Personal Knowledge Management: How to Set Up a PKM System That Thinks With You

Posted Wednesday, April 29, 2026.

Personal knowledge management (PKM) is how you turn the things you read, write, and learn into a body of thinking you can actually use. A good personal knowledge base is not just storage — it surfaces connections between ideas, reveals gaps in your understanding, and lets you reuse what you know across projects.

It is, however, quite difficult to extract insights from a PKM system as it grows bigger and more complex. That's why you can use InfraNodus to add knowledge graph capabilities to your knowledge base. Using advanced network science methods, InfraNodus will reveal the main topical clusters, the most important ideas, and also the structural gaps in your ideas, so you can use the enhanced intelligence of knowledge graphs to get more out of your PKM. InfraNodus can be added either as a standalone tool or as a layer on top of your existing PKM system or as an MCP server that you can connect with your favorite LLM client.

 

It is particularly important to have a good PKM system set up if you're planning to use it as a context for your AI agents. Well-structured PKM helps you focus your AI agents on the relevant pieces of content and avoid hallucinations. Adding the InfraNodus as a layer on top will add observability to your knowledge base and allow you to steer LLM's reasoning using the knowledge graph as an interface.

This guide walks through how to set up a personal knowledge base that goes beyond folders and tags: a graph-based PKM system where notes connect into a visible network and AI can reason with your knowledge directly. We'll cover the two common setup paths, the piece most PKM systems miss, a six-step practical setup, and how to use AI with your knowledge base — with InfraNodus as the graph and AI layer.

 

 

What a Personal Knowledge Base Actually Is

A personal knowledge base is the organized layer of your own thinking — the place where the books you read, articles you save, meeting notes you take, conversations you have, and ideas you generate accumulate into something coherent. Done well, it becomes a "second brain" you can query, develop, and reuse. Done poorly, it becomes a graveyard of folders you never open again.

There are three common ways to organize a personal knowledge base, each with different tradeoffs:

• Folders and files — hierarchical, familiar, but each note lives in only one place. Hard to retrieve across topics.
• Notes with tags and links — flexible, context-rich (Notion, Obsidian, Roam). Notes can belong to many contexts at once via backlinks and tags.
• A knowledge graph — every note and / or concept is a node, every link an edge. The structure of your knowledge becomes visible: clusters of related ideas, central concepts, and gaps between topics.

The first two are about storage. A graph-based PKM is about thinking: it shows you how your ideas relate, which concepts you keep returning to, and where your understanding is incomplete. This is the angle this guide takes — how to build a personal knowledge base that doesn't just hold notes but actively helps you think. We will show you how you can combine all of those approaches to get the best out of each of them.

The Two Ways People Set Up a PKM (and the Tradeoff)

If you research how to set up a PKM system, you'll quickly notice the field splits into two camps: Method-led and Tool-led. In our guide, we will show you how you can combine both approaches to get the best outcomes.

Method-led PKM

Method-led setups start with a philosophy: Zettelkasten (atomic notes that link to each other), PARA (Projects, Areas, Resources, Archives), or Building a Second Brain (BASB) (capture → organize → distill → express). The method tells you how to think about your notes; the tool is interchangeable.

Strength: conceptual clarity, durable across tools.
Weakness: high upfront cost. Many people spend more time learning the system than using it, very strict framework that may not suit everyone.

Tool-led PKM

Tool-led setups start with software: Notion for databases and templates, Obsidian for local-first markdown with backlinks, Roam Research for outliner-style block linking, Logseq for open-source outlining. The tool's affordances shape your workflow.

Strength: low friction, immediate feedback.
Weakness: easy to accumulate notes without ever connecting them. The tool stores; it doesn't think.

The honest truth: most "best PKM tool" articles miss the actual tradeoff. The question isn't Notion vs. Obsidian vs. Roam — it's whether your setup makes connections visible or just stores notes tidily. Which brings us to what most PKM setups miss.

What Most PKM Setups Miss: Connections and Gaps

The whole point of a personal knowledge base is to let ideas talk to each other. Yet most PKM systems — even the popular ones — leave the most important work to you: noticing when two notes are related, recognizing when a topic is becoming a cluster, and spotting the gaps where your knowledge is thin.

