Reshelve — AI Bookmark Organizer

Reorganize your Chrome bookmarks with AI — every change is previewed, confirmed, and undoable

Bookmarks pile up into the hundreds or thousands, folders get messier by the month, and finding anything means searching — Reshelve uses AI to reorganize them for you. It reorganizes your native Chrome bookmarks, not some separate system. When it's done, your bookmarks bar is still your bookmarks bar, and sync across your devices works exactly as before. ■ Four things it does · AI organize A model reads the folders you checked, proposes a folder tree, and places every bookmark. You approve it move by move before anything happens. · Local cleanup — no model needed Duplicates, empty folders, and bookmarks you stopped opening long ago. None of this involves a model or a network request, so it works before you have set up an endpoint at all, and works just as well if you never set one up. · Stats Which sites your bookmarks actually come from, broken down by folder. Optionally ranked by how often you really visit them — including the pages you keep going back to and never saved. · Import and export Move bookmarks in and out as a file, including a format other browsers can read. ■ You're in control, not the AI · You choose the scope Check the folders you want reorganized. Unchecked folders are never read and never modified — no bookmark moves out of them, and none moves in. · Review every move Before anything is applied, every move is listed for you to see: where each bookmark came from, where it's going, and why. Cancel any you disagree with individually, or filter in bulk by confidence. · Undo in one click Not happy with the result? Restore everything to how it was before, in one click. ■ Cleanup, without any model at all The cleanup side never calls a model. Three of the four scans below make no network request of any kind, and none of them needs a permission beyond the bookmarks access the extension already has. If you never configure an endpoint, this half of the extension still works. · Duplicate bookmarks Graded rather than lumped together: identical URLs, addresses that differ only in ways that don't change the page, and near-matches flagged as merely suspicious. You pick which copy to keep; the grade tells you how much to trust the match. · Empty folders Folders left holding nothing, including the ones that will only become empty once the rest of this cleanup runs. · Bookmarks you stopped opening Sorted into 3+ months, 6+ months, over a year, and over two years. This reads the last-opened timestamp Chrome already keeps on each bookmark — it does not read your browsing history and asks for no extra permission. Bookmarks Chrome has no timestamp for are shown separately rather than guessed at. · Dead links Checks whether each bookmarked page still answers. This one does make network requests — that is the whole point — so it is the one feature that asks for access to all websites, only when you start it, and skipping it leaves the other three fully working. Anything found can be deleted, or moved to a folder for you to go through later. Cleanups are undoable too. ■ Where your bookmarks actually come from A board of the domains behind your collection, with the folders each one is scattered across. Long paths that only differ by an ID are folded together so a hundred rows from one site don't drown out everything else. Grant the optional history permission and it can also rank by visits: which bookmarked sites you still return to, which have gone cold, and — the useful one — which pages you visit constantly and never bookmarked. Choose the window from the past month out to all time. The counting happens on your device and nothing is transmitted. ■ Import and export · Export Selected folders to JSON, either as the full folder structure or a flat list of links. Or to an HTML bookmark file with site icons included — the format Chrome, Firefox, Safari and Edge all import. · Import Read back a bookmark file someone shared with you. Before anything is written, you see what's inside and how many you already have. Everything lands in one new folder — your existing structure is untouched. Imported files come from someone else, so Reshelve blocks any javascript: and data: links inside them — that kind of "bookmark" runs a script on whatever page you're on when clicked. Blocked entries are called out explicitly, not silently dropped. ■ Bring your own model — only the AI half needs one Reshelve has no server of its own and doesn't collect any fee. For the AI reorganization you provide your own endpoint: · The official OpenAI API · Any OpenAI-compatible service (DeepSeek, Moonshot, a self-hosted proxy, etc.) · Ollama or LM Studio running locally on your machine — your data never leaves your computer Your API key is stored locally in your browser only. ■ Privacy, specifically "We value your privacy" doesn't mean much on its own, so here's what's actually verifiable: · URLs sent to the model are trimmed. Query parameters (everything after ?), anchors, and any username/password embedded in the URL are stripped before anything is sent — only the domain and path remain. Search terms and tracking identifiers never leave your browser. · Your bookmarks leave the browser in exactly two situations, both of which you start yourself: an AI reorganization sends the trimmed bookmark data to the endpoint you configured, and the dead-link check requests each bookmarked page to see whether it still answers. Nothing else makes an outbound request, and there is no analytics, telemetry, or tracking code anywhere. · No host access is granted when you install. The model endpoint asks only for the single domain you typed — enter api.openai.com and it only ever asks about api.openai.com. The dead-link check does ask for all websites, because bookmarks can point anywhere, and it asks only when you start that check. · The developer receives none of your data, because there's no server to receive it. ■ Who it's for · Anyone with hundreds of bookmarks who's given up on organizing them by hand · Anyone who wants AI's help but doesn't want it making changes unsupervised · Anyone who cares where their data goes and wants to choose their own model · Anyone who wants no AI in this at all — the cleanup and the stats board never call a model ■ Permissions it needs · Bookmarks: reads the bookmarks within the scope you checked, and reorganizes them according to the plan you confirmed · Storage: saves settings, analysis caches, and the snapshots needed for undo · Side panel: the entire interface lives in the side panel · Favicon: reads the site icons Chrome has already cached, so the HTML export can carry them. Makes no network request · Browsing history (optional): ranking bookmarked sites by visit count, requested only if you turn that on · Host access (optional): reaching the model endpoint you configured, and separately the dead-link check. Never granted at install time Privacy policy: https://gist.github.com/gaotiesuanna/239c067efd9cc7d98f25ed5daa4c3ef7