Creating Concept Maps with Obsidian Copilot

[WetHat]

The Background Story

To build an effective personal knowledge graph, I frequently use Obsidian Canvases to create concept maps, which help me better understand and remember technical material.

What are Concept Maps (CMAPs) Good For?

Concept maps are visual tools for organizing and representing knowledge. They typically include concepts enclosed in boxes (nodes), and relationships between concepts indicated by connecting arrows (edges). These relationships are articulated using linking phrases, specifying the connection between two concepts.

Benefits:

• Concept maps provide a visual framework that makes it easier to see the structure of information.
• By explicitly linking concepts, they promote a deeper understanding of the relationships between different ideas.
• Concept maps communicate complex ideas in a clear and concise manner.

Downside:

• Producing them requires a significant amount of manual effort.

Why use Obsidian Canvas for Concept Maps?

Obsidian Canvas offers a flexible, visual workspace where you can freely arrange and connect ideas. This allows you to:

• integrate notes into canvases.
• create visually appealing maps by leveraging full Markdown rendering capabilities.

Using Obsidian Copilot to make Concept Maps

I have previously created custom prompts for Obsidian Copilot that generate Mermaid diagrams. While effective, these diagrams are challenging to edit manually, making them less suitable for the inherently dynamic and interactive nature of concept maps. As a result, I decided to experiment with AI generated Obsidian canvases, albeit with modest expectations regarding their usability.

Surprisingly, both the copilot-plus-flash and GPT-4.1 models can generate quite useful concept maps in Obsidian Canvas format when provided with the custom prompt included at the end of this post. The examples below illustrate concept maps generated from a chapter of a tutorial on TypeScript Generics using both models.

It is important to note that both models encounter challenges with map layouts and node sizing, often generating dense maps that require some manual adjustments to separate the nodes.

Concept Map made by 'copilot-plus-flash'

The map displayed below has undergone initial manual cleanup and currently appears satisfactory. However, further manual corrections and enhancements will be necessary. For example, the representation of the variance concept does not accurately capture its intended use and requires revision.

Concept Map made by 'gpt-4.1'

The map displayed below has also undergone initial manual cleanup. Its current state accurately represents the key points and their relationships. However, still further manual adjustments are needed to align the map more closely with my preferred way of thinking.

Conclusion

Using Obsidian Copilot to generate concept canvases significantly reduces the manual effort (grinding) involved in mapping out concepts. However, it remains essential to review each concept and relationship within the maps to identify and correct any inaccuracies, omissions, or misrepresentations. This critical review process is not a waste of time; it not only ensures the accuracy of your concept map but also deepens your understanding of the material and aids in memorization.

Custom Prompt: Concept Maps in Obsidian Canvas Format

Note: You need to manually copy the LLM response in to a .canvas file using a text editor.

## PURPOSE & SCOPE
This prompt instructs the LLM to transform the key points, ideas, concepts, and relationships found in the provided content into a comprehensive, visually structured concept map using the Obsidian canvas JSON format. The map should make complex information accessible, trustworthy, and visually intuitive for academic and professional audiences by clearly illustrating how ideas interconnect.

## OBJECTIVES
- Extract and interpret all key points, ideas, concepts, subpoints, and nuanced relationships from the provided content.
- Represent the concept map in valid Obsidian canvas JSON format, ensuring all nodes and edges are uniquely identified (`id` property).
- Clearly label each relationship with descriptive, meaningful terms (e.g. "leads to", "enables", "contrasts with").
- Design a layout that:
	- is visually clear.
	- minimizes edge/edge and edge/node intersections.
	- clusters related nodes to indicate relationship strength.
	- always avoids node/node overlaps.
- Format edge labels as plain text and node text as well-structured Markdown (using headings, lists, tables, LaTex math expressions, and code blocks if they add clarity).
- Ensure the output is readable, reliable, and suitable for researchers, students, and professionals.

## INSTRUCTIONS
1. **Content Analysis**: Carefully read and analyze the content provided below `INPUT` and extract all key points, ideas, concepts, subpoints, and both explicit and implicit relationships.
2. **Depth and Detail**: Where possible enrich the map by:
	- highlighting hierarchies.
	- including examples or subpoints.
3. **Concept Map Construction**:
	- For each concept, create a node with a unique `id` and add a description of the point, idea or concept it represents as well-structured Markdown.
	- For each relationship, create a edge with a unique `id`, and label each edge with a clear, descriptive term that accurately reflects the nature of the relationship .
	- Ensure each node’s four connection points (top, right, bottom, left) have only incoming or outgoing connections (any number), not both. This makes it easier for the user to follow the connections and manually modify the node layout.
4.  **Layout and Design**:
	- Arrange nodes to minimize edge intersections and overlaps, using a cluster-shaped layout where proximity indicates relationship strength.
	- Space nodes apart for readability and balance the map visually.
	- Place a summary node at the top, containing 2-3 sentences that explain the map’s main structure and highlight key connections.
5. **Tone and Style**: Maintain a formal, precise, and informative tone suitable for academic and professional contexts.
6. **Response Format**: 
	- Output only a Markdown code block containing the Obsidian canvas JSON.
	- Do not include any text outside the code block.
	- The JSON must include all nodes, edges, and the summary node as described.
7. **Response Evaluation:** The response must:
	- Fulfill all objectives listed above
	- Accurately reflect the content and relationships in the input notes.
	- Be clear, well-organized, and easy to interpret.
	- Demonstrate depth by including nuanced or non-obvious connections.
	- Remain relevant to the needs of the target audience.

## EXAMPLE OUTPUT
```json
{
	"nodes":[
		{"id":"2","type":"text","text":"Design","x":380,"y":100,"width":170,"height":50},
		{"id":"5","type":"text","text":"Deployment","x":380,"y":300,"width":170,"height":50},
		{"id":"6","type":"text","text":"Maintenance","x":-40,"y":300,"width":248,"height":50},
		{"id":"3","type":"text","text":"Implementation","x":720,"y":100,"width":200,"height":50},
		{"id":"4","type":"text","text":"Testing","x":720,"y":300,"width":200,"height":50},
		{"id":"1","type":"text","text":"Requirements Gathering","x":-40,"y":100,"width":248,"height":50},
		{"id":"96348cd5822d56b8","type":"text","text":"# Summary\nThe concept map illustrates a simplified software development lifecycle, outlining the sequential and iterative relationships between key stages.","x":296,"y":-140,"width":338,"height":180}
	],
	"edges":[
		{"id":"e1","fromNode":"1","fromSide":"right","toNode":"2","toSide":"left","label":"informs"},
		{"id":"e2","fromNode":"2","fromSide":"right","toNode":"3","toSide":"left","label":"leads to"},
		{"id":"e3","fromNode":"3","fromSide":"bottom","toNode":"4","toSide":"top","label":"results in"},
		{"id":"e5","fromNode":"5","fromSide":"left","toNode":"6","toSide":"right","label":"requires"},
		{"id":"e4","fromNode":"4","fromSide":"left","toNode":"5","toSide":"right","label":"precedes"},
		{"id":"e6","fromNode":"6","fromSide":"top","toNode":"1","toSide":"bottom","label":"provides feedback for"}
	]
}
```

## INPUT
{}