The graph view: reading your 3D knowledge graph

You've been saving notes for a few weeks. The file tree gives you a list, and nothing more. It won't tell you that your supplier-meeting note names the same engineer as your note about last spring's outage.

The knowledge graph shows you exactly that: every note, every entity pulled from them (people, organizations, places, ideas), and every relationship between them. They appear as glowing spheres, wired together by colored lines, turning slowly on screen.

This lesson is about learning to read that picture: what the colors mean, how to move the camera, and how to click any entity to see the exact sentence that put it on screen. The floating control menus (filters, view settings, export) get their own lesson next.

Open it and move the camera

Click the waypoints icon in the top-right of the titlebar (its tooltip reads "Knowledge graph"), or press ⌘⇧G (Mod+Shift+G, which you can rebind).

The graph opens as a single tab in the tab strip. Press the shortcut again and you're back in your most recent document. Right beside it sits the "Entity table" button, which shows the same entities and relations as sortable rows and columns.

• One shared 3D scene. It also sits behind an empty new chat as its backdrop.
• "Open on the graph" on an entity page, and the neighbor chips throughout the app, bring you here with that entity already isolated.
• In graph view, a Settings-style left rail switches between "Graph" (the 3D canvas) and the tuning sections. Those sections belong to the lesson on graph controls.

Left-click drag     rotate (free trackball, no pole lock)
Right-click drag    pan
Shift + left drag   pan
Scroll              zoom
⌘⇧G                 open / leave the graph tab
⌘F                  "Find in graph" while the graph tab is active

The graph is never frozen: left alone, it spins slowly, about one turn every 21 seconds. Hover over it and that eases to about a third of the speed. Any drag, zoom, or click pauses it for 3 seconds.

A hint in the corner spells out those same three moves. The whole floating layer fades away while you interact, or while a card is open.

The floating circular-arrow Reset button (its tooltip reads "Reset to home view") re-frames the whole graph. But if you're inside a focus, say an isolated entity or a drilled-in theme, it peels back one layer at a time instead of jumping straight home.

Three kinds of spheres, seven kinds of lines

Every sphere is one of three node kinds, and each responds to a click differently.

The lines are typed and color-coded too. Directional edges are drawn as gradients, and relation edges print their predicate, like "works at" or "founded by", in small muted text along the line.

Hover, click, isolate

Hover any entity and a little card follows your cursor, showing its title and a summary like "3 connections · 2 notes". The node you're hovering and its immediate neighbors stay bright while everything else dims. That makes a crowded region easier to read.

• Click an entity node and the graph isolates it. The node and its immediate neighbors stay, everything else shrinks away, and the selection card opens. Click the isolated entity again to leave isolation.
• Click a note node and it isolates and opens the file in the editor as a blurred preview. A second click switches you fully to the editor.
• Click empty space to dismiss the open card, or to step back out of the current isolation layer.
• Click a theme super-node (in Map mode) to open the theme inspector card.

⌘F searches the graph itself. Type a name, press Enter, and the camera flies to the match. The search box and its shortcuts belong to the lesson on graph controls.

The entity profile card

Click an entity and the glass card in the top-left fills with its details. A plain list of edges appears instantly, before the full profile loads. A shimmer reading "Ranking connections…" then settles into the ranked view.

• A meta line, "Appears in {count} notes", plus a colored type badge. The badge is an editable pill listing all 11 entity types and "Clear type override". Untyped entities show a dashed "Set type…" instead.
• "Connections": neighbors grouped by predicate and ranked by relevance, with direction arrows (→ / ←). Click a chip to hop to that entity. Long groups end with "more…".
• "In your words": your verbatim statements about this entity, grouped by kind (Decisions, Claims, Quotes, and more). Where a later note revised one, it earns a "Superseded" or "Retracted" badge.
• "Also known as": merged synonym chips. The unlink button splits a wrong merge and rebuilds entity resolution.
• "Found in": the source notes. Click one to open it in the editor.
• "Related code": code symbols related in meaning to the entity, shown only when code intelligence is on.

Themes, communities, and colors

Out of the box, nodes aren't colored by topic. Build the topic map once (left rail → "Topic map" → "Build topic map") and Consilience finds themes: clusters of related entities, each one named and summarized for you. Once it's built, nodes take on their theme's color and "Map mode" shows up in the view settings.

• "Map mode" collapses the graph to one super-node per theme, so you're looking at continents instead of cities. Click a theme to open its inspector card.
• "Focus" drills into that theme's subgraph. A floating "← {theme} · Back to map" pill leads you back out.
• "MOC" creates a summary note that gathers the theme's notes and concepts.
• An "Isolated" badge marks a theme that barely touches the rest of the workspace.

Theme colors are separate from "Color by type", a view-settings toggle that colors entities by their kind (Person, Organization, Place, and so on). It lights up once enough of your entities carry a type. On an older workspace, run "Fill in missing types" first and the toggle appears.

Isolation

Shrinking everything except one node and its immediate neighbors. The Reset button peels focus layers back one at a time.

Selection card

The per-entity profile that opens on click: type, connections, statements, aliases, source notes.

Source sentence

The exact sentence in a source note that states a connection, shown when you hover a neighbor chip.

Theme (community)

A named, summarized cluster of related entities that the topic map detects across the workspace.

Map mode

A view that collapses the graph to one super-node per theme.

Ambient rotation

The graph's permanent slow spin. It eases on hover and pauses for 3 seconds after you interact.

Performance at scale

The graph renders on your GPU, and your camera and layout survive leaving and coming back. Settled node positions are cached per view (workspace, map, theme), so returning never re-scatters the layout.

Even at several thousand nodes it stays readable. Sphere detail eases off on its own as the graph grows, and labels are budgeted per screen cell. A crowded region shows a few readable names rather than a blur, and raising the "Detail" slider draws more.

You can read the picture now. Next up: the floating menus themselves, one by one, from filters to view settings to export.