The example below is an AI-generated ontology of Ray Dalio’s economic principles. Open it in a new tab and follow along. To build your own, generate an ontology for any topic or document on the Apps page, or ask your LLM to do it through the InfraNodus MCP server. The order of the steps follows the cognitive variability framework: shift the scale, then shift the intent.
1. Get an Overview of the Structure
Think of the graph as a social network of ideas: which concepts hang out together, which live at the edge, which groups would gain from meeting. Bigger nodes are more influential, each color is a topical cluster, and the Discourse Structure at the bottom of Analytics tells you what to do next: bridge the clusters if the graph is dispersed, develop the smaller ones if it is focused or biased.
Do: open the graph and look at it before clicking anything.
Unlike a chat AI: a chat lists the main topics without telling you how they connect, which ones sit at the periphery, or where the gaps are. The graph shows all three at once.
2. Zoom Into a Cluster
Read the graph in a non-linear way: pick the topic that matters to you and look only at it. You check whether the cluster covers what it should, read the statements it was built from, and ask the AI to add ideas to this one topic instead of reading another long summary of everything.
Do: click a topical cluster, then AI: Summary.
Unlike a chat AI: the cluster comes from the algorithm, so it is the same every time you open the graph. You see the statements behind it and send only those to the AI, without filling its context with the rest of the text.
3. Find the Gaps
Two clusters that should be connected but are not mark a blind spot. Bridging it gives you a new question to work on, and it fixes the knowledge graph so it stops missing that link. The same move steers an agent’s reasoning toward original ideas.
Do: open Analytics › Gaps, click the circle between two clusters, and choose AI: Question for a research question that bridges them.
Unlike a chat AI: LLMs give you the most predictable answer, which is how they are built. The gap points them at what is missing, so the question, and the ideas that follow, are less generic.
4. Explore the Relations
The detailed view shows the statements that link two concepts, so you see how the clusters connect and can correct a relation that is wrong or missing.
Do: switch to Analytics › Concepts and click any two nodes to see how they are related.
Unlike a chat AI: you check and edit the relations yourself, so the graph you hand to your AI tools is one you have verified.
5. Use the Graph as a Reasoning Map in LLM Tools
Once you have checked and fixed the graph, connect it to Claude, Claude Code, Cursor, or ChatGPT through the InfraNodus MCP server. The structure becomes a set of rules your agent can follow and you can verify. The same graph works as agent memory for procedures, rules, and learnings.
Do: click the Brain icon next to the graph’s name and follow the steps to add it to your LLM as a reasoning graph.
Unlike a chat AI: instead of a plain-text prompt or skill, your agent gets a graph of the principles and rules to follow in this context, available from any client or session. You can change it any time, in the interface or through MCP.
6. Save Your Insights to Project Notes
Notes can become a new graph or an extra system prompt for an LLM that reasons with this graph.
If the graph is a tree of knowledge, Project Notes is the basket where you keep the best fruit.
Do: open Project Notes and write down your thoughts, or click Save on an answer in the AI panel.
Unlike a chat AI: a long chat gets messy as it grows. Here you keep only the best ideas in a separate folder and build on them later.