PredictWind is the best tool for this and we're not pretending otherwise. But almost everything underneath it is public, and most sailors have no idea how much of it they could be running themselves.
Nothing here is autonomous and nothing here replaces your judgement. An AI can fetch a file, open an app and set up a calculation. It must never be the thing that decides your route or your course to steer.
AI gets things wrong, and it gets stubborn about it. It will give you a confident answer, accept a correction, then drift straight back to the same mistake. Offshore that is not an inconvenience. Every route gets checked by a human against the charts before anybody sails it.
It's simpler than it looks
Two inputs. A forecast, and how fast your boat goes.
Routing software takes a weather forecast laid out on a grid, takes a table of your boat's speed at every wind angle and strength, and then brute-forces the question: from here, where could I be in an hour? Then from each of those places, where in the hour after that? It fans outwards, hour by hour, until some of those paths reach the destination.
The one that gets there first wins. That's isochrone routing, and every router from the free ones to the expensive ones is doing a version of it.
Which means the quality of the answer comes entirely from the quality of those two inputs — the forecast, and your polars.
The forecast, as numbers instead of a picture
GRIB is just a file format — GRIdded Binary. A weather model divides the world into a grid, works out wind, pressure, waves and rain at every grid point for every few hours into the future, and writes that out as numbers.
That's the difference between a GRIB and a weather map. A map is a picture somebody drew for you. A GRIB is the raw data, so your software can compute with it.
They are also small, deliberately. You can ask for just your patch of ocean, just wind, just the next three days, and get a file measured in kilobytes — which matters enormously when your connection is a satellite link or an SSB radio.
GFS is the American global model, run by NOAA, and it is free to everybody. ECMWF is the European one, generally considered better, and historically the expensive one. The paid services largely sell you access to more models, higher resolution, and a nicer way to compare them.
The part nobody tells you
GFS data is public. You are allowed to have it. You do not need a subscription to get a forecast.
Then open them in qtVlm, OpenCPN or Freeboard SK — all free — and you have a routing setup that cost you nothing but the download.
Your boat's speed, as a table
A polar is just: at this true wind angle and this wind speed, this boat does this many knots.
Plot it and you get the familiar lopsided shape. Dead downwind is slow. Close-hauled is slow. Somewhere around a broad reach is where most boats do their best work, and the polar shows exactly where.
Routing software lives or dies on this table. Feed it an optimistic polar and it will route you as though you're a racing boat, promise a four-day passage, and leave you short of a weather window.
Nobody published polars for a forty-year-old heavy-displacement ketch. So you build your own, and the data is already going past you: log true wind angle, true wind speed and boat speed every few seconds on every sail you do, then take the median for each combination.
After a season you have a polar that describes your boat with your sails and your bottom — which is worth far more than the designer's numbers.
And be honest with it. A cruising polar should reflect what you'll actually carry at two in the morning with two people aboard, not what the boat could theoretically do fully crewed in daylight.
A careful line
The useful part is the donkey work. Fetching the right GRIB for the right area, loading it, setting up the calculation, logging sail data to build the polar, comparing a few departure times and summarising what changed. All of that is tedious and all of it is safe to automate.
What it must never do is decide. The route it hands you is a proposal. A human checks it against the charts for depth, hazards and traffic, checks the weather assumptions still hold, and makes the call.
That distinction is the whole thing: AI operates the tools, people make the decisions.
And you should be sharing it
You already measure wind, pressure and temperature. That data is worth something to everybody around you.
Forecast models are only as good as the observations fed into them, and the ocean is enormous and almost empty of sensors. A boat offshore reporting real pressure and real wind is genuinely rare data.
Windy accepts data from personal weather stations, and a boat counts. Exodus sends hers up automatically — the instruments feed Signal K, and a Signal K plugin pushes the observations to Windy. Once it's configured it simply runs.
It costs nothing, it takes an evening to set up, and if you're already running Signal K you have everything you need.
Here's the idea. Ten boats offshore within a few hundred miles of each other, each measuring real conditions, each sharing them. Not a forecast — what is actually happening, right now, where other boats actually are.
A model tells you what it thinks the wind will do. Another boat forty miles upwind tells you what the wind is doing. On passage that second thing is worth more, and the pieces to build it already exist.
That one's on the list.
Straight answer
Yes, and we pay for it.
They run their own high-resolution models, they have the computing power behind them, and for a real passage decision that is what we use. We're not going to pretend a free stack matches it.
But "the best tool costs money" and "you can't do this without paying" are different statements. The forecast data is public. The routing maths is well understood. The software is free. If the budget isn't there, you are not locked out of weather routing — you just have to understand it a bit better.
Exodus