The browser is becoming a data engine. In-browser databases and on-device AI are no longer niche experiments—SQLite in WebAssembly, and AI/RAG workflows via WebGPU, are driving the narrative [1][2].
In-Browser Databases and SQLite in WebAssembly — A Show HN project demonstrates a WebAssembly version of SQLite running entirely in the browser, powering a mailbox viewer with filtering, sorting, searching, and a Gmail-like thread view. It’s shareable without a server [1].
AI/RAG Pipelines in the Browser with WebGPU — WebPizza runs a full RAG pipeline in-browser, with no backend. Models include Phi-3, Llama 3, and Mistral 7B; embeddings come from Transformers.js, and IndexedDB serves as the vector store with PDF.js for parsing [2]. Performance hints land in the mid-range: ~3-6 tokens/sec on WebLLM, rising to ~12-20 tokens/sec with WeInfer, and Llama 3.2 1B clocks ~8-12 tokens/sec [2].
Browser-Scale Data Parsing — Limits and Experiments — The Ask HN thread asks how far the browser can push as a data engine beyond visuals, spotlighting open-source tests with Hyparquet and HighTable that probe browser-native data processing and its tradeoffs [3].
SQLite Ecosystem and Updates — SQLite 3.51 is out [4].
Closing thought: the browser is inching toward a fuller data stack, but practical limits and security considerations still shape how far we go.
References
Showcases an in-browser SQLite via WebAssembly for an agent mailbox viewer; exports/shares mailbox with GitHub Pages deployment and workflow ease
View sourceShow HN: WebPizza – AI/RAG pipeline running in the browser with WebGPU
Proof-of-concept runs Retrieval-Augmented Generation entirely in browser; uses WebGPU, IndexedDB vector store, embeddings, and local models; seeks feedback on optimizations.
View sourceAsk HN: How far can we push the browser for large-scale data parsing?
Explores browser-based data processing beyond visuals toward a real data engine; discusses limits, tradeoffs, and alternatives to backend stacks today
View sourceSQLite 3.51 Is Out
Announcement of SQLite 3.51 release with release notes link
View source