Two principled, opposite designs
Anytype is one of the more thoughtful note apps in the privacy-first corner: open-source, local-first, end-to-end encrypted, P2P sync, and an object model that lets you describe what each note is. If you want to own every byte and never trust a server with your knowledge, Anytype was built for you.
Hjarni starts from a different assumption. AI assistants should be able to read and write your notes through an open standard. That requires a server they can talk to. Hjarni embraces that constraint: hosted in the EU, Markdown by default, with a built-in MCP server.
Objects versus documents
Anytype's object model is powerful. Each note is a typed object with relations to other objects, much like Notion's database concept but local-first. Anytype now ships an official MCP server, but it lives outside the app: you install the npm package, generate an API key in the desktop app, and configure your AI client.
Hjarni keeps notes as plain Markdown documents in folders. The model is simpler, less expressive, and much easier for AI assistants to consume. The MCP server is built into the product and connects via OAuth instead of a CLI install.
Anytype is for people who put privacy first. Hjarni is for people who want their AI to read what they write.
A concrete workflow difference
You're managing a research project. In Anytype, you define object types (Paper, Author, Argument) and link them through relations. Synthesis happens in the app, on your devices, encrypted in transit.
In Hjarni, you keep the same research as a folder of Markdown notes. You ask Claude to read across the folder, find recurring arguments, and draft a synthesis with citations to specific notes. Claude reads through the MCP server and writes the summary back as a new note in the same folder.
When Anytype is the better fit
If end-to-end encryption, local-first, and open-source are non-negotiable, Anytype is the stronger choice. It is also a good fit for people who enjoy modeling their knowledge with objects and relations instead of folders and prose.
Why some Anytype users add Hjarni
Anytype's MCP server closes the biggest AI gap, but it stays a separate install. The people who add Hjarni want a knowledge base that connects to ChatGPT, Claude, Cursor, or Claude Code through OAuth in a couple of clicks, with folder-level AI instructions and team sharing baked into the same product. They are comfortable with EU-hosted SaaS as long as they can export their notes as Markdown at any time.
Migration and ownership
Anytype exports as Markdown. Hjarni's Markdown ZIP importer takes that export and preserves structure. The reverse is also true: Hjarni exports the full knowledge base as a Markdown ZIP at any time. No lock-in either direction.