A knowledge base with connections beats a knowledge base with neat folders every time. Here's why:

• Folders force one location per note. A note about "feedback loops" might belong under biology, design, and management — but folders make you pick one.
• Tags and backlinks help, but you have to remember to use them. The signal degrades as the base grows.
• Nothing surfaces what's missing. No folder system tells you "you keep writing about X but you've never written about how X relates to Y." That blind spot is exactly where the next interesting idea lives.

A PKM that has a knowledge graph addresses all three. Every note and or concept can become a node, every connection between them is a relation. Run a community-detection algorithm and topical clusters emerge automatically. Apply node ranking algorithm from graph science and you can identify that notes / concepts that are more influential in your particular context. Run a structural gap analysis and you see which clusters of your thinking aren't yet talking to each other — those gaps are your next research questions, your next essay topics, your next breakthroughs.

 

This is the differentiator: a knowledge base that surfaces gaps does the connection-noticing work for you, so you can focus on the thinking.

How to Set Up a Personal Knowledge Base in 6 Steps

The following six-step setup works whether you're starting from scratch or already have hundreds of notes in Notion, Obsidian, or Apple Notes. Each step explains the principle first, then how InfraNodus implements it (and how a tool-led PKM partially does the same).

Step 1: Capture — get everything into one place

The first job of a PKM is to stop ideas from leaking. Pick a single capture surface and use it for everything: articles, quotes, meeting notes, half-formed ideas. The tool matters less than the habit.

Local Storage

If you prefer to store your ideas and documents locally, you can use a tool like Obsidian or Cursor AI. They both work with local folders, so you can simply create a folder and copy all the files there and use it when you need to add new notes and information.

Interestingly, while Cursor AI is an IDE (basically a fancy text editor for coding), it has a one of the best and most powerful AI agents that will beat any Obsidian AI plugin you'd ever find. So it works really well if you want to use AI to interact with your data.

Cloud Storage

If you prefer to store your ideas and documents in the cloud, you can use a tool like Notion or sign up for Obsidian Sync (which are both paid). In the case of Notion, you won't have any local files, but it has much better capabilities for collaboration than Obsidian. In the case of Obsidian, you still will have local access but your files will also sync to the cloud, so you'll have extra backup and also be able to access your data through the phone, which can be very useful when you need to capture an idea or get an insight on the go.

You can also store all your ideas in InfraNodus, where capture happens in plain text. Paste articles, drop in PDFs, import a Markdown vault from Obsidian, or feed in a YouTube URL — the system extracts the underlying concepts automatically. You can also write notes directly in the editor using [[wiki links]] for entities you want to track explicitly. All of your data is stored in the cloud, but you can also export it as markdown files and use it in your favorite PKM.

Recommended: Hybrid Approach

If you are unsure whether to go local or cloud, we'd recommend a hybrid approach:

• Set up Obsidian as your main PKM tool for capturing ideas.
• Use Obsidian Sync to sync your notes to the cloud and to have mobile access to them.
• Add the InfraNodus Obsidian Plugin to get an overview of your knowledge base.
• Use the InfraNodus Desktop App to add content manually or using its multiple import sources.
• Set up Cursor AI so you can use it to interact with your knowledge base.
• Add the InfraNodus Cursor Extension so you can visualize your content and use the graph to steer your LLM's reasoning.
• Add the InfraNodus MCP server to your Cursor AI so you can extract the graphs captured in InfraNodus and save ideas back to your InfraNodus graphs.
• The InfraNodus MCP server also gives your AI advanced capabilities to retrieve structural gaps and missing ideas.

Step 2: Link — connect notes and concepts as you write

One of the most important elements of a PKM system is the ability to connect ideas and concepts to each other. This is what allows you to see the big picture and to find novel connections you haven't thought of before. There are two main ways to do this: manually and automatically. You can also choose to use the LLM Wiki framework popularized by Andrej Karpathy, which creates an additional layer of wiki-links on top of your existing project.

Manual Linking

Manual linking can be useful as it forces you to prioritize ideas, but it can also take a considerable cognitive effort. You can do that by simply writing a note and tagging the concepts, ideas and entities that you consider to be important. For example, when you write a new note, ask: what existing concepts does this connect to? In Obsidian, Notion, Roam, or InfraNodus, this is a [[backlink]] or [[wiki link]]. You can also do this after writing a note: can simply select the concepts you want to be tagged as entities and add the [[ ]] syntax to them. Then when you open the page again, you'll see the pages it links to in the Backlinks panel or on the graph view.

