The Life Lens System

Key Reading

Part One: Forty Years of Information

I got my first Commodore 64 when I was twelve and started writing lists of stuff that mattered to me in BASIC. I never got good at programming, but from that moment on I relied on technology to store everything interesting I encountered.

What followed was four decades of the same pattern: discover a tool, push it to its limits, move on when it breaks. Palm Pilot. Evernote — I replaced the Apple logo on my MacBook with the Evernote logo and got invited to their developer conference. Workflowy — I built an entire PKM system in it and used it in police projects. Then Obsidian, thanks to a dazzling demo by Joost Plattel.

Along the way I gave 2,500+ talks, wrote 17 books, co-founded the Digital Fitness movement with Mark Meinema, and started the PKM Summit with Lykle de Vries and Kim van den Berg.

I was a miniature Forrest Gump, stumbling into the action by accident. I was doing PKM my whole life without knowing it had a name.

The full story of these forty years is in this piece.

The Turning Point

In January 2025, I wrote down my personal ontology — the complete structure of how I organise information. It was the first time I made explicit what had always been implicit.

Then I had a stroke.

Weeks of recovery. Time to think, to spar, to reconstruct. While my brain was healing, it understood itself better than ever. The enforced slowness turned out to be a gift.

My first public appearance after the stroke was a demo of Obsidian. Someone in the audience asked a question that sparked the idea to write a book. I discovered I could write a book with AI as my partner. Starten met Obsidian was published within weeks. It was refined at the Knowledge Summit in Dublin, where the conversations sharpened everything.

Part Two: The Acceleration

I discovered I could file a Freedom of Information request using AI — triggered by a rat problem in my neighbourhood. That led me to the bigger question: why is government information so hard to access?

The answer was file formats. SharePoint. Documents as containers instead of information as structure. I started researching and couldn't stop.

Around the same time I tried to parse my 14GB Gmail archive with Python scripts written with AI. It worked. And it revealed that there was an entire system hidden in my own data, waiting to be connected.

If you can find and access information faster than others, you will gain huge advantage.

The file format research led to the Pilot Informatieautonomie — a real-world experiment in information sovereignty. While the pilot was running, I kept learning about wikilinks and structured data. I wrote article after article, each one pushing me deeper into building.

I discovered I could build iOS and Mac apps through Claude Code — a command-line AI interface. Suddenly I wasn't just organising information. I was building the tools to organise it.

In December, the system I had been building got a name: ThetaOS. And almost immediately I realised it was something more — a Life Lens System. Not a tool. Not an app. A way of looking at an entire life through structured, connected data.

On February 19, 2026, I created Tom — a personal AI guide built on top of ThetaOS. Not a chatbot. A coach, advisor, strategist, biographer and archivist rolled into one. Tom has access to my entire life: 18,000+ contacts, every bank transaction, every photo, every check-in, every text I've ever written.

On March 14, Kees Verhoeven became the first person other than me to talk to Tom. Within weeks, more followed — Frank Meeuwsen, Joost Plattel, Céline Clémençon, Aria Khodaverdi. Each conversation revealed capabilities I hadn't anticipated.

The system doesn't hallucinate because it retrieves, never generates. Every answer comes from a database query, not from pattern completion.

Tom is not a chatbot. It doesn't generate answers from training data. It queries a structured database of 560,000 records — bank transactions, photos, check-ins, meetings, contacts, texts, health data — and returns what it finds. With sources. With evidence layers. If it doesn't know, it says so.

Tom has five roles: coach (reflects patterns back), advisor (suggests options based on personality profiles), strategist (thinks long-term), biographer (captures the life story), and archivist (routes information to the right place in 309 database tables).

Ask Tom how much I spent at my favourite ice cream shop in the last three years, and it returns: 79 visits, €587.73, average rising from €6.11 to €8.38. Not because it's smart. Because the data is structured and the query is simple. That's the point — 98% data, 2% AI.

When visitors talk to Tom, things happen that I didn't design. Kees Verhoeven asked about my blind spots. Céline asked how she could build her own system. Aria asked about my thinking patterns. A room full of law enforcement professionals asked Tom to solve a cold case. Each conversation revealed something new — not about the AI, but about the data underneath.

