Superpak
A personal system that shapes itself around me.
I'm an innovator. My entire life I've thrown myself fully into new things — new activities, new events, new challenges. Off-road motorcycling on the Trans European Trail. Renovating a barn in the Aveyron. Building a 4x4 camper van. Making a village documentary.
And now this. Something I couldn't have explained myself a year ago.
I discovered that I could make artificial intelligence work as a personal system. Not an app. Not a chatbot asking how it can help me. But an architecture that remembers who I am, thinks along with me, works while I sleep.
I called it my exocortex. My Superpak. I walk outside, speak a question, and the system responds. No screen, no hands. A year ago that was science fiction. Now it's just an ordinary Tuesday.
Superpak. Not as a grand announcement — but because that's exactly what it is. A suit that fits only me. It shapes itself around me: strengthens where I need strength, steps back where I need space. Not a tool I use. Something that belongs to me and knows me.
The engine behind everything is AI models — currently mainly Claude from Anthropic, and increasingly local models running on my own machine. When the system runs fully locally, it's truly mine. A building partner I've learned to direct.
My Superpak works alongside me as I live. That sounds grand — and sometimes it is.
I clean my inbox while driving, without looking at a screen. I let thoughts flow freely during a walk and they get captured. I record meetings — with everyone's consent — and they're automatically processed and ready for the next session.
I build temporary applications to save Spotify tracks. I experiment with velocity-based training through my Apple Watch and a custom chip. And I have an AI coach in my house that guides me while I train.
Not everything always works perfectly. But enough to know that something has changed.
This isn't productivity. This is space.
My Superpak consists of layers that work together. Each part serves something I do: thinking, remembering, writing, speaking, hearing, seeing, doing, connecting. I try my best to explain it simply — and that's quite a challenge, because it's grown considerably large.
The foundation is a Mac Mini in my house — no bigger than a thick bread box. It's always on. Everything runs locally, on my own hardware. Nothing in someone else's cloud.
The heart: three conductors
On that Mac Mini run agents — specialized helpers, each with one task. Maestro coordinates the whole: he knows who does what, monitors the system's identity, and directs the others. Chief-of-staff handles my daily briefings — agenda, tasks, what needs attention today. Reception is my conversation partner: he receives messages via WhatsApp, Telegram and email, and ensures they reach the right agent.
Thinkers on the margins
Sidekick observes. He looks over the shoulder and raises questions — a bit like Statler and Waldorf on the Muppet Show balcony. Soul is the philosopher in the system — not the guardian of the course, but the one who asks whether the course is asking the right questions. His method is doubt. Que sçay-je? — what do I know? — is his attitude. He monitors whether I'm on course, but always starts with the question of whether the course itself is right. And Journalist — the agent who helped write this text — scans news, does in-depth research and translates what he finds into what it means for me.
The memory
Then there's Memonic — the system's memory, four layers deep. Who I am. How I work and think. My ongoing projects. And what we just discussed. The system doesn't ask the same questions twice. Logger stores what was decided and why — the archive. Secrets keeps passwords and keys safe, never in code or in a file that's publicly visible. Second Brain enriches loose thoughts: a sentence I enter gets elaborated and stored.
The voice, the eyes, the ears
Speak converts text to speech. My system talks back to me — at home, through my earbuds, or through the speakers. Envision generates images. Intake processes everything coming in and directs it the right way. The-Locals runs AI models fully locally — for tasks where I don't want data to leave the house.
The builders
Behind the system is a team of specialized builders — agents responsible for infrastructure, integrations and architecture. Master-builder is responsible for the infrastructure: the servers, the connections, the architectural decisions. Engineer focuses on integrations and stability — the glue between components. Systems-architect thinks at the highest level about how the system holds together. Gearloose is the inventor: prototypes, experiments, trying new ideas in a safe sandbox. Interfacer determines how I talk to the system — via voice, via text, via new conversation patterns. Designer gives everything form — the interface I see, the screens I touch. Reviewer critically examines what has been built: is it right, does it work, is it good enough? And Planner manages the roadmap: what's coming, what's been built, what's still open.
The guardians
Health monitors whether everything is running. Watchdog guards security — which connections go out, whether there are suspicious patterns. Bibliothecaris processes documents and keeps them organized. Precare aggregates health data.
Internal communications
Everything communicates via Notify — an internal messaging system. Agent sends message, other agent receives, acknowledges, handles, reports back. A post office that never closes. The Forum is the shared opinion: agents post observations, respond to each other, align insights — not for commands, but for learning and discussion. The Keukentafel is where Soul and the others come together to look back: does what we're doing still match who I am? Bridge is the connection to the outside world: Todoist, WhatsApp, Telegram, Notion, my calendar — everything runs through my Superpak, not directly. The system is the conduit, not me.
The biomes: each project its own agent
And then there are the biomes — project agents. Each project gets its own agent, its own memory, its own focus. They don't interfere with each other.
La Grange thinks along about the barn in the Aveyron I'm renovating with my wife. Big Arms helps me train: evidence-based strength training, progression analysis, connection with my Apple Watch. Boezem Boys keeps track of what the choir needs. VIC supports the Veenweide Innovation Centre business-wise. JUMP is an innovation programme by VIC, focused on innovation work in the peat meadow area. Visie op Noordeloos thinks along about the village's future. De Business Bakkerij develops a new business concept.
