Hi PH — I built PDFLoom, a PDF editor that runs entirely in the browser: pdf.js for rendering, pdf-lib for mutation, and a small local AI suite (t5-small for summarization, running via transformers.js on WebGPU with a WASM fallback) for summarize/translate/redact/chat-with-your-PDF. No file ever leaves the device, no API key, no account. The interesting constraint: there's no way to locate-and-rewrite an existing text run in a PDF's content stream from the client, so "editing" text works by covering the original with a new opaque appearance at the same spot rather than true content-stream editing — the UI says so explicitly rather than pretending otherwise. Would love feedback, especially from anyone who's fought pdf.js/pdf-lib edge cases before.

PDFLoom

A premium PDF editor with real AI — 100% private, free

A premium PDF editor with real AI — 100% private, free