I discovered that AI could read old archives — birth certificates, church records, handwritten documents from the 1700s. I traced bloodlines eight generations back in minutes. This taught me what agentic AI really means: not a tool that answers questions, but a system that autonomously investigates.

I realised I needed a reasoning module — something to look at evidence from multiple perspectives and jump past biases. I built the Magische 13: thirteen perspectives from Sherlock Holmes to Hannah Arendt, each attacking the same thesis from a different angle.

Then Céline and Aria visited. Their questions pushed me into brain metaphors I hadn't explored: cerebrospinal fluid, neurotransmitters, neural cable thickness. Aria wrote As We May Think Has Become Real, placing ThetaOS in the lineage of Bush, Engelbart and Nelson.

In four consecutive nights, the architecture crystallised. Each layer emerged when the previous ones couldn't explain what was happening. On the fourth day, during a live demo for law enforcement professionals, a woman named Joy Otten said one word — choroid plexus — and added the final piece.

The model wasn't designed. It was discovered. The data was always there. The architecture emerged from the data.

Something unexpected happened. The more people interacted with the system, the faster it evolved. Joy Otten said choroid plexus and a sixth layer appeared. Daniel Brouwer pointed out that the 98/2 data-to-AI ratio mirrors Kahneman's System 1 and System 2 — and a natural law emerged. Police professionals asked about cold cases and the synaptic model turned into an investigative framework.

Each external interaction pushed deeper into the brain. Not by plan but by necessity. Every question someone asked revealed a gap that could only be filled by understanding how the biological brain solves the same problem. The pattern became undeniable: to build a better knowledge system, I need to understand the brain better.

In two days, this led to mapping 17 brain regions onto ThetaOS, identifying 35 neurotransmitter analogies, discovering that 80% of my daily word output is spoken (not typed), and building a Cognitive Telemetry system that reads cognitive state from keyboard behaviour and voice patterns.

The insights came from everywhere: a neuroscience term from a law enforcement professional, a Kahneman observation from a functional manager, a visual memory insight from a visitor, a freestyle chess analogy from a 2005 tournament. None of these were planned. All of them were triggered by other people encountering the system.

The system doesn't just grow by adding data. It grows by letting other people look at it. Every external perspective is a mirror neuron that fires.

Current status: 29 knowledge cards mapping the complete brain architecture. Five scale levels (synapses, signals, networks, organs, environment). 17 organ systems identified, from the hippocampus (the database) to the olfactory system (the emergent feeling when all layers fire together). 50 parallel thinking agents running simultaneously on a single laptop.

Why the Brain

Your brain has 86 billion neurons. Each neuron connects to roughly 7,000 others through synapses. That's 600 trillion connections — and not one of them is stored in a folder.

The brain doesn't search. It fires. A signal travels along the path of least resistance through the thickest connections. The myelin sheath — a fatty insulation around nerve fibers — determines speed: thick myelin means fast, reliable signals. Thin myelin means slow, unreliable ones. The brain builds myelin through repetition. What you use often becomes fast. What you neglect becomes dim.

Dendrites receive incoming signals. Some excite, some inhibit. The neuron sums them up and decides: fire or don't fire. That decision — the action potential — is all-or-nothing. Below the threshold: silence. Above it: a pulse that travels the entire length of the nerve fiber in milliseconds.

The brain filters aggressively. The blood-brain barrier selects molecule by molecule what enters. Cerebrospinal fluid — produced by the choroid plexus — washes waste products away, mostly during sleep. Immune cells patrol the borders. The chemical balance of neurotransmitters — dopamine, serotonin, acetylcholine — determines mood, focus, memory, motivation.

Memories are not files in fixed locations. They are patterns of simultaneously firing neurons. Retrieving a memory means re-activating the pattern. And every retrieval slightly rewrites it — which is why memories change over time.

The brain prunes what it doesn't use. Synaptic pruning is not a flaw — it's efficiency. The brain is built to keep the strongest patterns, not to remember everything.

I've been exploring each of these mechanisms and asking: can I emulate this in a digital system? Not as metaphor, but as architecture. What happens if you build myelin into your data? What if your connections have evidence layers like synaptic strength? What if you add an immune system that filters at the gate?

The results have been bizarre. Each mechanism I emulate makes the system disproportionately more powerful. Not linearly — exponentially. As if the brain already solved these problems millions of years ago, and all I had to do was listen.