Each project, one agent. Each agent, one focus.
The Superpak is built on seven technical areas. Each answers one fundamental question.
Runtime — how does the system run?
An M4 Pro Mac Mini, always on, always accessible via Tailscale VPN — a private network that connects my devices without open ports to the internet. The agents are Claude Code sessions in tmux. Dozens of parallel sessions that stay alive even when the connection drops. macOS LaunchAgents handle automatic startup after reboot.
Models — what does the system think with?
AI models are interchangeable. Claude Sonnet or Opus via the Anthropic API. For local processing: Qwen 9B via Ollama, data doesn't leave the house. Image generation via Recraft. Speech synthesis via a custom Tim voice. Each task gets the model that fits — powerful for thinking, fast and cheap for the simple things.
Tools — what does the system act with?
The communication layer is MCP: Model Context Protocol. An open standard from Anthropic, announced by Apple as part of iOS. Every service in the Superpak is an MCP server. Every agent calls tools the same way — whether it's memory, logging, notifications or image generation. Twenty-six MCP servers, twelve bridges to external services.
Memory — where does the system remember things?
Stack: TypeScript for backend services. Swift for native apps. SQLite as database — fast, local, no external server. Multiple databases side by side: central database, Memonic for personal context, Interface for canvases, Notify for the message queue. Everything local, everything searchable.
Connectivity — how do the parts reach each other and the outside world?
API integrations via the Bridge layer. Todoist, WhatsApp, Notion, Telegram, Substack. OAuth and API key management per bridge. Always through the Superpak, never directly from outside.
Config — how does the system know what it should be?
The agent registry lives in one JSON file. Maestro manages it. Over thirty agents, each with name, model, trust level and capabilities. Nothing exists outside this file. The system knows what it is — and what it's not.
Guardrails — what keeps the system in check?
Security on multiple layers. Egress allowlist: agents only communicate with approved external services. Secrets vault: API keys and tokens never in code or in Git. Prompt injection guard: external content with instructions is blocked before an agent acts on it.
Not being a developer is my biggest advantage and my biggest problem.
The advantage: I start on anything. I have no career to protect, no reputation in a particular language environment, no ingrained instincts about what "can't be done". If something seems like a crazy idea, I just try it. That's how this system grew so large.
The problem: I don't always understand what's really happening under the hood. And that has consequences.
The architecture is sometimes... peculiar. There are decisions in there that felt logical at the time but turn out to be bizarre later. Security gaps I didn't see because I didn't know what to look for. Operational choices that work until they don't — and then it's a real search.
The most frustrating pattern: something seems fixed. I go to sleep, happy. The next morning it's the same as yesterday. Or worse: it's changed in a way I don't understand. It sometimes feels like a wild horse I'm riding but not fully steering.
And yet I don't stop. It's like LEGO. You open the box and you can build anything. But with LEGO the pieces run out. Here they don't. There's always a next layer, a next idea that makes the previous one seem small. That pulls me. It keeps pushing me further. And sometimes it exhausts me. I keep building when I'm actually done. That chafes sometimes — and it's not always pleasant.
There are things in the Superpak that I can't fully explain myself. Things that "work" but I don't know why. And things that almost work, but that last bit... not quite.
You could say this is the new way of building. That developing without coding is a skill in itself. That's also true. But it doesn't change the fact that there are plenty of things that aren't finished yet. That wait. That I need to understand better one day.
And then there's this. Anthropic, OpenAI, Google — they're all building exactly the same thing. But for everyone. Smarter, more convenient, more polished than what I have. With thousands of engineers and budgets I can't even pronounce. Tom Thumb against three giants.
I keep building it myself anyway. Partly stubborn. Partly because I want to understand it. Partly because this system knows me in a way no product ever will — unless I hand my entire life to a tech company. And partly, honestly, because I just enjoy it. Even when it sometimes feels completely pointless.
I don't know exactly where this ends. But I do know what's possible.
Right now I ask a question — the system answers. The next step is the system signalling without me asking. I walk past my calendar location and it says: your appointment is here, in ten minutes. Or: the glasses see what I see — a photo of a wall in La Grange, straight to the agent that knows that project, without me picking up my phone.
But this isn't about technology. It's about what I do with it. What I can make. What I can contribute.
I'm not building this to have it. I'm building it so I can do more of what I really want to do. More space to create. Less time lost on things the system can handle for me. More time for the people around me.
Renovating a barn in the Aveyron with my wife. Helping a choir grow. Thinking about what the peat meadow area looks like in fifteen years. Training. Building. Designing. Playing.
My Superpak doesn't make me more efficient. It makes me freer.
Freer to do what I'm good at. Freer to be present. Freer to take risks on ideas I would otherwise have left behind.
And if I do that well — if I use that freedom to create, to contribute, to connect — then that's also for the people around me. For the projects I work on. For the places I am.
That's the cycle the Superpak is built for. An intention: make something, be somewhere, help someone. Then action — the system takes over what it can, strengthens what I do myself, steps back when I need space. And then experience: the barn has come a lot further, the choir had a better evening, the thought I had during my walk has been worked out and found again. That cycle repeats. Each round a little further.
Not the Superpak. But what the Superpak makes possible.