> ## Documentation Index
> Fetch the complete documentation index at: https://docs.bowerlabs.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Knowledge graph visualizer

> See your entire workspace as a force-directed graph — every note, protocol, attachment, and the relationships between them — with permission filtering, light and dark mode, and an opt-in 3D view.

**Shipped 2026-05-09 · Visualisation**

We've talked about Bower's knowledge graph for a while — the GRI-based
edges, the inferred relationships, the link chips on every note. Today
you can finally *see* it.

## What's new

A new **Knowledge Graph** page renders your entire workspace as a
force-directed graph. Notes, protocols, attachments, projects, and
people show up as nodes; explicit links and inferred relationships show
up as edges between them.

* **Click a node** to select it and dim the rest of the graph to the
  selected node and its 1-hop neighbours.
* **Hover a node** for a quick label preview.
* **Click an edge** to inspect the relationship type and origin
  (manual link, inferred, or system-derived from `created_by` /
  `parent collection`).
* **Filter by node type** in the side panel — show just protocols,
  hide attachments, etc.
* **Search** to fly the camera to a specific entity.
* **Open the entity** from the detail panel to jump straight into the
  underlying note.

## Two renderers, one graph

The default is a **2D Canvas2D renderer** — fast, readable in light or
dark mode, and at home alongside the rest of Bower's UI. A toggle in
the filter panel switches to an **opt-in 3D view** powered by Three.js,
with a dark backdrop and bloom-style glow on nodes — useful for getting
a sense of the shape of a large graph.

Mobile and reduced-motion users always get the 2D renderer.

## Permission-filtered

The graph respects every layer of your workspace permissions. You only
see nodes and edges you'd be allowed to read in the underlying notes —
restricted notes are filtered out at query time, not just hidden in the
client. This is checked against the same canonical permission resolver
that gates note reads, so the visualizer can't drift from the rest of
the app.

## Inferred edges, clearly marked

Edges that came from [Bird's inferred relationships](/changelog/2026-04-26-inferred-relationships)
are drawn with a dashed line so they read differently from the explicit
links you (or the system) created. Same convention as the link chips on
the note view.

## Limits

The visualizer is built for workspaces up to **5,000 nodes**. Beyond
that, performance starts to degrade — we'll add server-side clustering
and viewport-windowed loading as workspaces approach that ceiling.

## Related docs

* [Inferred links](/ai-and-search/inferred-links)
* [Linked entities](/organisation/linked-entities)