32°N › Vision

01 · The shift

AI changed what software can do. Most of it happened in the last few years — and almost none of it has reached a boat.

Software that used to need a person now reads, reasons, and acts on its own. That’s not a distant prediction — it’s the software most of us already use on land, every day. The shift already happened. The question left is who it reaches, and when.

02 · The gap

Marine software is still running on 1990s architecture.

Walk onto most boats built in the last five years and the electronics underneath are still individual, closed systems from a generation before AI existed — a chartplotter, an autopilot, a monitoring panel, each its own island, each certified against a standard written for a different era. They weren’t designed to share data. They weren’t designed to be reasoned over. Bolting AI onto that later means bolting it onto architecture that was never built to carry it.

That’s the gap, and it splits the fleet in two. Platforms built AI- native from the data layer up get faster and more capable the longer they run. Platforms that retrofit AI onto thirty-year-old foundations spend that same time paying interest on the debt. I don’t think that’s a controversial bet — I think it’s just early.

03 · The promised land

A boat that anticipates, not just reports.

Picture the outcome, not the feature list. Instruments that don’t just display a reading but notice when it’s wrong. A system that watches the weather, the battery bank, and the passage plan together, and tells you what’s about to matter — before it does. Software that works the way good software works everywhere else now: fast, clear, and built around what you’re actually trying to do.

NOT BUILT YET RENTED — NOT OURS RENTED CHART PLOTTER their glass, on the bulkhead OURS WEB ashore OURS PHONE in your hand, on watch ON THE BOAT — THE HUB IS AUTHORITATIVE all logic and all data live here · fully functional with no internet SCENARIOS what the crew actually needs to decide on the boat’s own network AI PROPOSES · A HUMAN DECIDES · NUMBERS COME FROM THE ENGINES, NEVER THE MODEL SCENE RUNTIME · AI-NATIVE APPS the contract · live → stale → absent · capability gating · takeover 2 of 27 catalogued scenes are built ANALYSIS ENGINES pure functions · electrical · tides routing · weather · alarms PERSISTENT STORE boat and world together, over time the history nothing else on a boat keeps trends · reconciliation THE BUS one model · where every value came from · how fresh · which source won what is true now everything is kept THE BOAT’S OWN SYSTEMS NMEA 0183 over UDP — the one live source NMEA 2000 · serial · Victron — no transport yet decode + provenance PERCEPTION — ON A GPU ABOARD camera · radar · later LiDAR no satellite round-trip is survivable here just another source THE CLOUD OPTIONAL. NEVER REQUIRED. WORLD DATA weather · tides · charts COGNITION the copilot · ML weather too heavy for the boat THE RELAY a tunnel to the hub, not a server SYNC · BACKUP only if the owner opts in cover this column and the boat still works when remote
Everything the boat needs sits inside the box. The hub holds all the logic and all the data, and it keeps working when the cloud is unreachable. Two paths lead out of the bus. One goes straight to the engines for what is true right now; the other goes into the store, which is where trends and reconciliation get their history. The cloud column adds reach and heavy compute. Put your hand over it and nothing essential stops. Scenarios surface in three places: someone else’s plotter, which I draw onto rather than build, plus my own web and phone, reached either directly on the boat’s network or through the relay when you are away from her.

A boat that acts, not just alarms

Detection and monitoring are the easy part. The harder, more useful part is a platform that can take the next step itself — when you’ve told it it’s allowed to.

Software that’s a joy to use

Not “acceptable for marine electronics.” Actually good — the kind of interface you’d expect from something built this decade, because it was.

A platform that’s yours

Yours to read, run, change, and extend — not a black box you rent access to, and not one vendor’s roadmap deciding what your own boat is allowed to do.

AI-native, not bolted on. UX built for this decade, not the last three. 100% free and open source — MIT, no exceptions.

DESIGN-TIME INPUTS estimates — rated power, duty cycle, a typical year. No bus. Works offline RUN-TIME INPUTS measurements — actual watts, observed hours, measured irradiance. From the bus THE SAME ENGINE a pure function — it never learns whether a value came from a form or a sensor A RECOMMENDATION “what should I build or change?” A SCENARIO “what is happening, and what do I do?” expectation vs observation RECONCILIATION “why is it not what you expected?” learned values
The design model is an expectation; live data is an observation; comparing them is the third thing neither gives alone. You sized this for three days at anchor and you are getting 1.8. This panel is making 60% of what its neighbours make. Your fridge duty cycle is 0.62, not the 0.4 you estimated. Nobody in this market has the same model at design time and at run time.

04 · Come along for the ride

I’m building this in the open. Come along.

There’s no company behind this and nothing being sold. It’s one person’s R&D project, built in public, tested on a real boat. Every line of code is shared as it’s written, under MIT — read it, run it, change it, fork it, even sell it if you want to. I’d rather you follow the actual work than take my word for where it’s going.