Pocket shut down. Omnivore shut down. Del.icio.us, Google Reader, and a dozen others shut down before them. If you have been saving links for more than a decade, you have watched your archive evaporate at least once.
That is the whole argument for self-hosting. Not ideology, not tinkering for its own sake. Just the observation that a bookmark manager you do not control is a bookmark manager on a countdown you cannot see.
The good news is that this category got genuinely good in the last couple of years. The tools below are not rough hobby projects. Several are more capable than the commercial products they replace, and one of them does AI tagging locally, which no hosted competitor offers at any price.
I have deployed each of these. Here is what they are actually like to run, and which one fits which job.
- Linkwarden wins overall, with collections, tags, and automatic full-page archiving to PDF and screenshot, in a polished and actively maintained package.
- Karakeep wins for automatic organization, including AI tagging through a local Ollama model so nothing leaves your box.
- Wallabag wins as the read-it-later replacement, and it is the most mature project here.
- Shiori wins on simplicity, one Go binary, no database setup, running in five minutes.
- ArchiveBox wins if you want a true archive rather than a bookmark list, saving multiple full copies of every page.
- The shared blind spot: none of these can capture your X, Reddit, or LinkedIn saves. They store URLs you give them. ContextBolt is the one that does. It is not self-hosted, though Basic is free, needs no account, and keeps everything on your own machine, so it runs beside whichever server you pick.
What you actually gain by self-hosting bookmarks
Three things, and it is worth being precise because the marketing around data ownership tends to overclaim.
Your archive outlives the vendor: this is the real reason. When Pocket announced its shutdown, users had weeks to export or lose a decade of saves. A self-hosted instance has no shutdown date.
Your reading history stays private: every link you save to a hosted service is a data point about what you are thinking about. On your own server, that list is nobody’s business.
Full-page archiving without a subscription: most hosted tools put permanent page copies behind a paid tier, if they offer them at all. Several tools here archive every page by default, for free, because storage is just your disk.
And one thing you do not gain, despite what the category’s enthusiasm suggests, is convenience. You are taking on updates, backups, and 3am breakage. That trade is worth it for some people and genuinely not worth it for others. Be honest about which you are before you spend a Saturday on this.
How I judged these self-hosted bookmark managers
- Archiving: does it keep a real copy of the page, or just the URL?
- Search: full text of archived pages, or titles and tags only?
- Organization: manual tags, or something that files things for you?
- Setup cost: how long from
docker compose upto a working instance? - Maintenance burden: how often does it break, and how active is the project?
- Clients: browser extension and mobile apps, or web only?
I excluded projects with no commit activity in the past year, anything without a documented Docker path, and pure link-shortener or start-page tools that are not really bookmark managers.
Self-hosted bookmark managers compared
| Tool | Best for | Stack | Full-page archive | Auto tagging | Setup effort |
|---|---|---|---|---|---|
| Linkwarden | All-round archival | Next.js + Postgres | PDF, screenshot, HTML | AI tags, optional | Low |
| Karakeep | Hands-off organization | Node + Postgres | Yes | Yes, incl. local Ollama | Medium |
| Wallabag | Reading articles later | PHP + Symfony | Article text | No | Medium |
| Shiori | Simplicity | Single Go binary | Article + archive | No | Very low |
| LinkAce | Link rot monitoring | PHP + Laravel | Via Wayback | No | Medium |
| Readeck | Clean reading, low resources | Single Go binary | Article text | No | Very low |
| ArchiveBox | True web archiving | Python + Django | Multiple formats | No | High |
The 7 best self-hosted bookmark managers
1. Linkwarden
Best all-round self-hosted pickLinkwarden is the one I would point most people at. It is the closest thing in this category to a commercial product, and it does the single most important job properly: every link you save gets archived as a PDF, a screenshot, and readable HTML, automatically, at save time.
That matters more than any feature list. A bookmark to a dead page is worthless, and roughly a tenth of the links you save will rot within a few years. Linkwarden solves that by default rather than as a paid extra.
Organization is collections plus tags, with nested collections and the ability to share a collection publicly or with team members. The interface is clean, the browser extension works, and there is a mobile app. Deployment is a Docker Compose file with Postgres, and it is genuinely uneventful.
- Automatic PDF, screenshot, and HTML archiving of every link
- Nested collections plus tags
- Team sharing and public collections
- Browser extension and mobile app
- Active development and a real community
- Optional AI tagging
- Archiving eats disk fast, budget for it
- Needs Postgres, so not a single-binary deploy
- Search is keyword based, not semantic
Best for: Anyone who wants one self-hosted tool to save, archive, and organize links and does not want to think about it again.
