AI & Autonomous Agents•7 min read•2026-10-15
Offline Semantic Search: In-Browser Vector Embeddings with Transformers.js and Voy
Building fast, client-side semantic search engines using miniLM embeddings, cosine similarity, and zero external vector databases.
C
Castov Engineering
Core Platform & DevSecOps Team
Table of Contents
Quantized Embedding Models in the Browser
Embedding models like all-MiniLM-L6-v2 convert text into 384-dimensional dense vectors. Quantized to 8 bits, this model occupies less than 30MB of memory and executes inference in under 15ms per chunk.
Interactive Castov Tools for This Topic
Local-First & Zero-Telemetry Computing
Automate Your Development Workflows Without Cloud Telemetry
Castov provides 61+ client-side utilities, local secret vaults, and CLI tools that execute entirely in your browser or native Windows workstation. Zero tracking, zero tokens sent to third-party clouds.