Semantra: Serverless Semantic Search Running Entirely in Your Browser via WebGPU
init0 · reddit · 2026-08-01
A developer introduced Semantra, a semantic search tool that runs entirely locally within the browser.
Key Features:
- Zero Server Dependency: Executes text chunking, embedding, and cosine similarity ranking directly on the client side via WebGPU (with a WASM fallback).
- Lightweight Model: Bakes in Snowflake's Arctic-Embed-S model (33M parameters, 30MB quantized) and supports local caching for instant reloads.
- Easy Integration: Provides an npm package and React Hook for quick integration into frontend projects.
The author tested it on 100 Wikipedia articles and noted excellent performance given the complete absence of server infrastructure.
More from coding & agent
- Building Custom Eval Benchmarks for Code Agents Using Real-World PRs — Hacubu · 2026-08-01
- Launching AI Agent Trading Hackathon with $20k Prize Pool — cardosofede · 2026-08-01
- Y Combinator Open-Sources QM, Its Internal Multi-Agent Harness — ycombinator · 2026-08-01
- Dev Opinion: One-Shot Your Entire Codebase Today If Requirements Are Clear — shauseth · 2026-08-01
- Dev Builds Aura Local Agent: Demoting LLMs to a Voice Box Component — bryany97 · 2026-08-01
- The Compute Bottlenecks of Agentic AI: Inference vs. Execution — charles_irl · 2026-08-01