Pricing: Free and open source, and the hosting bill is yours. Cloud is $4 per user monthly, or $36 per user annually, if you would rather not host it.
Verdict: Start here. If it does not fit, the reason will tell you which of the others to try.
2. Karakeep (formerly Hoarder)
Best for automatic organizationKarakeep is the project that was called Hoarder until it was renamed, so plenty of older guides and Reddit threads still use the old name. It is the most interesting tool in this category because of one feature. It tags your bookmarks for you using an AI model.
Crucially, that model can be a local one. Point it at Ollama running on the same machine and your bookmarks get automatically categorized without a single request leaving your network. No hosted bookmark manager offers that, because the whole point of a hosted product is that your data goes to them.
Beyond tagging you get full-text search across archived content, a browser extension, mobile apps, and support for saving notes and images alongside links. It also handles bulk imports well, which matters when you are migrating years of saves.
The trade-off is complexity. Running AI tagging means either an API key or a local model with the RAM to match, and there are more moving parts to go wrong than in Linkwarden.
- AI tagging that can run fully locally via Ollama
- Full-text search across archived pages
- Saves links, notes, and images
- Good browser extension and mobile apps
- Solid bulk import
- More moving parts than the alternatives
- Local AI tagging needs real RAM
- The Hoarder rename makes older documentation confusing
Best for: People who never keep up with manual tagging and want the filing done for them, without sending their reading list to a third party.
Pricing: Free and open source.
Verdict: The most future-facing tool here. Choose it over Linkwarden if automatic organization matters more to you than polish.
3. Wallabag
Best read-it-later replacementWallabag has been around since 2013 and it shows, in the good way. It is the most mature project in this list, and it was built from the start as a self-hosted answer to Pocket.
The focus is reading, not filing. Wallabag extracts the article text from a page, strips the clutter, and gives you a clean reading view with tags and an archive state. It exports to ePub, MOBI, and PDF, so you can send a week of saved articles to an e-reader.
It is the right pick if what you lost when Pocket died was a reading queue rather than a link library. If you want to manage 5,000 reference links, Linkwarden is the better shape.
- Over a decade of development, very stable
- Excellent article extraction and reading view
- ePub, MOBI, and PDF export for e-readers
- Wide client support including third-party mobile apps
- Straightforward Pocket import
- Interface looks its age
- Built for articles, awkward with other link types
- PHP and Symfony stack is heavier to maintain
- No automatic tagging
Best for: Ex-Pocket users who want their reading queue back, on their own server, with e-reader export.
Pricing: Free and open source, plus a low-cost hosted option from the project.
Verdict: The safest long-term bet in the whole category, purely on track record.
4. Shiori
Best for simplicityShiori is a single Go binary. Download it, run it, and you have a working bookmark manager with no database to configure and no container orchestration to reason about. For a lot of people that is the entire pitch and it is enough.
It saves the page content for offline reading, gives you tags and search, and ships a CLI alongside the web interface, which is genuinely handy for scripting bulk imports. It is deliberately minimal and does not pretend otherwise.
Development is slower than Linkwarden’s and the feature set is smaller. That is the trade you are making for something that will still start up cleanly in three years with no migration drama.
- Single binary, no database required
- Running in about five minutes
- Saves page content for offline reading
- Useful CLI alongside the web UI
- Very low resource use
- Minimal feature set
- Slower development pace
- No official mobile app
Best for: Anyone who wants this solved in an afternoon with the least possible ongoing maintenance.
Pricing: Free and open source.
Verdict: The lowest-effort entry point. A perfectly good permanent answer for light users.
5. LinkAce
Best for link rot monitoringLinkAce has one feature the others mostly lack. It checks your saved links on a schedule and tells you which ones have died. For anyone maintaining a reference library that other people rely on, that is the difference between a resource and a graveyard.
It also integrates with the Internet Archive, so a dead link can be swapped for a Wayback snapshot rather than deleted. Organization is tags plus lists, with a clean interface and a proper REST API.
It leans on Wayback rather than archiving pages itself, which is lighter on disk but means you depend on somebody else having crawled the page.
- Scheduled dead-link checking
- Internet Archive integration
- Clean interface, tags and lists
- Well-documented REST API
- Relies on Wayback instead of archiving itself
- No automatic tagging
- Laravel stack needs more care to update
Best for: Curators keeping a reference library healthy over years.
Pricing: Free and open source.
Verdict: Niche but excellent at its niche. Pair it with something else if you want real archiving.
6. Readeck
Best for clean reading on low resourcesReadeck is a newer Go project that occupies similar ground to Wallabag with a much more modern interface and a far lighter footprint. Single binary, tiny memory use, happy on a Raspberry Pi.
