Open Source Readwise Reader Alternatives

Karakeep, previously called Hoarder, is a bookmark-everything app for people who save far more than they ever revisit. Links, quick notes, images and PDFs all land in the same place, and the pile stays usable because tagging and summarization happen automatically rather than waiting on your discipline.

The web app is built on Next.js with Drizzle for the database, tRPC between client and server, NextAuth for sign-in and Meilisearch for full content search. Puppeteer crawls saved links to pull titles, descriptions and images, and the tagging model can be a hosted provider or a local one served through Ollama.

Saving is only the first half of what it does.

• Full page archival: pages stored with monolith and videos with yt-dlp, so link rot cannot erase what you kept.
• Search: full text and semantic search over everything stored, including text pulled out of images by OCR.
• Capture everywhere: Chrome, Firefox and Safari extensions, iOS and Android apps, RSS hoarding and importers from Pocket, Linkwarden, Omnivore and Chrome.
• Rules engine: rule-based automation that files bookmarks without manual sorting.
• Collaboration: lists can be shared and worked on with other people.

A REST API and a command line client make the collection scriptable, and official agent skills let LLM agents work with it directly. Offline reading, highlights, bulk actions, SSO, dark mode and multi-language support are all present. Self-hosting comes first, under an AGPL-3.0 license, with a managed cloud for anyone who would rather not run it.

Linkwarden is a self-hosted bookmark manager for collecting, reading, annotating and preserving the pages that matter. Links rot as sites disappear, so every bookmark you save also becomes an archived copy that you keep.

Each link is captured as a screenshot, a PDF and a single HTML file, and can optionally be pushed to the Wayback Machine for a second snapshot. A reader view strips the page down for reading, with highlighting and annotation on top of it.

Organization and sharing run through collections rather than a flat list of saved pages.

• Collections and tags: links sit in collections and sub-collections with names, descriptions and multiple tags.
• Member permissions: several people gather links into one collection, and each member has their own permission set.
• Full text search: search, filter and sort across everything you have saved.
• Local AI tagging: an optional model tags links from their content instead of you doing it by hand.
• Apps and extensions: native iOS and Android apps, a browser extension, API keys and RSS feed subscriptions.

Self-hosting is documented for anyone who wants to run their own instance, and a hosted cloud plan exists for people who would rather not. Interface translations, SSO for self-hosted and enterprise users, bulk actions, user administration and image and PDF uploads fill out the rest.

wallabag is a web application for saving web pages to read later. Click, save, and come back when you have time. It extracts the article content so pop-ups and the rest of the page furniture do not follow you into the reading view.

Extraction quality is what decides whether a tool like this is worth keeping, and wallabag leans on three well-known libraries for it, Graby, php-readability and the ftr-site-config rule set. Site-specific configuration is what makes awkward pages come out clean.

Saved articles are reachable from more than the web interface.

• Android app: an official client maintained in its own repository.
• iOS app: likewise official and developed alongside the main project.
• Browser extension: wallabagger, for saving whatever page you are currently reading.
• GNOME app: read-it-later for Linux desktops, maintained outside the project.

Installation is a git clone followed by a make install, then a virtual host pointed at the result, with other methods documented for anyone who would rather not build from source. If running a server is not appealing, the maintainers operate a hosted instance you can sign up for instead.

Maintained since 2013 and distributed under the MIT license, the project takes translations through Weblate and welcomes anyone who wants to add a language. One request from the authors: the name is always lowercase.

Most applications handle one media type well, notes or web pages or PDFs, and leave you shuttling between windows. Deta Surf is an AI notebook that pulls your files and the web into the same stream of thought, cutting out the searching, tab juggling and copy-pasting that research usually involves.

Data stays on your machine in open formats, held by a local storage engine the project calls SFFS. Surf is built in Svelte, TypeScript and Rust, runs on macOS, Windows and Linux, and works with whatever language model you point it at, a cloud model with your own key or a local one.

Several features build on that private library.

• Notebooks: local files, websites, YouTube videos and tweets are stored and organized, much of it usable offline.
• Smart notes: any tab or library item can be mentioned inline, with citations deep-linked to a page section, a video timestamp or a PDF page.
• Web search: research results come back into the note you are already writing.
• Surflets: interactive applets are generated from a prompt to visualize data or explore a concept.
• Tabs and split view: notes, files and web pages open side by side, with a sidebar for navigation.

Installers are documented in the repository, and building from source is covered by the contributing guide. The source is under Apache 2.0 with one exception noted in the license section. Surf suits people who want their reading, their notes and their AI in one window, without any of it leaving the machine unless they choose.

Bookmark It is an open source bookmark manager. Saved links are arranged in a timeline instead of a folder tree, so looking back over what you kept works like scrolling through your own history rather than digging through nested directories.

It was built from scratch on Next.js 14, with shadcn components and Tailwind for the interface. Supabase supplies the Postgres database and Google sign in, hosting is on Vercel, and LemonSqueezy handles payments for the paid tier.

The product page describes what happens to a link once it is saved.

• One click saving: browser extensions add a bookmark and browse the collection without leaving the tab.
• Tags: bookmarks are grouped by tag, with the tag allowance depending on the plan.
• Favorites: links worth returning to are marked and kept in their own list.
• Public sharing: a tag can be published as a shared collection.
• Import and export: browser bookmarks import in bulk, and the paid plan exports to CSV or HTML.
• Theme customization: the appearance of the interface is adjustable.

A free tier covers a monthly bookmark allowance with one unlimited import, and a yearly paid plan raises every limit and adds email support. The code is AGPL 3.0 and the author takes issues and pull requests for bugs, features and quality of life work, so the whole thing can be read, reviewed or extended by anyone who wants to.

Capd keeps what you come across on a Mac and makes it findable later. There is no account, no subscription and no telemetry, and the library stays on the machine inside a local SQLite database and an assets folder.

Captures start from a keyboard shortcut, a drag onto the menu bar or the notch, the macOS share sheet, or the bundled command line tool. Browser captures keep the page and, where available, its readable article text, and re-capturing an item updates the existing entry instead of adding a duplicate.

• Search beyond bookmarks: titles, article text, selections, notes and text recognized inside images are all indexed.
• Filtered queries: narrow results by site, tag and date from the same field you type keywords into.
• Ask Cap: Apple Intelligence assembles an on-device answer whose claims link back to the captures behind them.
• Automatic tagging: captures are grouped into a small vocabulary on device, without their contents leaving the Mac.
• CLI and Raycast: the capd command adds, searches and exports, and a Raycast extension reads the same library.

A read-only MCP server offers search and cited answers to compatible AI assistants, and the CLI prints stable JSON for scripts. Network behavior is limited and written down in the privacy guide, and several of those features can be switched off.

Released under the MIT license, the app needs macOS 26 or later and runs on Apple silicon and Intel machines. Installation comes from a notarized disk image or through Homebrew.