Where data lives

There is a simple idea behind this connector that took us a while to appreciate fully. The assistant stays the assistant. The data stays where the data already lives. A protocol sits between them, and the protocol is boring on purpose. That boringness is the feature.

Forecasts. Marine conditions come from sources that already exist for navigators. Wind, wave height, period, direction, on a grid that resolves coastal effects at a usable scale. We did not pick them because they are fashionable. We picked them because the variables a sailor reads at the chart table — wind at ten meters, swell direction, peak period — are the variables they return, in units we do not have to apologize for.

Tides. From real measurements at real stations, where stations exist. A tide is a measurement at a place, not an average over a region. If there is no observation near you, the assistant should say so, and it does. Pretending otherwise is the kind of small, well-meaning lie that compounds at sea.

Places. Marinas, anchorages, fuel stations, bays, drawn from open data with provenance you can trace, plus a curated layer we maintain. Open licensing is the only kind of data we are willing to put behind a sentence the assistant will say with confidence.

Routing. A sea network that respects coastline geometry. A leg is computed against real distances and real obstacles, not narrated. The route the assistant returns is the route the network actually permits.

Why the protocol. Because we did not want to build a new assistant. The one you already use understands you, holds your context, speaks your language. What it lacks is reach. The protocol is the reach. It lets the model call out for the forecast, the tide, the place, the route, and lets each of those answers come from a system that was built to be right about that one thing.

None of this is glamorous. It is plumbing. But plumbing is where the truth of a system lives. The protocol does not make the assistant smarter. It makes the assistant reachable to things that are already true.

A short distance between a question and a fact. That is what we wanted to build.