Replace Pocket, which no longer exists

This is not a page about leaving a service. Pocket left you. Mozilla announced the phase-out in May 2025, stopped the service on 8 July 2025, kept exports available for three months, and deleted accounts on 8 October 2025. Everything below assumes you are rebuilding rather than migrating.

What you are actually giving up#

Pocket was free, it synced to every device without configuration, and when it broke it was not your evening. Those three properties are what you are replacing, and none of the tools here has all of them.

Pocket's text extraction was also better than what you are about to run. It had been tuned against the open web for over a decade, with per-site rules maintained by people who were paid to do it. wallabag uses the same class of extractor with a community site-config set, and it will occasionally give you a page of navigation links where Pocket gave you the article. Karakeep runs a real headless Chrome and gets closer, at three containers and roughly 700 MB idle before you save anything.

Then there is the thing that catches people out: these are three different products and picking wrong is why people abandon the category.

• Read-later archiving (wallabag) means the article body is fetched, extracted and stored. The link can rot and you still have the text. This is what Pocket did.
• Bookmarking (linkding) means the URL, a title, tags and a note. Fast, tiny, searchable, and if the page dies the bookmark is a tombstone.
• Full-page archival with tagging (Karakeep) means the whole rendered page: text, screenshot, PDF, images, indexed in Meilisearch and optionally tagged by a language model.

Saving 4,000 links into a read-later tool and never reading them is the classic failure. If you are honest that you are hoarding references rather than queuing reading, you want the bookmark manager and you want it at one tenth the operational cost.

The last loss is offline. Pocket's mobile apps cached your queue. wallabag has apps that sync, Karakeep has a mobile app, linkding has community clients, and all three need your server reachable. There is no self-hosted equivalent of "it just works on a plane" that does not involve you configuring remote access first.

Get your data out first#

For once, this section is mostly an obituary.

If you exported during the window, you have a CSV with columns for URL, title, time added, tags and status (unread or archive). That is the whole thing: Pocket's export never included the cached article text, only the metadata. Every tool here imports it:

• wallabag 2.6.14 added a Pocket CSV importer specifically for this. Use the asynchronous import path with Redis or RabbitMQ if you have more than a few hundred entries, because a synchronous import runs the content extractor once per entry inside the web request and will time out half way through, leaving you deduplicating.
• Karakeep ships importers for Pocket, Chrome, Linkwarden, Omnivore and Tab Session Manager. Budget hours, not minutes: CRAWLER_NUM_WORKERS defaults to 1 on purpose, so 2,000 bookmarks is 2,000 sequential Chrome renders.
• linkding imports the Netscape HTML bookmark format, so convert the CSV or export from an intermediate tool.

If you did not export, accept it. There is no API, no grace period and no support queue. What you can still salvage: your browser history, which has the URLs if not the tags; any Pocket digest emails in your inbox; and the Firefox reading list if you used that separately. Do not spend a weekend on this. Spend twenty minutes, then start clean.

The lesson worth taking forward is the one that made this page necessary. Whatever you run next, put a scheduled export on it from day one. wallabag exports your whole library as JSON including the extracted article text. linkding has a full_backup management command producing one consistent zip. Karakeep has a backup API you can trigger without stopping anything.

Which one to run#

Pick by which of the three jobs you actually do. If you genuinely read things later, wallabag. If you are building a searchable reference pile, linkding. If you save screenshots, PDFs and images as often as links and want full-text search across page content, Karakeep, and pay for it in RAM.

Two naming traps. Karakeep was called Hoarder until the 2025 rebrand, and the old ghcr.io/hoarder-app/hoarder image stopped getting updates, so anyone following a pre-rebrand tutorial is quietly frozen on an old build. And linkding has two tags: latest is lean, latest-plus bundles Chromium for local HTML snapshots and needs substantially more memory and disk. Choose deliberately.

What the migration actually costs#

• wallabag: about an hour to stand up, one container on port 80 with two volumes. The variable you must get right on first boot is SYMFONY__ENV__DOMAIN_NAME, which is baked into every generated link and password reset email; changing it later means editing the variable and clearing var/cache. Default credentials are wallabag:wallabag and they are in the public docs, so change them in the first session. Ongoing: a Symfony app with a database and a cache directory that is a genuine footgun on restore.
• linkding: twenty minutes. One container, port 9090, one volume. The one thing that will trip you is a 403 CSRF error on login behind a reverse proxy, fixed by forwarding X-Forwarded-Proto and Host properly or setting LD_CSRF_TRUSTED_ORIGINS. Ongoing: a pull and a restart, and one cron line running manage.py full_backup.
• Karakeep: two to four hours, three containers, NEXTAUTH_SECRET and MEILI_MASTER_KEY generated, NEXTAUTH_URL matching the address you actually browse to, and DISABLE_SIGNUPS=true immediately after you create the first account, which is automatically promoted to administrator. Ongoing: a pre-1.0 project that has already merged and removed containers once, so pin KARAKEEP_VERSION and read release notes.

Whichever you pick, you now personally maintain a service whose entire value is that it still has content nobody else is storing for you.

What breaks and how to tell early#

• Mobile. Install the app in the first hour. wallabag has official apps; Karakeep has one; linkding relies on community clients and the web app. If saving from a phone takes more than two taps, you will stop saving.
• The share sheet. Pocket's iOS and Android share targets were the whole product for most people. Test that specific flow on day one.
• Browser extensions. wallabag and linkding both have Firefox and Chrome extensions; Karakeep has one too. Check that yours authenticates, because if you put an auth proxy in front of linkding it will intercept API requests and every extension and mobile client using an API token starts failing.
• Extraction quality. Save ten pages you actually care about, from ten different sites, and read them. This is the honest test and it takes fifteen minutes.
• Losing the admin password. Karakeep has no CLI reset: the documented recovery is stopping the app and running an UPDATE against db.db with sqlite3. Create a second admin account now.

The point of no return#

There is no subscription to cancel here, so the real threshold is when you stop treating this as an experiment:

1. You have restored a backup onto a different machine and your saved articles are there. On linkding, note that copying db.sqlite3 is documented as not transaction safe and can hand you a corrupted database, so use full_backup and open the zip to confirm.
2. You have saved from a phone, from a browser and from your feed reader, and all three worked without you thinking about it.
3. You have a scheduled export in a format that is not this tool's database, because that is the mistake Pocket taught everyone.

Next#

If you are also rebuilding a feed reader, Replace Feedly covers the pairing, since both Miniflux and FreshRSS push saves directly into wallabag and linkding. Read Backups that actually restore first, and size the stack with the Stack planner before you commit to Karakeep's three containers.