Two AI-aware tools, two different bets
Capacities is an object-based note app with a beautiful interface, strong daily notes, and AI features you pay for inside the product. It is opinionated about structure: every note is a typed object with properties.
Hjarni is simpler and AI-neutral. Plain Markdown documents in folders, a built-in MCP server, and folder-level AI instructions. The AI is whatever you already pay for: Claude, ChatGPT, Cursor, Claude Code.
Bundled AI versus bring-your-own
Capacities bundles its own AI features. Convenient if you want one product and one bill, but you're tied to whichever models Capacities chooses. If a better model ships next week, you wait for the integration.
Hjarni keeps the AI layer outside the product on purpose. The MCP standard means any compatible client can read your notes today and any future one tomorrow. You pick the assistant, the model, and the chat surface.
Capacities ships an AI inside the app. Hjarni hands your notes to the AI you already use.
Objects versus folders
Capacities models knowledge as typed objects: a Book has different properties from a Person, which has different properties from a Meeting. That structure is great when you actually want to query your knowledge like a database.
Hjarni leaves the modeling out. Folders, tags, and prose. The bet is that AI assistants are better at reading well-organized Markdown than at navigating a custom object schema, especially across many users and use cases.
A concrete workflow difference
You're keeping notes on books you read. In Capacities, each book is an object with author, rating, and quotes. You query that object type to find unfinished reads.
In Hjarni, you keep a folder of book notes in Markdown. You ask Claude to find recurring themes across them, draft a reading roundup, or surface unfinished books. Claude reads the folder through MCP and writes back into the same folder.
A capped AI meter versus an open memory
Capacities' AI runs as a chat inside the app on a daily budget. When you reach the cap, the path forward is to plug in your own model-provider API key, from OpenAI, Anthropic, Gemini, or another provider. Either way it is a single in-app assistant, useful for quick questions but not something an outside agent can reach.
Hjarni adds no AI meter of its own. The MCP server lets ChatGPT, Claude, and custom agents search, create, and update the whole knowledge base directly, with no per-day budget from us and no lock-in to one assistant. The case is not that Capacities has no AI. It is that its AI is a metered chat in one app, while Hjarni's notes are an open memory any agent can read and write back into as visible, named notes you can review.
When Capacities is the better fit
If you enjoy modeling notes as objects, want strong daily notes, and prefer a single product with bundled AI, Capacities is well designed for that workflow. The aesthetic is also a real factor: many people stay for the UI alone.
Why some Capacities users switch
The shift usually starts when teams get involved or when an assistant outside Capacities becomes the main thinking tool. Wanting Claude to read the same notes ChatGPT can read. Wanting folder-level instructions so AI behaves differently across personal journaling and team documentation. Hjarni trades the object model for an MCP-native knowledge base.