Karakeep vs linkding

Karakeep and linkding are both common picks for replacing Pocket, and both will hold your bookmarks on your own server. They are not the same kind of tool, though. Karakeep calls itself a "bookmark-everything app with a touch of AI": it saves links, notes, images and PDFs, crawls each page with a headless Chrome, indexes the content in Meilisearch, and can tag and summarize with a language model. linkding describes itself as "minimal, fast, and easy to set up using Docker": one Django container that keeps a clean, searchable, tagged list of links. The choice comes down to whether you want an archive of what you saved or a list of where it lives.

An archive of content vs. a list of links

Karakeep (previously named Hoarder) treats each bookmark as content to keep. Its feature list covers links, simple notes, images and PDFs; automatic titles, descriptions and images; lists you can share with other users; highlights; OCR for text in images; auto-saving from RSS feeds; and a rule engine. Full-page archival with monolith and video downloads with yt-dlp are both there as switches (CRAWLER_FULL_PAGE_ARCHIVE, CRAWLER_VIDEO_DOWNLOAD), and both are off by default. It has Chrome, Firefox and Safari extensions and official iOS and Android apps, including offline reading on mobile.

linkding is built around the bookmark itself: tags, bulk editing, Markdown notes, a read-it-later flag, sharing with other users or guests, and automatic titles, descriptions and icons. It imports and exports the Netscape HTML format every browser understands, installs as a Progressive Web App, and has Firefox and Chrome extensions plus a bookmarklet. Mobile apps exist, but upstream lists them under community projects rather than shipping its own.

Search and tagging: AI vs. rules

This is the biggest practical difference. Karakeep indexes the page content it crawls, so its README promises "full text & semantic search of all the content stored." Full-text search runs on Meilisearch; semantic search additionally needs an embedding model configured. Automatic tagging needs either OPENAI_API_KEY (which also works with OpenAI-compatible providers through OPENAI_BASE_URL) or OLLAMA_BASE_URL for local models. Without one, the docs say, "automatic tagging will be skipped." Summarization is a separate switch that is off by default. Tag quality, upstream warns, "will depend on the quality of the model you choose."

linkding's search matches words and phrases in the title, description, notes and URL, plus tags, with and, or, not and parentheses. It does not search the text of the saved page. Its auto-tagging is rule-based: you map URL patterns to tags (youtube.com video), and the docs are explicit that matching "only works based on the URL, not on the content of the website." No model, no API key and no per-bookmark inference cost. The trade-off is that you do the thinking.

Footprint and setup

Karakeep's recommended install is three containers from its own Compose file: the web app, a karakeep-chrome browser for crawling, and Meilisearch. You fetch the file and write a .env with NEXTAUTH_SECRET, MEILI_MASTER_KEY and NEXTAUTH_URL, then run docker compose up -d. A minimal single-container install exists, but the docs warn you give up a lot. Without Meilisearch, search is "completely disabled". Without Chrome, you lose screenshots and JavaScript-heavy pages don't crawl correctly. Karakeep publishes no RAM requirement. Our 2 GB floor is an estimate for the full three-container stack, since a headless browser and a search engine run next to the Node app. Add more if you run Ollama on the same box.

linkding is a single docker run with one volume, SQLite by default (PostgreSQL is an LD_DB_ENGINE option), and an image that runs on ARM, Raspberry Pi included. It publishes no RAM figure either; the 256 MB in our catalog is also our estimate. The exception is the latest-plus image, which bundles Chromium to take local HTML snapshots. linkding's archiving docs say that needs "at least 1 GB of RAM." Our linkding page has the one-line install.

We rate linkding 1 out of 5 to deploy and Karakeep 2 out of 5. Neither is hard. Karakeep just has more services, secrets and optional providers to wire up.

Running it safely

Two defaults matter on a public server. Karakeep makes the first account to sign up the administrator, and new signups stay open until you set DISABLE_SIGNUPS=true, so register immediately and then close the door. Its security page also warns that the crawler fetches whatever URL a user submits from your own network. An untrusted user could use it to reach internal endpoints or expose your origin IP. Karakeep has basic SSRF protections, but upstream's advice is to limit access to trusted users or route the crawler through a proxy outside your network.

linkding has the opposite problem: no account exists after first start. You create one with python manage.py createsuperuser inside the container, or with LD_SUPERUSER_NAME and LD_SUPERUSER_PASSWORD. Both support OIDC single sign-on. linkding can also sit behind an authentication proxy.

Licensing

Karakeep is AGPL-3.0; linkding is MIT. For running either one for yourself or your team, neither license asks anything of you. The AGPL only matters if you modify Karakeep and offer it to others over a network.

Which to choose

Choose Karakeep if you save more than links, such as articles, screenshots, PDFs and notes, and want to find them later by what they say rather than by what you tagged them. Also pick it if you want AI tagging, native mobile apps or RSS auto-saving, and can give it a small VPS with room for Chrome and Meilisearch. Choose linkding if you want a fast, dependable bookmark list you can run on almost anything, with tags you control and no model in the loop. Neither wins outright. The question is whether you want an archive or an index.

Choosing between linkding and Karakeep

Put the other way round, start with linkding if your old Pocket or browser bookmarks were mostly "links I'll come back to". It is the smaller install and the one with less to maintain. Move to Karakeep when searching the saved content itself, or letting a model do the tagging, is something you would actually use.

Not the only two options

If what you miss about Pocket is reading saved articles offline in a clean view, Wallabag is the dedicated read-it-later app in this catalog. See all the self-hosted Pocket alternatives for the full list.