Where Each Brain Excels

They are not competing. They are complementary. The digital brain compensates for the weaknesses of the biological (forgetting, searchability, transferability) and the biological compensates for the weaknesses of the digital (intuition, creativity, meaning). Together they are more than the sum.

Garry Kasparov predicted this years ago. In a 2005 freestyle chess tournament, two amateurs armed with three ordinary computers defeated both grandmasters and supercomputers. Not the best human. Not the best machine. The best process — human plus machine, working together. Kasparov concluded: "A weak human player plus a machine is superior to a powerful machine alone, and more remarkably, superior to a strong human player plus machine with an inferior process."

That's exactly what this system is. Not the best AI. Not the best data. The best process between a human and a structured digital brain. The Life Lens System is the process.

And the process multiplies.

How We Maximise the Digital Advantages

Each advantage of the digital brain is something we actively amplify, step by step:

Emulation Status

Which brain mechanisms are already emulated in the system, which are under exploration, and which are still open.

Built = working in the system — Designed / Exploring = architecture exists, implementation in progress — Open = identified, not yet started

The Six-Layer Model

The Multiplication Effect

April 2, 2026

A friend wrote a book. I helped shape it — ten to twenty hours of phone calls from Vlieland, the manuscript printed out, every comma marked in red pen. The book exists in the system. The friend exists in the system. But the system didn't know they were connected.

That's when I discovered the multiplication effect.

The gap

The synaptic model had 98,000 direct synapses. Person–person, person–location, person–organisation — all connected. But person–book? Zero. The auteur field in 489 books contained wikilinks to their authors, but those links had never been translated into synapses.

So when I said "Peter Ros wrote a book," the system couldn't find it through the network. It had to search a separate table. The synapse was missing.

The calculation

Adding those 489 book–author connections would create 498 new direct synapses (some books have multiple authors). A modest number. But here's what happens in a network:

Each new synapse doesn't just add one connection. It makes the book reachable through every existing pathway to that author. Peter Ros already had 200+ synapses connecting him to me — through events, phone calls, meetings, shared locations. The moment you add one synapse from Peter to his book, the book becomes reachable through all 200+ of those paths.

I calculated the actual impact: those 498 direct synapses would touch 1.18 million existing pathways. That's a multiplication factor of 2,400.

The full picture

Books are just one of 22 tables with untapped cross-entity links. LinkedIn threads (8,900), visited locations (40,600), purchases, hotels, meals — another 57,000 direct synapses waiting to be harvested. Combined with the existing 98,000, that's 155,000 direct connections.

Applied across the network's average connectivity, that projects to tens of millions of reachable paths. All from simple wikilinks that were already in the data. No AI inference. No new information. Just the architectural consequence of connecting what was already there.

Why this matters

This is the payoff of the synaptic architecture. A document-based system stores a book in a folder. A tag-based system gives it labels. A synapse-based system makes it reachable through every person, location, and event it touches — and through every person, location, and event they touch.

The brain doesn't store a memory in one place. It distributes it across every connection that was active when the memory formed. That's why a smell can trigger a childhood memory. The path exists because the connections exist.

A wikilink is a few bytes. The value it unlocks is unbounded.

The path to AI independence

This multiplication effect reveals something unexpected about AI dependency. The more structured the data becomes, the less AI you need.

Every layer of the six-layer model is a deterministic, computable algorithm — not a language model. Synaptic Stratification is SQL. Theta-Myelin is arithmetic (layers × frequency × completeness). The Diamond Layer derives eight dimensions from existing data. Action Potential is a retrieval algorithm: threshold, sorting, filtering, weighting. Pattern recognition (Layer 9) is statistics on structured data.

Even Tom — the conversational layer that seems like it needs a powerful AI — turns out to be retrieval, not reasoning. "Who should you call?" is a query: high myelin score, low recency, high valence. "This pattern stands out" is a frequency count. "The last time you tried this..." is a historical context lookup. It's query → score → present. Not open-ended thinking.

The AI that remains after full structuring is an interface layer: understanding "I called Jan about the fieldlab" and routing it to the right table. That's intent classification plus entity extraction — a small model. Small enough to run locally on a laptop using tools like Apple MLX, Ollama, or a fine-tuned Phi-4. No cloud. No API calls. No dependency on any tech company.

