Knowledge Graph

The Knowledge Graph page (/graph in Analytics) is a live, zoomable canvas showing your ontology's actual entities (instances) and the relationships between them. You can navigate it like a vault — zooming out reveals the type-level structure; zooming in reveals individual records and how they connect.

What the graph shows

Each glowing dot is an entity from your ontology. Node size reflects how many connections that entity has: highly connected hubs appear larger; isolated records appear smaller. Edges represent the declared relationships between entities from your link types.

When you zoom far out, the canvas shifts to a cluster overview — object types and communities of related entities appear as labeled super-nodes, giving you the shape of your data at a glance. Zoom back in and individual entities resolve again.

Colors and sizes

Navigating the canvas

• Pan — click and drag on an empty area of the canvas.
• Zoom — scroll wheel or the zoom controls (bottom-right). Far zoom shows the type/cluster overview; near zoom shows individual entities.
• Drag a node — grab any entity and drag it. The physics simulation responds live and the graph settles naturally when you release.

Searching for an entity

The search pill at the top of the canvas searches across all bound text values in your ontology. As you type, matching entities are suggested with their type label. Selecting a result:

• Jumps the camera to that entity.
• Loads its immediate neighborhood — all entities it is linked to via any declared relationship.

Investigating connections

Ego-expand and pin

Click any node to select it and open the Inspector (right panel), which shows its properties and connections. From the inspector — or by double-clicking the node directly — you can expand it to load its neighbors onto the canvas.

Pinning a node fixes it in place so it does not drift while you continue exploring. Double-click a pinned node to expand it further (its connections' neighbors). Unpin by clicking the pin icon in the inspector.

Path mode — tracing A to B

Switch to Path mode in the investigation bar to find how two entities connect:

1. Click a starting entity (A).
2. Click a destination entity (B).

The canvas highlights the shortest path between them and dims everything else. Shared neighbors — entities connected to both A and B — are highlighted in a distinct color so you can see who sits between the two.

Hubs — finding the most connected entities

Switch to Hubs mode to size and color nodes by their connection count (degree) or betweenness centrality. A ranked list of the top entities appears in the side panel so you can jump directly to the most central records in your ontology.

Clusters

Switch to Clusters mode to highlight the Louvain communities in your data. Each cluster gets a distinct color and a hull outline; the cluster coloring is also reflected in the legend. Use this to identify groups of tightly related entities.

Map mode

The Graph / Map toggle (bottom-center) switches to a geographic view of every bound object type that carries coordinates. Instead of loading every row, the map shows count bubbles — each bubble is an exact, clearance-correct count of the records inside that area at the current zoom. Zooming in (or clicking a bubble) subdivides the areas until individual records appear as real markers; relationships render as arcs between areas, weighted by the exact number of linked pairs.

The badge strip along the bottom states the exact in-view total per type. When a list had to be cut short, the total is shown as a lower bound (for example 12,400+) — nothing is sampled silently.

Observation trails

Types whose data is a stream of timestamped observations (position fixes, sensor readings, sightings) collapse to one node per real-world entity on the canvas, with an exact observation count in the inspector. To see the movement history behind such an entity:

1. Click the entity node to open the Inspector.
2. Click Show trail on map.

The view switches to Map mode and draws the entity's observations as a time-ordered track, fitted to the viewport, ordered by the binding's mapped timestamp property. The trail chip at the top states exactly how many observations are drawn out of the entity's total; Load more extends the track page by page. Observations without readable coordinates are counted and declared on the chip rather than being invented on the map.

Trails require the binding to declare identitySpec.entityColumns (entity collapse) and to map a timestamp property (such as observedAt or seenAt). If either is missing, the graph tells you why the trail is unavailable instead of showing an empty map.

Controls and settings

Settings panel

The collapsible Settings panel (top-left) has five sections:

Adjusting a Forces slider takes effect immediately — the physics simulation reacts in real time.

Legend

The legend (bottom-left) lists the active color groupings with live counts. Click a type or cluster name to toggle its visibility; hiding a group hides all its nodes and the edges that connect only to hidden nodes.

Other controls

Diagnostics

The diagnostics chip in the page header separates problems from by-design notes:

• A red N issues chip means something needs fixing — reference link types with zero or unknown edge counts, or bindings that are not ready.
• A quiet N notes chip appears when there is nothing to fix but there are informational entries (geo-inferred link types, skipped bindings).

Click the chip to open the grouped diagnostics panel. Actionable groups come first:

• Bindings not ready — a required column or table is missing; fix it in Bindings.
• Link types without edges — 0 means the join ran but no column values matched (check the column map); ? means the count could not be computed (usually an endpoint type with no binding).

Informational groups follow:

• Geo-inferred link types — proximity links compute their edges from instance coordinates when the graph renders, so a schema-level edge count is not applicable. These are notes, not failures.
• Skipped bindings — binding kinds the graph engine does not project (for example memex or manual bindings).

Each actionable row carries a fix link that takes you straight to the right place in the Bindings tab:

• A ? link type (an endpoint with no binding) shows Bind <Type>, which opens the New Binding dialog pre-targeted to that object type.
• A 0 link type shows Edit column map, which scrolls to the source binding and expands its column map so you can repoint the join column.
• A not-ready binding shows Fix binding, and the footer has Open Bindings editor as a catch-all.

Selecting an ontology and branch

The header dropdowns let you switch between ontologies (org-wide and workspace-level) and between branches. The graph always reflects the schema and data on the selected branch, including unpublished proposal branches — useful for previewing a schema change before publishing it.

Related

• Object types — define the entity types (nodes) shown on the graph.
• Link types — define the relationships (edges) between them.
• Bindings — connect object types to real data and enable edge inference.
• Branches & proposals — the branch shown in the graph picker.