Self-hosted bookmarks and read-later in 2026: revisit Linkwarden, Karakeep and Wallabag before you save another link to Pocket

Every technical team collects links. A design document refers to a vendor page. A support engineer saves the forum thread that finally fixed a customer issue. A manager keeps a list of articles for the next team discussion. Most of these links live in browser bookmarks, chat messages and personal read-later apps. When a page disappears or a person leaves, the knowledge goes with it.

For many people the read-later app was Pocket. Mozilla announced in May 2025 that it would phase out Pocket, and the service shut down on 8 July 2025. The original export deadline was 8 October 2025, which Mozilla later extended, finally to 12 November 2025. After that, user data was scheduled for deletion. If your team depended on Pocket, that chapter is closed. The useful question now is where saved links should live so the same thing does not happen again.

In 2026 three open source projects deserve a calm look: Linkwarden, a collaborative bookmark manager focused on preserving pages; Karakeep (previously called Hoarder), a “bookmark everything” app with optional AI tagging; and Wallabag, the long running read-later application that many Pocket users already know.

The revisit habit is the same as always. Learn what each tool really stores. Unlearn the idea that a bookmark is the page itself, because a URL is only a pointer. Re-learn how link rot and archiving work. Practise an export and restore. Then apply it to the team’s real links.

A scenario to keep us honest

This is an invented scenario, not a real company. Pune Ledger Labs is a seven person team building accounting integrations for small businesses. They save a lot of reference material: GST portal notices, API documentation from banks and payment gateways, Stack Overflow answers and blog posts about PostgreSQL tuning. Today each engineer keeps links in their own browser, and two of them used Pocket until it closed. The team lead wants:

  • One shared, searchable place for team reference links, organised by project
  • A saved copy of important pages, because government and vendor pages change or move
  • Phone and browser saving that takes less than five seconds, or nobody will use it

We will come back to this team at the end.

Where these tools came from

Linkwarden

Linkwarden’s GitHub repository was created in April 2022. The project describes itself as a self-hosted, open source collaborative bookmark manager to collect, read, annotate and fully preserve what matters. Its README speaks openly about link rot as the reason for its archiving features. The team also runs a paid hosted offering, Linkwarden Cloud, which funds development. At the time of writing, the latest release on GitHub is v2.16.3, published in September 2026.

Karakeep (formerly Hoarder)

Karakeep began as Hoarder; its repository was created in February 2024. The author explains in the README that he browses on his phone, finds things he wants to read later on a laptop, and wanted a project to keep his web development skills fresh. The project was renamed to Karakeep in 2025, and the README says the name comes from a colloquial Arabic word for miscellaneous clutter. Some older paths still show the old name; for example, the Android app package is still app.hoarder.hoardermobile. The latest GitHub release at the time of writing is v0.34.1, published in October 2026.

Wallabag

Wallabag is the oldest of the three. Its repository dates back to April 2013, when the project started under the name “poche” before being renamed wallabag. It is a PHP application and has been a common self-hosted Pocket alternative for more than a decade. There is also a paid hosted service, wallabag.it, run separately from the open source project. The latest release at the time of writing is 2.6.15, published in late October 2026.

Licences, in plain words

Linkwarden

Linkwarden is licensed under AGPL-3.0. You can self-host it freely. If you modify it and offer it to others over a network, you must make your modified source available to those users. The README notes that SSO integration is available for Enterprise and self-hosted users.

Karakeep

Karakeep is also AGPL-3.0, with the same obligation for modified network services. For internal team use without code changes, this has little practical effect.

Wallabag

Wallabag uses the MIT licence, which is permissive. You can modify and redistribute it with very few conditions. For a small team this is the simplest licence of the three.

How each is built

Linkwarden

Linkwarden is a TypeScript application built on Next.js with a PostgreSQL database. A background worker uses a headless browser to capture each saved page as a screenshot, a PDF and a single HTML file, and to produce a readable version. Newer versions use Meilisearch for search. The documentation offers Docker Compose as the main installation path, and the archives live on disk or in S3 compatible storage, depending on configuration.

