Roadmap
This is a direction, not a schedule. There are no dates here on purpose — a date is a promise you then have to keep, whereas the order below reflects priority and changes far less often than a calendar would.
Next up: a graph you can actually read
On a large base the link graph turns into a hairball, and that is the next thing being fixed. Zoomed out you get clusters by topic instead of a spray of dots; a click drills into one, and search opens the right place on its own. Around the note you have open, its connections read at a glance, and you start getting "related but not linked yet" hints.
Alongside it: the note feed gets finished, and the interface gets more responsive.
After that: AI deeper inside the base
Today an agent arrives from the outside over MCP. The next step is working with AI inside the knowledge base:
- a chat next to the note you have open, aware of the context of your space;
- assistants that keep the base in shape — duplicates, missing tags, stale notes, orphans — opt-in, with every edit visible and reversible;
- control over what exactly the AI gets as context, and a way to inspect why a search missed;
- your base as a source for outside assistants, answering with links to your own notes.
The same stretch brings comfortable work from a phone or tablet, and an interface-language switch.
Then: a team on one server
Everything you need once there are many people and many spaces: an activity log with filters and an export, a split between owning the server and administering an organisation, storage quotas that warn you before they bite, and outbound notifications when notes change. Alongside those — keeping notes in external storage, and links that span different spaces.
Your own machine
The "Notarium on my laptop" scenario: notes you mark as local live on the device and sync back when the network returns, while recent and favourite ones stay readable offline. Search gains a compact model that will not cook your CPU, and the workspace becomes yours to arrange — panels drag between groups, and the layout is remembered.
A separate desktop application stays a direction: the core is already decoupled from the web, but the work itself has not started.
Working together
Discussion around a note: comments and threads anchored to a specific passage, @ mentions of colleagues, notifications for what concerns you personally, and a sense of who is around right now. Permissions get finer — a role in the space plus access to a particular folder.
Several people editing one note at the same time is a direction under exploration, with nothing promised.
Extensibility
The base stops being a closed box: add-ons install from a link, are signed by their author, ask for permissions in plain language, and switch off in one click. An add-on can contribute its own panel, settings tab or command, colour the graph its own way, and derive material that sits next to the source note. Styling by hand — folder colours and custom graph groupings — becomes available too.
What will not happen
These are not deferred items. They are settled positions.
No end-to-end encryption. Search, semantics, revision history and agents all require the server to see your content. Privacy here comes from the server and the files being yours, not from hiding the data from the server itself. If zero-knowledge is the actual requirement, the honest answer is that other products serve it.
No block editor. The unit of data is a file-note, not a block inside someone else's database. That is the price of portability: a folder of .md can be picked up and opened with anything.