Step 4 · AI Workflows

AI Workflows: Use InfraNodus Inside Claude, ChatGPT, and Your Agents

Connect InfraNodus to your LLM once. Then ask for a graph of any file, chat, or folder, and your AI comes back with the topics, the gaps between them, a map to reason with, and a memory it can read next time.

By Dmitry Paranyushkin · Updated

An InfraNodus knowledge graph used by an LLM as a reasoning map: GraphRAG retrieves the statements that relate the selected concepts, and the graph insights add context for the model
Your graph as your AI sees it: topics, the concepts that connect them, and the gaps between them.

What you can do once it is connected, each in one short section below, with a link to the full guide: connect, find gaps, reason with a map, keep a memory, and build graphs you reuse everywhere.

1Connect InfraNodus to Your AI

Get your API key from the InfraNodus API panel (needs a free InfraNodus account), then add this address to your AI client. It is the same for every client:

MCP serverhttps://mcp.infranodus.com

Settings → Connectors → Add custom connector, paste the address, click Connect, and sign in with your API key.

Then ask your AI for a graph. This is what a session looks like in Claude Code:

✓ infranodus MCP connected
> Make a knowledge graph of the text files in this folder and find what’s missing
✓ infranodus · generate_knowledge_graph
✓ infranodus · generate_content_gaps
Your notes have 3 main topics:
• Pricing 41%
• Onboarding 33%
• Support 26%
The gap: nobody connects onboarding to pricing.
Try asking: “What would onboarding look like if it explained the price?”
An example session in Claude Code with the InfraNodus MCP server connected.

Setup for every client, step by step →

2Find the Gaps from the Chat

A content gap is two topics in your material that should be connected but are not. Ask for them and your AI calls generate_content_gaps: it gets the two clusters that sit far apart and a question that would bridge them. This works on your notes, a document, a codebase, Google results, or a whole conversation.

Ask your AIUse InfraNodus to find the content gaps in this document / folder / conversation. Suggest a question that would bridge the main one.
InfraNodus Analytics showing a content gap between two topical clusters, highlighted in the graph
The gap your AI receives: two clusters that are rarely linked.

Do not stop at the first answer. Pick the gap that matters for your work and ask the AI to develop it:

Then askDevelop the text I’m working on and generate questions that bridge the gap between [topic A] and [topic B].

Then open the same graph in the web app to read the statements behind each cluster. That is where the idea gets checked against your actual material.

InfraNodus AI Insights generating a research question that bridges the selected content gap
The same gap in the web app: AI Insights turns it into a research question.

How content gap analysis works →

3Reasoning Maps

An LLM predicts one step at a time and has no structure to check its reasoning against. A reasoning map gives it one: a graph of concepts with typed relations such as [influences] or [constrains], written in plain text with [[wikilinks]]. Your AI retrieves the relations between the concepts it is working with via GraphRAG, instead of the most similar passages.

[[market sentiment]] [influences] [[price volatility]] [constrained by] [[liquidity]]

Build it in InfraNodus, check where it is thin or lopsided with the gap analysis from section 2, and connect it to Claude, Cursor, or your agents. Point the model at the graph and it calls retrieve_from_knowledge_base on it:

Ask your AIUse my InfraNodus graph [name] as a reasoning map: follow its relations to answer [your question], and tell me which relations you used.
An InfraNodus graph made available to LLM agents as a reasoning map over MCP
The graph your agent reasons with, opened in the web app.

Build a reasoning graph for your AI →

4Memory Graphs

Claude Code remembers with CLAUDE.md and auto memory: flat files on one machine. A memory graph stores what you and your agents learned as short statements in your InfraNodus account, so Claude Code, Cursor, Claude Desktop, and the web app all read the same memory.

  • Insight memory: the AI saves with memory_add_relations, and memory_get_relations brings it back in the next chat.
  • Agent learnings: with the Project Learnings skill, your agent reads the learn-<project> graph when a task starts and proposes new learnings when it ends. Nothing is saved until you approve the batch.
To saveSave what we figured out in this chat to my InfraNodus memory graph [name], as short statements with [[wikilinks]].
To recallWhat does my InfraNodus memory graph [name] say about [topic]?

Because the memory is a graph, you can see it. Open it in InfraNodus and the gap between two clusters shows what your agent keeps relearning: it knows the deploy routine and the caching rule, but not that one breaks the other.

MEMORY.md as a flat list on the left; the same learnings as a graph on the right, with a deploy cluster and a cache cluster and a content gap between them
The same learnings as a list and as a graph. The missing link is the lesson your agent keeps relearning.

Set up Claude Code memory with a knowledge graph →

More on memory: the memory MCP server, the AI memory stack, and agent memory tools compared.

5Create and Analyze Your Graphs

Any data can become a graph: a PDF, a folder of Markdown notes, a web page or a YouTube video, interview transcripts, a column from a CSV, the Google results for a query, or your codebase with the /infranodus skill. Ask your AI to save it under a name and it calls create_knowledge_graph. The graph is stored in your InfraNodus account, not in the chat.

To createCreate an InfraNodus graph called [name] from [these files / this URL / this conversation] and tell me its main topics.

Once it is saved, any session can analyze it by name: the topics and gaps with analyze_existing_graph_by_name, what one graph has that another lacks with difference_between_texts, and the graph you need with list_graphs and search.

To analyzeAnalyze my InfraNodus graph [name]: what are the main topics, where is the gap, and what does it miss compared to my graph [other name]?

The same graph works in every tool you connect. Build it from Claude today, query it from Claude Code in Cursor or ChatGPT next week, open it in the web app or in the Obsidian plugin to read the statements, or let an n8n agent use it as context. Each new session starts from what the graph already holds.

In a new sessionList my InfraNodus graphs about [topic] and use [name] as context for this task.
A knowledge graph in the InfraNodus graph view inside Cursor: find a gap, generate a question, and send it to Claude Code in the terminal
One graph, several tools: the gap found in the graph view goes straight to Claude Code in the same editor.

All MCP tools for creating and analyzing graphs →

Continue the Workflow


Setup
1.3 — Connect the MCP Server

Setup for every client: Claude, Claude Code, ChatGPT, Cursor, Codex, and local JSON config.

Reasoning
AI Reasoning Graph

Build a reasoning ontology with typed relations, check it for gaps, and hand it to a model.

Memory
Claude Code Memory

CLAUDE.md, auto memory, and a learnings graph shared across machines and clients.

Knowledge Base
Your AI Second Brain

GraphRAG, memory, and the LLM Wiki skill on top of your own knowledge base.