# Explore a graph on the canvas

Read a source's domain graph as a drawing, expanding one relationship at a time, inspecting an entity, saving a view and exporting what you can see.

**Canvas** draws the domain graph. It does not draw all of it: a graph of any size will not fit and would not be readable, so the canvas starts with an overview and you expand outwards from what interests you.

**Who can use it:** the **Builder** role, or a Member holding the **Knowledge Admin** grant.

## What is on the canvas

A node is an entity. An edge is either a real relationship or, where the canvas has folded a crowd of them away, a dashed edge carrying a count. That count is grounded: either the exact number the schema knows, or a lower bound written as "at least", never an estimate dressed up as a total.

Entities whose only role is to hang off something else, a label with no outgoing relationships, are not drawn as nodes at all. They fold into the inspector as facts of the entity they belong to, which is what keeps the drawing readable. They can still be found by search, and the inspector can put one on the canvas if you need to see it.

**Node types** on the left is both a legend and a filter: selecting a type hides or shows it. Types that hold only facts are tagged as such.

## Moving around

- **Search** the graph by name. The search sweeps the larger labels and caps what it returns, so it is a way in rather than a way to enumerate.
- **Overview**, **Fit** and **Re-layout** reset the view; the plus and minus zoom.
- The keyboard works throughout: the arrow keys pan, plus and minus zoom, N and P step through nodes, Enter expands the selected one, Escape clears the selection and 0 fits the graph to the window.

## Expanding

Expanding is one relationship at a time, by design. Selecting a node lists its relationships with the number of neighbours behind each; expanding one pulls in a page of them. **Expand all** does every relationship of that node. A folded crowd is drained page by page, because the graph service pages by key rather than by offset, so there is no way to skip ahead and no pretending otherwise.

Past a few thousand elements the canvas folds the heaviest types back into crowds by itself, to keep moving. It leaves whatever you have selected alone when it does.

## The inspector

Selecting a node opens the inspector: its properties, the facts folded into it (each type collapsible, with a way to put one on the canvas), and its connections, with an expand control and a count per relationship. The **Source** line names the knowledge source the entity came from, resolved to the source's name rather than an identifier.

## Saving and exporting

A view can be saved: the canvas records the steps you took and replays them, so applying a saved view rebuilds it rather than restoring a picture. **Copy link** shares one the same way, within a length the address bar will carry.

Export gives four things, all restricted to what is actually on the canvas: an image, a GraphML file for graph tools, two CSV files of nodes and edges, and the subgraph as JSON.

## Limits and behaviour

- The canvas is built for graphs of roughly ten thousand nodes. Beyond that it degrades deliberately: animation stops, then layout simplifies, then crowds fold.
- A saved view holds a bounded number of steps; a very long exploration cannot be replayed in full.
- Every count on screen is either exact or explicitly a lower bound.

## Related

- [The domain graph](/knowledge/knowledge-bases/domain-graph)
- [Query a graph in the explorer](/knowledge/knowledge-bases/graph-explore)
- [Govern knowledge graphs with schemas](/knowledge/knowledge-bases/graph-schemas)

Source: https://docs.katonic.ai/knowledge/knowledge-bases/graph-canvas
