One graph, and everything is in it

Cells

A cell is the unit of everything. Notes, todos, lists, events, files, folders, people, chats, messages, ideas, issues, concepts, libraries, slides and agents are all cells. They differ in how they render and what operations make sense for them, but they share one identity model, one storage format and one sync path.

The practical consequence is that you never hit a wall where a thought is the wrong shape for the tool. A todo that turns out to be a project does not need to be migrated into a different container — you link things to it and it becomes one.

Links, in both directions, always

A link is itself a cell, and every link type has a reverse. Link a note as an example of a concept, and the concept shows it as an example without anyone maintaining a second reference. There is one link; you are simply looking at it from one end or the other.

Because links are first-class, they carry meaning. parts, examples, related, references, questions, answers, problems, benefits, gotchas and others each say something different about why two cells belong together. Backlinks are free and never stale.

The lens

You view one cell at a time. Its leaders sit above it and its followers sit with it, and moving to a neighbour re-aims the lens rather than loading a page. Nothing is copied, nothing is duplicated, and there is no separate "backlinks panel" to go check — the other side of the graph is already on screen.

Sync that doesn't lose work

Your graph is held locally and synchronised in the background. Reads never touch the network, so opening a cell and searching are both instant. Writes are optimistic and queue while you are offline.

Conflict handling is per field rather than per document: if you edit a cell's body while a colleague renames it, both changes survive. Deleting is a "bury" rather than a destroy, so anything can be brought back.

Search

Two layers, both local. Full-text search runs over the graph already in your browser. Semantic search uses an embedding model that runs on your own device, with the vectors stored locally — the meaning of your notes is not shipped to a server to be indexed.

Libraries and the knowledge engine

A library is a shelf, and a shelf may hold other shelves. Libraries are how a team keeps curated material — standards, decisions, reference notes — organised without inventing a parallel filing system.

On top of that sits the knowledge engine: concepts act as doors into topics, and each concept links to the atoms of knowledge underneath it. When an agent works on something, the relevant material is assembled from the graph rather than pasted into a prompt by hand.

Agents share your graph

Cellara's agents are not a chat window with an API key. They read and write cells through the same operations you do, with the same permissions, and their work shows up in the graph where you can inspect it, link to it and correct it. You can interrupt one mid-task, steer it, or cancel it outright — and a cancelled task reports itself as cancelled rather than quietly claiming success.

Interoperability

Cellara exposes a scripting interface for reading and writing the graph programmatically, and there is a connector that lets ChatGPT create and query cells in your workspace.