AI workers for Open-Meteo

What is an AI worker for Open-Meteo?

An AI worker for weather reads public forecast, historical, air quality and marine data for a place and answers the question somebody was going to have to answer anyway: whether the crew goes out on Thursday, whether the delivery window holds, whether last month's numbers had a reason. There is no account to connect and no key to paste, because the data is public and the platform reaches the service on your behalf. Nothing of yours travels to the weather service either, since the only thing a lookup contains is the place being asked about.

What these workers do

Decide whether the work goes ahead

The forecast for the site, on the days the job is booked, read against what would actually stop it: rain, wind, gusts, freezing conditions. The answer arrives before the crew is dispatched rather than after.

Explain what the weather did to the numbers

Pulls the actual conditions for a past period and lines them up against demand, footfall, delays or cancellations, so a strange week has an explanation instead of a theory.

Protect an outdoor or sensitive shift

Reads air quality, particulates and pollen for the place people will be working, so a heavy air day is planned around rather than discovered by the people breathing it.

Plan deliveries and routes around the week

Checks conditions across the days and places a route touches, so the schedule is built around the bad afternoon instead of colliding with it.

Give another worker the conditions

Weather becomes context for a job somewhere else, so a briefing, a customer message or a staffing plan is written against the real forecast rather than an assumption about the season.

How an AI worker runs in Open-Meteo

There is nothing to connect

Nothing of yours goes out

Places become coordinates first

The forecast and the record are different things

Weather is an input to somebody else's decision

Times belong to the place

Open-Meteo questions, answered.

Do we need a weather account or an API key?

No, and there is nothing to set up. Weather is public data and the platform reaches the service on your behalf, so this is the rare app where a worker is useful the moment it exists. There is no consent screen, no admin to chase, no credential to store and no connection that can later expire or be revoked by somebody leaving. A kit that uses weather installs ready to run.

Does any of our data go to the weather service?

Only the place and the dates being asked about. A weather lookup is a coordinate and a time range, so there is nothing else in it to send: no customer records, no contacts, no documents, no account identifiers, and nothing that identifies your organisation. This runs in the opposite direction to most integrations, which read your data. Here a worker asks a public question about a public place.

How far ahead can it forecast, and how far back does the record go?

Forecasts run about two weeks out, with the useful detail concentrated in the first few days as forecasts always are, and the daily view sits alongside an hourly one for when the question is about a particular morning. The historical record reaches back to the 1940s, which is long enough to ask whether a season is genuinely unusual rather than merely memorable. The most recent days of the record settle as the observations are processed.

Does it cover the places we operate in?

Coverage is global for forecasts, current conditions and the historical record, so a worker handles sites in different countries in the same job without anything special. Two of the surfaces are narrower by their nature and say so rather than returning a confident blank: wave and swell conditions exist for oceans and seas, so an inland point has no marine answer, and pollen has limited model coverage, so outside covered areas those values come back absent rather than as zero.

Will it warn us about severe weather?

Not as an official alert, and it is worth being clear about the difference. This surface carries measurements and forecasts, not the warnings and watches issued by a national meteorological service, and nothing here should be treated as one. A worker can read that gusts, snowfall or heavy rain are forecast for a place and raise it on a schedule, which is genuinely useful for planning. For life-safety decisions the official warning from the national service is the authority, and it always outranks a reading from here.

What AI workers can do in Open-Meteo

The tools a worker can be granted in Open-Meteo once you connect it. A kit grants only the ones its job needs, and you can narrow them further on the worker.

Find Place Coordinates

Turn a place name or postal code into coordinates. START HERE - every other weather tool takes a latitude/longitude, not a name. Returns up to 20 matches, each with its country, admin area, population and IANA time zone so you can tell two same-named towns apart; pick one and pass its latitude/longitude on. Ambiguous names are common: append a country ('Springfield, US') or set `countryCode` to narrow.

Current Conditions

Read the weather RIGHT NOW at a coordinate - temperature, apparent temperature, humidity, precipitation, cloud cover, wind speed/direction/gusts and a plain-English condition. Use `weather_geocode` first if you only have a place name. For later today or any future day use `weather_forecast` instead. Each series comes back as ROWS, not parallel arrays: every row carries its own `time` with the UTC offset embedded, plus one key per variable. `units` maps each variable to its unit, and a `weather_code` is accompanied by a plain-English `weather_description`.

Weather Forecast

Forecast the weather at a coordinate, up to 16 days ahead - daily highs and lows, precipitation totals and probability, wind, sunrise/sunset, UV. Add `hourly` for intra-day detail and `pastDays` to prepend recent days in the same call. THE tool for planning: outdoor work, deliveries, site visits, events, staffing. Use `weather_geocode` first if you only have a place name. Each series comes back as ROWS, not parallel arrays: every row carries its own `time` with the UTC offset embedded, plus one key per variable. `units` maps each variable to its unit, and a `weather_code` is accompanied by a plain-English `weather_description`.

Historical Weather

Read what the weather ACTUALLY WAS at a coordinate, any day from 1940 to a few days ago. Use it to explain the past - correlate sales or demand with temperature, check whether weather caused a delay, or build a seasonal baseline to compare a forecast against. This is a reanalysis archive, so its variables differ from the forecast's: no precipitation probability, UV, visibility or CAPE, and the soil layers are named for their depths (soil_temperature_0_to_7cm). Asking for a variable it does not carry is rejected with the supported list, never answered with nulls. Windows are bounded: up to about ten years of daily rows, or 31 days when you ask for hourly. Each series comes back as ROWS, not parallel arrays: every row carries its own `time` with the UTC offset embedded, plus one key per variable. `units` maps each variable to its unit, and a `weather_code` is accompanied by a plain-English `weather_description`.

Air Quality and Pollen

Read air quality at a coordinate - US AQI, PM2.5/PM10, ozone, nitrogen dioxide, dust, UV index - plus tree, grass and weed pollen. Use it to advise on outdoor work, HVAC and filtration, or health-sensitive scheduling. Pollen has limited model coverage; outside covered areas those values are null rather than zero, and the response says so in `accessInfo`. Each series comes back as ROWS, not parallel arrays: every row carries its own `time` with the UTC offset embedded, plus one key per variable. `units` maps each variable to its unit, and a `weather_code` is accompanied by a plain-English `weather_description`.

Marine and Wave Conditions

Read wave and swell conditions at a coastal or open-water coordinate - significant wave height, direction and period, the wind-wave and swell components, sea surface temperature and current. For port operations, shipping, coastal works and water-based events. The wave model covers oceans and seas only: an inland point returns all-null values and says so in `accessInfo`. Each series comes back as ROWS, not parallel arrays: every row carries its own `time` with the UTC offset embedded, plus one key per variable. `units` maps each variable to its unit, and a `weather_code` is accompanied by a plain-English `weather_description`.