Automatic Linking

Many people don't like to do [[wiki link]] syntax as it takes a lot of time, so you have two ways to solve this problem. One is to use an automated LLM wiki setup that will automatically tag the concepts for you. You can also use an ontology creator skill to have LLM do it for you.

Alternatively, in InfraNodus, you can choose to have every co-occurrence of concepts in your text can act as an edge in the graph automatically — you don't have to remember to link, the structure emerges as you write. So you can simply add InfraNodus to Obsidian or Cursor AI and it will automatically find connections for you. You can also set it to process wiki-links only, in which case if you do use that syntax you will see connections between your tagged entities only. InfraNodus also can use its built-in AI to generate knoweldge graphs for any topic, which can be a useful starting point. You can then import these graphs using the MCP server back to your local vault and develop them further.

LLM Wiki Framework

The LLM Wiki framework popularized by Andrej Karpathy is a way to create an additional layer of wiki-links on top of your existing project. This can be useful if you want to use AI to interact with your data and you want to have a more structured way to link your notes to each other.

If you decide to set it up, we recommend you to separate your main knowledge base from the LLM Wiki. While your main knowledge base can contain information about all the possible projects you're working on as well as personal stuff, your LLM Wiki can be project-based. For instance, I have separate LLM Wikis for my research in fractal dynamics and for my research in finance. LLM Wiki uses a specialized LLM Wiki skill that will automatically create the wiki-links for you based on the content of your notes and organize them into folders like "concepts", "connections", and "questions".

This way you have one knowledge base with the source data and then another, smaller knowledge base for a particular project with a separate wiki that outlines all the relations and the main concepts that relate to that particular project. The LLM Wiki skill will help you extract relevant data from your main project folder as well as other documents and external sources, so you don't have to touch your personal data with AI. When the structure is set up, you can open that LLM Wiki in Cursor AI or Claude Code and start interacting with it to produce insights and content. Your LLM will have a better understanding of the context because it has that additional wiki layer that explains how the concepts relate to each other.

The result is a personal knowledge graph where every concept lives in a network of relationships instead of a single folder.

Step 3: Cluster — let topics emerge

Once you have a few dozen connected notes, topical clusters start to form. These are groups of concepts that frequently appear together — your actual areas of interest, as opposed to the categories you thought you cared about.

If you use InfraNodus (desktop version, Obsidian plugin, or Cursor extension), it will run community detection (Louvain modularity) on your graph and visualize each cluster in a different color. The nodes will also be ranked by their betweenness centrality: the higher it is, the more important is the concept for connecting the different clusters. The largest clusters and nodes with the highest betweenness centrality tell you what your knowledge base is really about. The smallest clusters often hold the most original thinking.

 

Step 4: Surface gaps — find what's missing

This is where a graph-based PKM pulls ahead of every folder, tag, or backlink system. InfraNodus detects structural gaps — pairs of clusters that are large and influential but barely connected to each other. Those are the spaces in your thinking where a single new idea could connect two whole domains.

You can extract the gaps using the Obsidian plugin or the Cursor extension (visually) or using the InfraNodus MCP server inside your favorite LLM pointed at the folder you work in. If you use Notion, you can use it together with their MCP server to gain access to ideas and knowledge in the cloud.

InfraNodus graph view / MCP server also generates AI-powered research questions that bridge the gap, so the next thing you read or write is targeted at expanding your knowledge base in the direction with the most leverage. This turns review from a chore into a generative practice.

Step 5: Ask questions — talk to your knowledge base

A modern PKM should be queryable in natural language. The advantage of having your knowledge base in InfraNodus is that it exposes your graph as an GraphRAG endpoint: this means you can then plug this graph into any LLM client (Claude, ChatGPT, or n8n) and use it as a context or a reasoning expert for your queries.

Connect Claude, Cursor, ChatGPT, or any MCP-compatible client to your graph at https://mcp.infranodus.com and your knowledge base becomes a thinking partner rather than a filing cabinet.

Step 6: Iterate — review and prune

A healthy personal knowledge base is not a strictly growing pile. It needs pruning, restructuring, and periodic review. InfraNodus measures the diversity score of your graph: too focused (all roads lead to a few dominant concepts) and it's biased; too dispersed (many disconnected fragments) and it lacks coherence.