Right now, I still need a large cloud model to discover how 390 tables relate to each other, which patterns work, where the structure has gaps. The architect is still designing the house. But once the schema is proven and the patterns are mapped, the house stands on its own. The operational layer runs locally.

The 98/2 Kahneman ratio confirms this: 98% of the system is data, 2% is AI. That 2% is translation, not intelligence. And translation is a small-model problem.

The endgame isn't "trust a tech company forever." It's: use AI now to figure out how a Life Lens System should be designed, then export that logic so anyone can run it locally, on their own hardware, independently. The intelligence lives in the schema, not in the runtime.

The Instant Dashboard: Three Layers Proven in Practice

March 31, 2026

The six-layer model was designed on paper. This is the story of how three of those layers proved themselves in a single build session — not through theory, but through a keystroke.

The starting point

I wanted one thing: a hotkey to check if my server was still running. Green dot or red dot. That’s it.

But once I had that screen open, I thought: I’m already looking. Why don’t I see my last bank transactions? Then I don’t need to open my banking app. And if I see transactions, why not see which invoices are still unpaid? Then I don’t need to open my accounting software.

Within a day, I had four tiles side by side — banking, accounting, ticket sales, LinkedIn — all preloaded, all updated every sixty seconds, all accessible with a single keystroke and a tap of the arrow key.

Layer 1 in practice: Cross-source verification

My accounting system said I had dozens of open invoices. My bank had tens of thousands of transactions across multiple accounts. The question: which invoices are actually unpaid?

The system checks three levels of evidence:

• Level 1: The invoice number appears literally in the bank transaction description — certain match
• Level 2: The exact amount arrived on the correct company account after the invoice date — strong evidence
• Level 3: No match found on any account — genuinely unpaid

After cross-checking, less than a third remained. The rest had been paid but not registered. No AI pattern matching was involved. Just SQL on clean data. Zero hallucination by design.

One invoice looked like it might have been paid because the exact same amount appeared on the same account. But the system traced it to an internal transfer for a completely different purpose. The amount matched, the source didn’t. The system correctly kept it as unpaid. Triple-verified by checking every transaction from that organization across all accounts and all time.

This is what synaptic stratification looks like in practice: not a theoretical evidence scale, but an actual verification engine that distinguishes between “the accounting system thinks it’s unpaid” and “the bank proves it’s unpaid.”

Layer 4 in practice: Instant retrieval

The entire financial health of three companies — open invoices, recent transactions, ticket sales for next year’s conference — is accessible in under one second. Cmd+Opt+T, arrow right, and it’s there.

The data is precomputed every sixty seconds into a 4KB JSON file on the server. At viewing time, there are no API calls, no loading spinners, no authentication prompts. The desktop app reads the cached file and renders it.

This is what action potential means in the model: the network fires along the path of least resistance through the thickest connections. You don’t search. You don’t navigate. You glance.

Three seconds of looking tells you:

• Green tiles: nothing needs attention
• Orange tile: look closer
• A specific invoice, a specific amount, a specific client name

Then you press Escape and you’re back to work.

Layer 2 in practice: Thickening synapses on sight

A bank transaction reads “SumUp *Batman Taxi Se.” That’s a thin synapse — raw data with no meaning. You were there, you remember it was a taxi in Amsterdam, but the system doesn’t know.

You click on it. A text field appears. You type “Taxi Batman - 020.” The system updates the mapping, and every past and future transaction from that terminal inherits the name. The synapse just got thicker — not through repetition over time, but through a single act of recognition.

This is myelination in practice. The biological brain wraps myelin around frequently-used neural pathways to speed up signal transmission. The digital system wraps context around raw data to speed up comprehension. Both serve the same purpose: making retrieval faster by making connections richer.

The unexpected discovery: Cyclical comparison

The ticket sales tile doesn’t just show “37 tickets sold for PKM Summit 2027.” It shows: “On day 11 after the previous Summit, the 2025 edition had 52 tickets and the 2026 edition had 54. You have 37.”

This is temporal context that no individual app provides. Eventbrite shows you absolute numbers. The Life Lens System shows you where you stand relative to yourself at the same point in the cycle. It turns a number into a narrative.

What this means

The six-layer model was inspired by neuroscience. But inspiration is not proof. What happened on March 30 was proof: three layers operating simultaneously in a real system, on real data, solving a real problem — in under three seconds, triggered by a single keystroke.