Karakeep

Karakeep is a TypeScript monorepo. The README lists Next.js for the web app, Drizzle for the database layer and migrations, Better Auth for authentication, tRPC between client and server, Puppeteer for crawling and Meilisearch for full text search. The standard Docker setup runs the main app, a headless Chrome container for crawling and a Meilisearch container. Full page archives are made with monolith, and videos can optionally be saved with yt-dlp. Optional AI features call either an OpenAI compatible API or a local model through Ollama.

Wallabag

Wallabag is a PHP application built on the Symfony framework. It works with SQLite, MySQL/MariaDB or PostgreSQL, and can use Redis or RabbitMQ for asynchronous imports. It extracts the readable article text from a page using site configuration rules, and stores that text rather than a full visual snapshot. An official Docker image is maintained, and classic installation on a PHP host is also documented.

Features that matter for a small team

Linkwarden

  • Automatic screenshot, PDF and single file HTML capture of each saved page
  • Reader view with highlights and annotations
  • Optional submission to the Wayback Machine at archive.org
  • Optional local AI tagging based on page content
  • Collections, sub collections and tags, with per member permissions for collaboration
  • Official iOS and Android apps, a browser extension, and browser sync through Floccus
  • Full text search, RSS subscriptions, API keys and user administration

Karakeep

  • Saves links, simple notes, images and PDFs in one place
  • Automatic titles, descriptions and preview images
  • Optional LLM based tagging and summarisation, with local model support through Ollama
  • Full text and semantic search
  • A rule engine for automatic organisation, and OCR for text inside images
  • Chrome, Firefox and Safari extensions, plus iOS and Android apps with offline reading
  • Importers for Chrome, Pocket, Linkwarden, Omnivore and Tab Session Manager
  • Shared lists, highlights, RSS auto saving, REST API, CLI and SSO

Wallabag

  • Clean reading view of saved articles, with tags, annotations and starring
  • Import from Pocket exports, browser bookmarks and several other services
  • Export to formats such as EPUB, PDF and CSV for offline reading
  • Browser extensions and mobile apps for Android and iOS
  • RSS feeds of your saved items and a REST API for third party clients
  • Mature, quiet and predictable; it focuses on reading, not on collaboration

AI tagging, carefully

AI tagging sounds attractive, but treat it as a helper, not a filing clerk. It is optional in both Linkwarden and Karakeep, and Wallabag does not depend on it. A few practical points:

  • If you send page content to a hosted AI API, you are sharing that content with a third party. For internal or client related links, that may not be acceptable.
  • A local model through Ollama keeps data on your network but needs CPU, memory and preferably a GPU. On a small server, tagging can be slow.
  • Automatic tags drift. Review them for a few weeks and keep a short list of agreed team tags.

No benchmark was run for this article, and we do not quote speed numbers. Test on your own hardware with your own links.

Archiving: a bookmark is not the page

This is the most important difference between the three tools.

  • Linkwarden keeps several preserved formats per link: screenshot, PDF and single HTML file. This is closest to a personal web archive.
  • Karakeep stores fetched content for search and can keep a full page archive with monolith, plus videos if you enable it.
  • Wallabag stores the extracted article text and images. This is excellent for reading, but it is not a visual record of how the page looked.

Archives use disk. Pages with heavy images or video grow quickly. Plan storage and decide which collections really need full preservation. Also remember that some sites block headless browsers or show cookie walls, so a capture can silently save the wrong thing. Spot check archives of important pages.

Operations

Installation

All three can run with Docker Compose on a small Linux server. Linkwarden and Karakeep need more moving parts (database or search engine, plus a headless browser). Wallabag can run with just PHP and a database, which suits low resource hosts.

Upgrades

  • Pin image tags to a specific version rather than latest.
  • Read release notes before upgrading; database migrations usually run automatically at startup, which makes rollback harder.
  • Take a backup immediately before every upgrade.
  • Karakeep is still on 0.x versions, so expect faster change and occasional breaking notes.

Monitoring

Watch disk usage, failed crawl or archive jobs, and container restarts. A crawler that fails quietly means your “saved” pages are not really saved.

