How to Make OpenClaw (and AI Agents) Smarter with InfraNodus

In this tutorial, we will demonstrate how to use InfraNodus with autonomous AI agents like OpenClaw to get more interesting, novel results from your personal notes and research papers.

We will first describe how you'd use OpenClaw normally, without InfraNodus, so you can see the difference in the output yourself. Then we will demonstrate how InfraNodus' ability to retrieve topics and find gaps between them can benefit your research workflows.

We also recommend you to read our article on setting up OpenClaw in a secure way before starting to ensure that you don't expose your sensitive data to the internet or malicious skills.

This workflow is applicable not only to OpenClaw but also to any AI agent like Claude Code, Antigravity, or anything else that can connect to the InfraNodus MCP server and skills.

1. Generating Insight from Your Data

We can use OpenClaw and AI agents to generate insights from data. You can use the OpenClaw web chat or Telegram chat for that. 

In order to use OpenClaw to generate interesting ideas, you need to add additional data analysis and insight generation capabilities to it. You can, of course, use your LLM for that, but the nature of LLMs is that they give the most probable and generic answers, so the insights you generate this way also will not be original. That's why you need a prompt, a skill, or an MCP server that will provide your LLM with additional steering logic and also access to real-time data. 

Let us demonstrate the difference. First, we just use the standard LLM setup to generate insight from our Obsidian vault (in 5/1) and then show you how you can get better insights using the MCP server (5/2)

1/1 Finding the Main Themes with OpenClaw

The best way to begin is to ask OpenClaw to find all the documents that mention a certain term, for instance, in my case I want to find all the documents that mention "fractal". Simply prompt OpenClaw:

Find me all the Obsidian documents where I mention the term "fractal"

As we have a super protected setup, OpenClaw will sometimes think it doesn't have access to those files, but you can simply tell it that it actually does and it will find them the second time:

You can then ask OpenClaw to tell you what are the recurrent ideas you tend to focus on when you think of this topic. Just prompt it:

What are the typical recurrent ideas I tend to focus on in relation to fractals?

You will get something like this as a response:

This information is super helpful, because it shows you recurrent ideas around the topic of your interest. While you may be aware of the most prominent topics, there might be smaller ones that you forgot about that are less obvious but also more interesting and novel because of that reason.

For instance, I see in this case that I may be missing the part where I'm talking about the aesthetic and artistic implications of fractal dynamics. This report reminds me of that. 

1/2 FInding Gaps In Your Thinking

I can further explore the gaps in my thinking to find some original angles to focus on by adding the prompt:

And what are the gaps in my thinking in this realm?

OpenClaw will then use your LLM (Claude in my case) to find what it considers to be the gaps in my thinking:

For instance, here it's proposing us to think of the failure modes of fractality (what happens if the heart / movement dynamics is not fractal) and also more rigorous research. As you can see, it relates more the "how" rather than "what", which is interesting too but also points to a fundamental limitation of pure LLM workflows when it comes to content. 

2. Using InfraNodus MCP Server to Improve the Quality of Insights in OpenClaw

In order to improve the quality of OpenClaw's output, you can connect it to the InfraNodus MCP server using the mcporter or the InfraNodus OpenClaw skill. This will enable you to 

• have better control over the model's decisions, 
• provide access to real-time search results and search intent data (that can help you optimize your content and writing), 
• use specialized GraphRAG knowledge graph system for expert ontology, persistent cross-project memory, and content retrival
• provide access to your existing InfraNodus graphs

2/1. Installing the InfraNodus Skill / MCP Server to OpenClaw

The easiest way to start using the InfraNodus MCP server is to install the official skill file that contains instructions on using InfraNodus with mcporter which basically turns InfraNodus into a CLI tool (and enables LLMs to invoke it in its calls easily).

The installation instructions can be obtained at https://infranodus.com/skills/cli-use-openclaw

You can basically run the following prompt in OpenClaw:

install this skill into the skills folder: https://github.com/infranodus/skills/releases/download/v1.0.7/skill-cli-use.zip

(alternatively, you can copy and paste the contents of the the actual skill file.

If you receive an error (because of too much security restrictions), you can install the skill manually:

Or also download the skill-cli-use from our official skills repo at https://github.com/infranodus/skills/releases

And then copy the skill and unzip it to your ~/.openclaw/workspace/skills folder (or ~/.openclaw/skills to avoid security restrictions and make that skill available to all the workspaces and projects globally):

cp skill-cli-use.zip ~/.openclaw/workspace/skills/skill-cli-use.zip
unzip ~/.openclaw/workspace/skills/skill-cli-use.zip -d ~/.openclaw/skills/skill-cli-use

SECURITY NOTE #8: CAREFUL WHICH SKILLS TO INSTALL AND WHERE TO INSTALL THEM

Skills is the most dangerous part of OpenClaw. Always review the skill's content to ensure that it doesn't contain any instructions to run malicious terminal commands and also make sure it doesn't send your data to some external URLs that you never heard of (even if the data it sends to the URL is innocent, the response may contain a prompt injection with a malicious command or code). 

Then restart OpenClaw:

openclaw gateway restart

2/2. Set Up the InfraNodus API Key

Note, that InfraNodus will work without the API key, but you will hit the rate limits and won't be able to access your graphs and store / retrieve memories. We recommend to get an API key for better experience.