It saves article text for offline reading, supports labels and collections, and handles ebook export. The reading experience is genuinely nice, which sounds trivial until you spend an hour a day in it.
Being newer, it has a smaller community and fewer third-party clients than Wallabag. If long-term stability is your main worry, Wallabag has the longer record.
- Modern, pleasant reading interface
- Single binary with very low resource use
- Labels, collections, and ebook export
- Runs comfortably on a Raspberry Pi
- Younger project, smaller community
- Fewer third-party mobile clients
- Articles only, weak with other link types
Best for: Home-server users who want Wallabag’s job done with a nicer interface and less overhead.
Pricing: Free and open source.
Verdict: The best-looking reader here. Choose Wallabag instead only if project maturity is your priority.
7. ArchiveBox
Best for true web archivingArchiveBox is not really a bookmark manager and including it needs that caveat. It is a self-hosted web archiver that happens to be fed by bookmarks.
Give it a URL and it saves everything, raw HTML, a screenshot, a PDF, a WARC file, extracted text, even the video if there is one. The result is a genuine archive that will still render in twenty years, not a link with a preview image.
The cost is complexity and disk. It pulls in Chromium, yt-dlp, and several other dependencies, and a few thousand pages will consume serious storage. Browsing and organization are weak because that is not the point of the project.
- Most thorough archiving of anything here
- Multiple independent formats per page, including WARC
- Downloads embedded media
- Built for genuine long-term preservation
- Heaviest setup on the list
- Consumes disk aggressively
- Weak browsing and organization
- Overkill for everyday bookmarking
Best for: Researchers and archivists who need evidence-grade copies of pages.
Pricing: Free and open source.
Verdict: Run it alongside a real bookmark manager, not instead of one.
The gap every self-hosted bookmark manager shares
Worth stating plainly, because it is the thing people discover a month after setting one of these up.
None of them can capture your platform-native saves. Not one. The bookmarks you tap on X, the posts you save on Reddit, the items you save on LinkedIn all stay locked inside those accounts. Self-hosted tools store URLs you deliberately hand them, through an extension or a paste box.
The reason is not laziness on the developers’ part. LinkedIn exposes no API for saved items at all, X put its bookmark access behind paid API tiers, and Reddit’s API terms tightened in 2023. Reaching platform saves means reading the rendered page in a logged-in browser, which is a fundamentally different architecture from a server that fetches URLs. It is also why every way to export LinkedIn saved posts runs through the browser instead of an API.
So if most of what you save is a tap inside a social app, a self-hosted manager will not capture it, and your carefully owned archive ends up holding only the links you remembered to file twice.
To be straight about our position here, we build ContextBolt, and it is not self-hosted, so it does not belong in the ranking above. It is the answer to the gap, though. A Chrome extension saves web pages with their text and catches your X, Reddit and LinkedIn saves automatically, AI-tags all of it, and searches it on your device. Basic is free, needs no account, and keeps everything on your own machine, so the privacy half of the self-hosting argument holds. Pro adds cloud sync and an MCP endpoint so Claude or Cursor can search the library by meaning. Karakeep ships its own MCP server, so agent access is not the difference; what sits behind the endpoint is. If data sovereignty on your own server is the priority, Linkwarden or Karakeep is the right server, and nothing here argues against running one. If the saves you keep losing are inside social apps, no tool on this page reaches them, and ContextBolt does, local by default and free, so it costs nothing to run alongside one of these. If you would rather have the AI do the filing without running a server at all, the best AI bookmark managers roundup covers the hosted options and ranks Karakeep beside them. The local-first middle ground is worth reading before you decide.
Which self-hosted bookmark manager should you run
- You want one tool that just works: Linkwarden. Best balance of archiving, organization, and polish.
- You will never tag anything manually: Karakeep, with a local Ollama model so the AI stays on your hardware.
- You miss Pocket: Wallabag. Same job, your server, e-reader export included.
- You want this done in an afternoon: Shiori. One binary, no database.
- You maintain a reference library others use: LinkAce, for the dead-link checking.
- You want a nice reader on a Pi: Readeck.
- You need evidence-grade page copies: ArchiveBox, alongside one of the others.
One last piece of advice from having run several of these. Set up your backups on day one, before you import ten years of bookmarks. Self-hosting moves the single point of failure from a company’s business model to your own disk, and that is only an upgrade if the disk is backed up.
And whichever server you pick, install ContextBolt Basic beside it. It is free, needs no account, stays on your machine, and catches the X, Reddit and LinkedIn saves that none of the seven can reach, so the archive you just built stops missing the half you save inside social apps.