Deepmark vs Karakeep, compared honestly

Karakeep facts from karakeep.app, docs.karakeep.app and the v0.30–v0.33 release notes, checked 8 August 2026. Karakeep moves fast; tell me if a line is stale and I'll fix it.

• Self-hosted, free, and unlimited. Open source, unlimited bookmarks and storage on your own hardware, with roughly 28k GitHub stars of community behind it. Deepmark has no self-host story at all.
• Nothing has to leave your network. AI tagging, summarization and OCR can run against a local Ollama model. Deepmark's pipeline runs on hosted infrastructure and named subprocessors; the privacy page lists exactly who.
• Archival is first-class: full-page copies, screenshots, PDF snapshots, and video downloads via yt-dlp. You keep the actual bytes against link rot. Deepmark keeps a screenshot and the extracted text, not a full archive.
• A bigger agent surface: Karakeep's MCP server exposes 29 tools, plus a REST API, CLI, webhooks and a rule engine. Deepmark's MCP server has three tools (search, latest, get) that cover retrieval and nothing else.
• Mobile apps, RSS auto-hoarding, and a cheap managed option: Cloud Pro is $4/month for 50,000 bookmarks if you want Karakeep without the server.

• Videos become searchable text. Karakeep can download a video for archival, but nothing transcribes it: the words said in a reel are not searchable. Deepmark transcribes every video and reel in some 99 languages, describes sampled frames, and reads on-screen text out of them. If your saves are Instagram reels and YouTube talks, this is the whole difference.
• Semantic search is the default, not an experiment. Karakeep shipped semantic and hybrid search modes in v0.33 as experimental: off by default, embeddings to regenerate yourself, and its release notes say results might not be accurate yet. Deepmark's search has been hybrid semantic + keyword from the start, tuned to 89% top-5 recall on vague from-memory queries (54 of 61 hand-written queries against my own library; small sample, take it as directional).
• Your saved things arrive on their own. X bookmarks, Instagram saves, YouTube Watch Later and Liked sync in the background from your own browser session. Karakeep saves social posts one link at a time; the platforms' own saved lists can't be pulled in.
• Zero operations. No Docker, no Meilisearch to keep alive, no API keys to rotate, no inference bill of your own. That's the actual product: the pipeline exists so you don't run one.

The trade: Deepmark costs more than a $4 cloud plan, can't be self-hosted, has no mobile apps yet, and export is by email. If those are dealbreakers, Karakeep is the better pick and I'd rather you knew now.

Is Karakeep the same thing as Hoarder?

Yes. Hoarder was renamed Karakeep in April 2025 after a trademark dispute. Same project, same maintainers.

Can Deepmark be self-hosted?

No, and there's no plan for it. The indexing pipeline is the product, and it runs hosted. If self-hosting is a requirement, use Karakeep or Linkwarden.

Can I import my Karakeep library?

Via bookmarks HTML: Karakeep exports it, Chrome imports it, and Deepmark's extension indexes the tree from there. No one-click importer yet. Getting data out of Deepmark is currently by email to support, which you should weigh honestly.