Backups

  • Back up the database and the archive or data directory together, at the same point in time.
  • For Karakeep, the data volume holds the database and assets; the Meilisearch index can usually be rebuilt, but confirm this with the current documentation.
  • For Linkwarden, dump PostgreSQL and copy the archive storage.
  • For Wallabag, dump the database; article content lives there.
  • Keep one copy off site, and use each tool’s own export as an extra, portable copy.
  • Practise a restore on another machine at least once.

Security

Exposure

Put the app behind a reverse proxy with HTTPS. If only the team needs it, keep it on a VPN or private network rather than the open internet. Disable open sign up after creating accounts.

Crawler risk

A bookmark tool fetches any URL a user gives it. That makes it a possible path to internal addresses (server side request forgery). Karakeep has a Security Considerations page in its docs; read it, and for all three, restrict what the crawler container can reach on your internal network.

Accounts and secrets

Use separate accounts per person, SSO where available, and remove leavers promptly. Keep API keys and AI provider keys in environment files with restricted permissions, never in a shared chat.

The Indian context: the DPDP Act

The Digital Personal Data Protection Act, 2023 applies to digital personal data. Most bookmarks are public web pages, so the Act may not be a big concern for a typical reference library. It becomes relevant when saved items contain personal data, for example screenshots of customer tickets, uploaded PDFs with names and phone numbers, or notes about individuals. In those cases:

  • Keep such material in restricted collections or lists, not shared widely.
  • Avoid sending it to an external AI API without a clear basis and agreement.
  • Have a retention rule and actually delete items when they are no longer needed.
  • Hosting on your own server in India keeps location under your control, but security duties still apply.

This is general awareness, not legal advice. Check the Act and the rules notified under it with your advisor.

When not to adopt each

Do not choose Linkwarden if

  • You only want a simple personal reading list and do not care about preserved copies
  • Your server is very small and cannot run a headless browser comfortably

Do not choose Karakeep if

  • You need a slow, stable release rhythm and dislike 0.x software
  • You want AI tagging but cannot use either an external API or a local model

Do not choose Wallabag if

  • You need a team space with shared collections and permissions
  • You need a faithful visual snapshot of pages, not just article text

Do not self-host at all if

  • Nobody will own upgrades and backups; a hosted plan is safer than an abandoned server

Back to Pune Ledger Labs

For this invented team, a sensible path is to pilot Linkwarden, because shared collections with member permissions and preserved PDF and screenshot copies match the need to keep GST notices and vendor documentation that may change. Create one collection per project, install the browser extension and mobile app for everyone, and import the two old Pocket exports if they still have them. Leave AI tagging off for the first month. If engineers later want quick capture of notes, images and screenshots with smart search, a short Karakeep trial is worth doing. Anyone who mainly wants a quiet personal reading queue can run Wallabag for themselves. No hands-on drill was run for this article; the plan is based on the projects’ own documentation.

Revisit checklist

  • Decide whether you need a reading queue, a team library or a web archive
  • Check the licence and which features are hosted only or enterprise only
  • Pin versions and read release notes before each upgrade
  • Back up database and archives together, and keep one copy off site
  • Restrict the crawler’s network access and keep the app behind HTTPS
  • Decide where AI tagging runs, or keep it off
  • Map any personal data to DPDP retention and deletion
  • Practise an export and restore before you trust it

Sources

comments powered by Disqus

Releted Posts

Self-hosted photo libraries in 2026: revisit Immich, PhotoPrism and Ente before you upload another album to Google Photos

Photos look like a personal topic, but in many small Indian businesses they are working data. A studio keeps client shoots.

Read more

Self-hosted document management in 2026: revisit Paperless-ngx, Mayan EDMS and Docspell before you scan another invoice into a shared drive

Every small business has a paper problem that slowly becomes a file problem. Purchase invoices, GST returns, rent agreements, salary slips, vendor KYC, insurance policies, bank statements.

Read more

Self-hosted forms and surveys in 2026: revisit Formbricks, LimeSurvey and OpnForm before you collect another response through Google Forms or Typeform

Almost every organisation collects data through forms. Admission enquiries, job applications, event registrations, customer feedback, employee pulse surveys, vendor onboarding. Most of this goes into Google Forms or Typeform because they are quick to start.

Read more