid prop drilling form input value

EMBED LATENCY

—ms

single embed() call (@ternlight/mini)

⧉Copy query embedding

MODEL

Embedding—Linear weights ternary (BitLinear)Transformer layers—Attention heads—d_model—Output—Max input—

PACKAGE

Runs on@ternlight/mini↗Bundle—Engine init—Throughput—License MIT After first load cached (0 network)

React docs from reactjs/react.dev · CC BY 4.0 · Demo built with ternlight