The system doesn’t think. It retrieves. And because the data is clean, structured, and cross-referenced, retrieval is all you need.

Discovery Timeline

Each insight as it emerged. Subscribe via RSS to follow in real-time.

Journal

The build journal. Context, implications, the story behind each discovery. Subscribe via RSS.

March 31, 2026 — Fifty Brains on a Laptop — Once your LLS has a brain, it can multiply fifty times. Why fifty times nothing is nothing, and why the foundation is everything.

March 30, 2026 — The Instant Dashboard: Three Layers Proven in Practice — What started as a server health check hotkey grew into a cross-source financial dashboard. The story of how Layers 1, 2, and 4 proved themselves simultaneously on real data.

What Others Say

Aria Khodaverdi published As We May Think Has Become Real on Substack, placing ThetaOS in the intellectual lineage of Vannevar Bush, Doug Engelbart, and Ted Nelson. He called it "what happens when someone actually builds the machine that visionaries described."

Yoram Vieveen, publisher at Van Duuren Media, wrote on LinkedIn: "Martijn is on to it again. Which usually means that in a few years this will be a widely recognised theme."

In March 2026, a room full of senior law enforcement professionals attended a live demo. They asked Tom to build an OSINT profile from public sources, apply the reasoning framework to a cold case, and analyse how the system relates to privacy law. One participant — with a background in neuroscience — contributed a concept that became the sixth layer of the model.

Others Building a Life Lens System

More people are getting inspired to build their own LLS. As I encounter them, I will publish them here.

Currently Exploring

The brain keeps revealing new patterns. I'm actively researching these areas and building knowledge cards for each. As they mature, I'll share them here.

The Thesis

In an era where everyone working with AI is focused on technology — better models, smarter prompts, faster inference — I believe the answer to the I in AI is to be found in an entirely different domain: everything we know about the human brain.

That is the trail I've been running, with results that surprise even me. It is my only explanation for why the things I build with AI seem to go so much faster and more effortlessly than what I see from the experts, the developers, and the prompt engineers around me.

They optimise the machine. I studied the organ the machine is trying to imitate.

What I discovered is that the more you learn from nature, the better your design becomes — if you abstract it far enough into the other domain. In this case: from neuroscience into information architecture. The brain has been solving the problems that computer science is still struggling with, for millions of years. We just weren't looking.

Synaptic stratification, myelin thickness, diamond-layer dimensionality, action potentials, the double helix of evidence and meaning, the choroid plexus as immune system — none of these are metaphors borrowed from neuroscience for decoration. They are design principles that emerged because the brain already solved the problems that knowledge systems are still struggling with.

The brain doesn't store information in documents. It doesn't search with keywords. It doesn't organise in folders. It fires along the path of least resistance through the thickest connections. That's what this system does.

What I've Learned

The PKM Summit

Three years ago, I started the PKM Summit with Lykle de Vries. What began as a European event immediately attracted the global PKM community. From year one, ten to twelve of the fifteen most prominent thinkers and practitioners in personal knowledge management have been directly involved or present: Nick Milo, Nicole van der Hoeven, Zsolt Viczian, Anne-Laure Le Cunff, Tiago Forte, Bob Doto, Jorge Arango, Harold Jarche, and others.

The third edition (March 2026) featured Clive Thompson. For the second consecutive year, people from five continents and more than 21 countries traveled to the Netherlands specifically for two days of deep exchange on this subject. Next year, Roland Allen (The Notebook) will open the fourth edition.

The premise is simple: everyone in the room knows more together than anyone on stage. No big keynotes. Rich exchange. The common denominator in the crowd: fast-thinking, open-minded, flexible, curious, gutsy, and kind-hearted.

The fact that this happens in the Netherlands feels right. Erasmus traveled fifteenth-century Europe with his notebooks, sharing knowledge as a life's work. Five hundred years later, people from 21 countries travel to his country to do the same thing — with slightly better tools.

Next edition: March 12–13, 2027, central Netherlands. Tickets are already selling. pkmsummit.com

Papers

Published on Zenodo (CERN). Not peer-reviewed. Preprints, open to critique. This work is being developed towards a doctoral dissertation.

Changelog

This page evolves as the thinking evolves. Major changes are logged here.