Herbarium stores AI-generated HTML pages in a searchable review vault

Summary

Herbarium is a desktop application for keeping HTML pages generated by AI tools. Users can import pages by drag and drop, file picker, or paste, then search their titles, tags, notes, and page text; organize them with folders, tags, and notes; edit the HTML; and read it with interactivity preserved. Pages are stored as ordinary .html files with adjacent .json metadata files in a vault folder owned by the user, while folders remain real directories that can be backed up, synced, or edited externally. The app includes user-selected review schedules, reminders, keyboard-driven review sessions, trash restoration, undo, and a command palette. Because generated HTML can execute code, each page runs in a sandbox that cannot access local files, the app, or other pages, and links require confirmation before opening in a browser. Network access is controlled per page; when enabled, pages may load scripts and fonts from well-known CDNs and images from any website, allowing the page author to know when it is opened. Herbarium also works as an MCP server, allowing AI agents to save, search, organize, and schedule pages. It supports Linux, macOS, and Windows, is built with Tauri 2, Rust, SQLite, and Svelte 5, and is released under the MIT license.