Semantic search on client with 5 lines of code, embeddings included.
We use an embedding model that outperforms OpenAI's text-embedding-ada-002 and can compute embeddings without any issues even in an apple watch!
Pay no one to get started, if u scale pay 10-20x less to compute/store/search 1M embedding with our API.
Client Vector Search provides client-side semantic search with embeddings in just five lines of code.
- for
- JavaScript/TypeScript developers needing fast, low-latency vector search in the browser or edge devices.
- pricing
- freemium
- license
- MIT
works withnpmTypeScript
Key features
6 features of Client Vector Search
- One-line integration — Install via npm and add semantic search with only five lines of TypeScript code.
- On-device embeddings — Compute embeddings locally, outperforming OpenAI's ada-002 model, even on an Apple Watch.
- Fast vector index — Search up to 100 K vectors in under 100 ms without network latency.
- Scalable API — Optional API lets you embed, store, and search up to 10 M vectors for $20 /mo.
- Persistable indexes — Save created EmbeddingIndex objects and reuse them across sessions.
- Playground demo — Interactive UI to test embedding speed and search against a Wikipedia dataset.
Use cases
- Add instant semantic search to a client-side web app
- Enable offline recommendation or similarity lookup on edge devices
- Build fast, low-cost search for small to medium datasets without server infrastructure
Client Vector Search FAQ
Do I need a backend to use Client Vector Search?+
No, the library runs entirely in the browser or any JavaScript runtime; a backend is only needed for the optional scaling API.
What embedding model is used?+
It uses a proprietary model that the provider states outperforms OpenAI's text-embedding-ada-002.
How many vectors can I store locally?+
Up to 100 K vectors can be indexed and searched locally with sub-100 ms latency.
What are the costs for larger datasets?+
The paid API plan costs $20 per month and supports up to 10 M vectors for embedding, storage, and search.
Can I run this on low-power devices?+
Yes, the embeddings can be computed on devices like an Apple Watch.
Summarized by DevHunt from clientvectorsearch.com · Sep 28, 2026. Details may change; check the official site.
.png?width=64)
