Retrace — your research trail
https://kartik-singhal26.github.io/retrace/ · Workflow & Planning
Media
2 items
Description
The page you read three weeks ago and didn't save — back when you need it. Fully offline: your history never leaves this device. The page you read three weeks ago and didn't save — back when you need it. Retrace is a memory prosthetic for researchers, grad students, and anyone who reads a lot. It quietly indexes your own browser history on your device and does two things: RELATED — open the side panel on any page and see pages you've ALREADY visited that relate to it: the paper you skimmed last month, the blog explainer from that late-night session, the docs page you swear you read. Results are grouped by "this session" (pages you previously visited together) and "earlier", with your own past searches shown separately as compact chips. TRAILS — your browsing, clustered into topic trails. Each trail groups the sessions you spent on a subject ("attention · transformer · paper", "sourdough · starter · bread") and reopens the whole set in one click. Close your 40 tabs; Retrace can get them back. PRIVATE BY ARCHITECTURE, NOT BY PROMISE - Zero network permissions — verifiable in the manifest. Retrace cannot phone home. - No accounts, no servers, no telemetry, no analytics. - The index lives in your browser's local storage. Delete it anytime. - Mail, social, and banking sites are excluded by default; add your own exclusions. - Incognito is never touched. - Deep indexing of page text is OFF by default and opt-in per domain (e.g. arxiv.org) — Chrome asks your permission for each domain you add. HOW MATCHING WORKS Retrace scores your history against the page you're on using keyword relevance (BM25), recency, co-visitation (pages you visited in the same sitting), and visit frequency — all computed locally. No AI cloud, no data leaves your machine. WHO IT'S FOR Researchers and grad students doing literature work, students assembling sources for essays, and anyone who has ever typed "that article about..." into Google trying to re-find something they already read. Free and open source.