To add the key, go to the Skills control panel http://127.0.0.1:18789/skills  to add the InfraNodus API key which you can get at https://infranodus.com/api-access

After you add this key, it will automatically be added to the main configuration file of OpenClaw at ~/.openclaw/openclaw.json. It will look something like that:

"skills": {
		"entries": {
			"infranodus": {
				"apiKey": "YOUR_INFRANODUS_API_KEY"
			}
		}
	}

 

IMPORTANT: If you run OpenClaw in a secure sandboxed environment, you will have to manually add this key as an environmental variable to your Docker container. In order to do that, go to the ~/.openclaw/openclaw.json file and then add the following line to the `agents.default.sandbox scope:

"env": {
	"INFRANODUS_API_KEY": "YOUR_INFRANODUS_API_KEY"
}
			

The final setup will look something like:

"sandbox": {
		"mode": "all",
		"workspaceAccess": "rw",
		"scope": "session",
		"docker": {
			"image": "openclaw-sandbox-mcporter:latest",
			"network": "bridge",
			"binds": [
				"/Users/dmt/Software/Second Brain:/mnt/Obsidian/Second Brain:rw",
				"/Users/dmt/Dropbox/Research:/mnt/Dropbox/Research:ro"
			],
	    	"dangerouslyAllowExternalBindSources": true,
			"env": {			
				"INFRANODUS_API_KEY": "YOUR_INFRANODUS_API_KEY"		
			}			
		}
	}

2/3. Revealing the Main Recurrent Topical Clusters with InfraNodus

After the InfraNodus skill is installed and you relaunched OpenClaw and added the InfraNodus API key, you can now invoke InfraNodus to generate better insights with OpenClaw. 

For instance, you can ask OpenClaw to list all the Obsidian documents where you mention "fractal" with the standard prompt like

What are the Obsidian docs where I mention "fractals"?

After OpenClaw lists all the docs, you can ask:

Pull out those documents and then use InfraNodus to find typical recurrent themes in these.

As a result, we will get a list of topics with the influence score:

Compare these to the standard generic LLM response we got in section 5/1 above. The topics obtained using InfraNodus are based on text network analysis. They are much more defined and — more importantly — they are reproducible and are based on a peer-reviewed algorithm, meaning you will always get consistent results.

Additionally, we have influence score for each, so we can directly see that, for instance, the Movement & Fractal Variability topic takes too much attention while the technical topics (Topology, dimension measurement) are much less influential — prompting us to focus on technical details. 

It is also quite interesting to see the separate more defined Breathing Meditation and Multi-Scale Body Sensitivity topics as they are much better defined in this analysis than in 5/1 and they are also the ones that contain the aesthetic aspects.

2/4. Revealing the Content Gaps

InfraNodus MCP has a content gap detection tool. It builds a graph of your data, identifies the topical clusters (as described above) and then finds gaps between these themes in order to identify what is missing and what new connections could be made.

This is how it looks in practice: the topics that are furthest apart could benefit from better connections —

And here's a report for our fractal research:

Interestingly, unlike the standard LLM output that we saw in section 5/2, InfraNodus focuses on the actual gaps between the topics, offering a more granular way to improve this discourse.

The focus is on connecting the math to embodied experience: conducting specific research that would link HRV / DFA parameters to the body states and various movement regimes to understand the relation between those.

2/5. Saving Ideas as Memory / Knowledge Graphs

A really interesting feature of InfraNodus is that it can save your extracted data as a knowledge graph, so you can explore it using the visual interface.

You can ask OpenClaw to

Ok, the data you used to create this report can you save it in infranodus graph please

Then after a few seconds you'll get the InfraNodus link with the graph generated. It will look something like this:

2/6. Transcending the Discourse and Optimizing the Bias

Another really interesting feature of InfraNodus is its ability to transcend the discourse and optimize the structure of the text so that if it's too focused, then it would focus on peripheral ideas and connect them to other interesting topics outside of the realm of the particular discourse we're analyzing.

This is very useful for creative thinking and generating novel ideas, especially in research and writing. 

You can use this prompt to get these results:

Ok, now use infranodus to trancend and optimize this discourse

The output  will look something like this:

We can then save these ideas into the graph by telling OpenClaw

Save these ideas into the openclaw-fractal graph

We will have a graph memory generated in InfraNodus which we can then visually explore or use in OpenClaw later or in any other LLM tool.

The cool thing is that once the graph is saved, InfraNodus provides information about its structural gaps automatically back in the response, so you don't just get a topical summary but also a proposition of what you could focus on further.

3. Connecting Your Ideas to Existing Research

The power of OpenClaw is that it can connect different silos of data and tools to each other. 

In this case, as we're exploring fractal body dynamics, we can ask OpenClaw to search through our other mounted folder (with external research papers) to find some relevant content that would bridge the gap identified above. 

For instance, we can use a prompt like this:

Search the the Research Papers folder and find the papers that can be used to bridge the gap identified

If OpenClaw says it doesn't have access to the folder, you can nudge it again, or manually copy the folder with all the papers to the ~/.openclaw/workspace that it does have access to and try agian.

After a while, OpenClaw will make a list of PDF documents and select the ones that could be useful for your work. 

You can then ask it to set a reminder to explore these ideas using the Apple reminders skill.