helgesverre/bookmarking-book

If you've ever bookmarked hundreds of pages in Chrome and then never looked at them again, Bookmark Book is a tool that fixes that problem. It takes your entire bookmark collection and turns it into a readable, organized markdown document, like a personalized book built from everything you've ever meant to read. The tool runs on your computer through the command line. You give it access to your Chrome bookmarks file, and it walks through a multi-step pipeline. First it visits each bookmarked URL and pulls the actual text from the page. If a link is dead, it tries the Wayback Machine to recover the content. Then it sends everything to Claude (Anthropic's AI) in two passes: first to tag each bookmark by topic, type, and quality, and then to normalize those tags into a consistent taxonomy. After that, it groups bookmarks into 15 to 25 thematic chapters, writes narrative introductions and summaries for each, and assembles the whole thing into a single markdown file with a table of contents. The output isn't just a flat list. It includes an "Interest Graph" that maps how your topics connect, a "Time Capsule" showing how your interests changed over the years, a "Graveyard" section for dead links that were rescued via archived copies, and a "Rabbit Hole Index" that suggests what to read next based on how much time you want to invest. Each chapter groups bookmarks by subtopic and includes two to three paragraph summaries of individual pages. This is for people who have accumulated a large bookmark collection and want to actually make sense of it, researchers, developers, founders, or anyone who treats bookmarks as a reading backlog. If you've ever wanted a curated digest of your own saved links rather than scrolling through Chrome's nested folders, this produces exactly that. It costs real money to run, around $10 to $18 for roughly 1,700 bookmarks, because the classification and generation steps call Claude's API. All processing happens locally and through direct calls to Anthropic, so your data doesn't go to any third-party servers beyond that. The crawl results are cached, so if the process gets interrupted you can resume without re-fetching everything.