Canvas Knowledge Graph

An Obsidian plugin that reads a JSON knowledge graph and renders it as a native Obsidian canvas — spatial layout, colored tiers, typed edges, live file embeds. Built to pair with LLM Wiki, but works with any graph file that matches the expected schema.

How it works

1. Reads a graph.json file from your vault (by default produced by LLM Wiki at _schema/graph.json).
2. Computes a spatial layout using one of three algorithms: force-directed, radial tiers, or grid-by-type.
3. Writes a standard Obsidian .canvas file with positioned nodes and typed edges.
4. Opens the canvas so you can pan, zoom, and click straight through to the underlying notes.

Because the output is a real .canvas file, everything you'd expect from Obsidian canvas works — live file previews, manual edits, selection, search, linking into other notes.

Commands

• Render full knowledge graph — lay out and render every node in the graph
• Render graph filtered by tier — pick a memory tier (working, episodic, semantic, procedural)
• Render graph filtered by entity type — pick any entity type present in the graph
• Render neighbourhood of current note — render a 1-hop subgraph centered on the open page
• Open last rendered canvas — jump back to the most recent output
• Show graph statistics — counts by tier, entity type, and average confidence

Layout algorithms

Algorithm	When to use
Force-directed	Organic clusters, good default for exploratory browsing
Radial	Tier-centric view — each memory tier becomes a concentric ring
Grid	Categorical view — nodes grouped into columns by entity type

All three support grouping by community, scaling nodes by connection count, and a configurable spring length / repulsion for the force-directed case.

Appearance

Nodes can render as either:

• File embed — a live, interactive Obsidian preview of the underlying wiki page
• Text card — a compact summary card with title, tier badge, and confidence score

Edges are colored by type (related-to, part-of, derived-from, contradicts, supersedes, supports, example-of, prerequisite) and can optionally show their relationship label.

Setup

Prerequisites

• Obsidian 1.5.0+
• A graph.json file somewhere in your vault. LLM Wiki produces one automatically, or you can author your own — see the schema below.

Install

1. Copy the canvas-renderer folder into your vault's .obsidian/plugins/ directory
2. Run npm install && npm run build inside the plugin folder
3. Enable "Canvas Knowledge Graph" in Obsidian Settings → Community Plugins
4. Open the plugin settings and confirm the path to your graph.json
5. Run the "Render full knowledge graph" command

Graph schema

The plugin expects a JSON file with this shape:

{
  "version": "1.0",
  "nodes": {
    "wiki/retrieval-augmented-generation.md": {
      "title": "Retrieval-augmented generation",
      "entityType": "concept",
      "tier": "semantic",
      "confidence": 0.78,
      "community": "llm-architecture"
    }
  },
  "edges": [
    {
      "fromPage": "wiki/retrieval-augmented-generation.md",
      "toPage": "wiki/vector-databases.md",
      "relation": "related-to",
      "label": "uses"
    }
  ]
}

Node paths double as unique IDs and as vault paths for file embeds.

Settings

Setting	Default	Description
Graph data file	_schema/graph.json	Vault-relative path to the graph JSON
Canvas output folder	—	Where generated canvas files land
Canvas filename	knowledge-graph.canvas	Default filename for the full graph
Layout algorithm	Force-directed	Force-directed, radial, or grid
Node width / height	400 × 180	Base card dimensions in pixels
Spacing multiplier	1.0	Spreads nodes further apart when increased
Force iterations	200	More iterations means cleaner layout, slower render
Spring length	300	Ideal distance between connected nodes
Repulsion	6000	Push strength between unconnected nodes
Scale by connections	On	Heavily-connected nodes render larger
Group by community	On	Cluster nodes that share a community ID
Node style	File embed	Live previews or summary text cards
Show edge labels	On	Display relationship labels on arrows
Show confidence / tier	On	Show badges on text cards
Minimum confidence	0.0	Hide nodes below this threshold
Filter tiers / types	—	Comma-separated include lists

Pairing with LLM Wiki

LLM Wiki maintains the graph automatically as you ingest, lint, and crystallize sources. Install both plugins and the canvas stays in sync with your wiki as it grows.

Development

cd canvas-renderer
npm install
npm run dev    # watch mode — rebuilds on save

Symlink the plugin folder into your test vault's .obsidian/plugins/ directory. Reload Obsidian (Ctrl+R / Cmd+R) after rebuilds.

License

MIT