Show HN: Open-Source Semantic AI Chat Search – 100% Locally (github.com)

A new open-source browser extension (Chrome/Firefox) provides 100% local semantic search over your AI chats from ChatGPT, Claude and Perplexity by turning conversations into vector embeddings in the browser and storing them in IndexedDB. The extension bundles an ONNX Runtime WebAssembly module (ort-wasm-simd-threaded.jsep.mjs) to generate embeddings locally, auto-syncs new conversations when you visit supported sites, converts queries to embeddings, and returns similar conversations with similarity scores — all without sending data to external servers. It requires Node.js 18+/Yarn to build and runs as a typical unpacked extension or CRX, and the project is available under CC BY-NC-SA 4.0. This is significant for the AI/ML community because it demonstrates practical, privacy-first on-device semantic search using WASM-based models and a browser-side vector store, lowering barriers to private retrieval-augmented workflows. Key technical implications: local embedding generation is CPU-intensive (initial indexing can take ~25–50 minutes per site), IndexedDB handles storage, and Chrome CSP constraints currently force use of 'wasm-unsafe-eval' (workaround: offscreen document) which blocks Web Store submission. Known limitations include missed old-thread capture, rate limits on chat fetches, and scalability concerns for large datasets — all useful signals for researchers and engineers working on edge/clientside ML and browser-native retrieval systems.