The framework behind this comes from ecological thinking — treating your knowledge base as a living system that goes through phases of growth, saturation, and restructuring — and cognitive variability, which gives you a concrete way to nudge the graph between focus and exploration depending on what you need.

Using AI With Your Personal Knowledge Base

AI personal knowledge management is the next stage of PKM. The shift is simple: instead of chatting with a generic LLM and re-explaining context every time, you let the model read directly from your own knowledge base. However, you need to decide where AI is applied: at the level of capturing and organization, at the level of providing responses, or both.

AI for Capturing and Linking

As we described in the Steps 1 and 2, you can use AI to help capture, ingest, and link your knowledge. We recommend a hybrid approach: adding ideas manually into your favorite PKM and then setting up an additional AI-powered process like LLM Wiki for specific projects. This way you don't pollute your original repository of knowledge with AI-generated content and instead create a separate repo for ideation where you can use AI to ingest, capture, link, and generate content.

AI for Providing Responses

Once you point your LLM client to your knowledge base, there are several ways that you can use it to query this data:

• Using AI's own context window: In this case, every time you ask a question using an AI agent in Claude Code or Cursor AI pointed at your knowledge base folder, it will use its built-in mechanisms (often similar to vector RAG below) to find the most relevant chunks of content in order to provide better responses. Most people will get satisfying results with this approach, however, there are additional methods you can use to ensure the output is more precise and to avoid hallucinations.
• Vector RAG (basic): chunk your notes, embed them, retrieve similar chunks per query. Works for narrow questions, fails for "what's the overview of my thinking on X?"
• Database layer: you can use a database layer like PostgreSQL or Supabase (with an MCP server) to store data and numbers along with the relevant meta-data. This will let you point your LLM to specific data points and generate queries that will filter and segment the data according to your requests.
• GraphRAG (recommended): use a knowledge graph to retrieve not just chunks but relationships — main topics, key concepts, structural overview. InfraNodus produces this automatically from your notes.
• Reasoning ontology (advanced): a hand-shaped graph where concepts are typed and relationships are explicit, used for multi-step reasoning. Useful for domain experts and structured AI reasoning.

For most people, GraphRAG is the sweet spot. Connect any MCP-compatible client (Claude, Cursor, Copilot) to https://mcp.infranodus.com, ask it to generate insights from your project folder, save the graph with the results to your InfraNodus account. Then you can use this graph as an "expert" from any LLM workflow in order to get additional context or — in case of reasoning ontologies - to add additional reasoning logic to your projects.

This InfraNodus graph can also be used via the InfraNodus n8n node: every time you publish a note, an article, or a meeting transcript, an n8n workflow can route it into your knowledge base, run gap analysis, and email you the new questions to investigate.

 

Common PKM Workflows by Use Case

The same six-step setup adapts to very different kinds of work. Here are four patterns we see most often:

Researcher / Academic

Import paper PDFs and reading notes into InfraNodus, extract concept graphs per paper, then merge them into a single domain graph. The cluster view becomes your literature map; the gap view becomes your list of unexplored research questions. Pair with the AI Research Assistant flow for systematic literature analysis.

Writer / Content Creator

Capture every quote, observation, and half-formed idea into the graph. Use cluster detection to surface the themes you keep returning to (those become your articles). Use gap detection to find non-obvious combinations of themes — those become the essays only you can write. See the creative thinking workflow for the writing-side practice.

Founder / Consultant

Feed in customer interviews, market research, and competitor notes. The graph reveals which themes recur across customers (product signal), which themes appear in your notes but not in competitors' positioning (differentiation), and where your strategic thinking is thin. Combine with SEO and content gap analysis for marketing applications.

Learner / Lifelong Student

When learning a new field, the hardest part isn't finding sources — it's knowing what you don't yet know. Capture notes from books, courses, and YouTube transcripts (InfraNodus accepts YouTube URLs directly). The clusters show you what you've already absorbed; the gaps show you what to study next. The journaling and diary workflow is a related daily-practice variant.

 

Frequently Asked Questions

 

Start Building a Personal Knowledge Base That Thinks With You

Stop storing notes you'll never reread. Set up a personal knowledge graph that surfaces connections, reveals what you're missing, and lets AI reason directly with your own knowledge — no folders, no formal ontology, just plain text.

Log In    Sign Up