Weather

Weather from Open-Meteo, which serves national weather models and the ERA5 reanalysis through one key-less API. The script resolves the place, asks the right endpoint in the place's local time zone, and prints a compact table or JSON.

• Forecast: now, then up to 16 days, and hour by hour on request.
• History: any day since 1 January 1940, with a summary: mean highs and lows, total precipitation, wet days, hottest and coldest day. Long ranges also get monthly rows.
• Air quality: European and US AQI with their levels, PM2.5, PM10, ozone, NO2, SO2 and UV, plus pollen in Europe. The daily worst AQI is shown for the coming days.

Every answer starts with the place it resolved to: name, region, country, coordinates, elevation and time zone. When a name is ambiguous, check that line before you answer.

When to use

• "What's the weather in X", "do I need a jacket", "weekend forecast" → the default command.
• "Will it rain this afternoon / tomorrow" → --hours 24 / --hours 48.
• "What was the weather on 12 June 2019", "how hot was last July" → history. For "is this November unusually warm", compare with a 30-year baseline (history PLACE --from 1991-11-01 --to 2020-11-30; its monthly rows give the normal range).
• "Is the air bad today", "can I run outside", "smog", "pollen" → air.

Not for:

• severe-weather warnings or alerts (those come from national weather services);
• climate facts and classifications (Köppen, climate zones) → wiki-lookup;
• climate projections;
• marine, surf or aviation (METAR/TAF) forecasts.

Quick start

The script is scripts/weather.py relative to this skill. It uses only the standard library.

uv run scripts/weather.py Berlin                                   # now + 7 days
uv run scripts/weather.py "Paris, Texas" --days 3 --hourly --hours 12
uv run scripts/weather.py 40.71,-74.01 --units imperial
uv run scripts/weather.py "Eiffel Tower" --days 1                  # landmarks go through OpenStreetMap
uv run scripts/weather.py history Moscow --from 2024-07-01 --to 2024-07-31
uv run scripts/weather.py air Delhi

Output of the first command (shortened):

# Berlin, State of Berlin, Germany · 52.53°N 13.42°E · 46 m · Europe/Berlin (GMT+1)
now (20:00): 8.6°C, feels 6.8°C · Light drizzle · wind 8 km/h WNW (gusts 18) · humidity 95% · 1032 hPa

date       weather     min/max °C  feels  precip mm (chance)  wind/gusts km/h  UV   sun
Fri 10-30  Light rain  7 / 9       5 / 8  0.0 (0%)            10/22 WNW        2.1  06:58–16:40
Sat 10-31  Light rain  6 / 11      4 / 9  0.0 (3%)            11/27 NNW        2.1  07:00–16:38

model: best_match (Open-Meteo's regional blend of national models) · Weather data by Open-Meteo.com (CC BY 4.0) · free tier for non-commercial use

Places

• Name: "Berlin". The most prominent match is taken. When another place of that name is comparable in size, a note: lists the others.
• Name with region or country: "Paris, Texas", "Springfield, Illinois", "Valencia, Venezuela". Only matches whose region or country fits are kept. --country US does the same with a code.
• Addresses, landmarks, districts that the place index lacks fall back to OpenStreetMap Nominatim, and the output credits it.
• Region and country codes: US state codes (Salem, IN is Indiana), Canadian province codes (Victoria, BC), and country codes (Paris, FR, London, UK).
• Coordinates: 52.52,13.41 or -33.87,151.21, latitude first, in decimal degrees. Decimal commas also work: 52,52 13,41.
• --lang de gives place names in another language. Weather descriptions stay in English.

The output shows the model's grid point. It can be a few kilometres from the place, and its elevation can differ in the mountains.

Options

JSON output

Shapes, values shortened:

// forecast
{"kind": "forecast",
 "place": {"name": "Berlin", "admin1": "State of Berlin", "country": "Germany", "country_code": "DE",
           "latitude": 52.52, "longitude": 13.41, "grid_latitude": 52.53, "grid_longitude": 13.42,
           "elevation": 46.0, "timezone": "Europe/Berlin", "timezone_abbreviation": "GMT+1",
           "source": "Open-Meteo geocoding (GeoNames)", "alternatives": ["Berlin, New Hampshire, United States"]},
 "units": {"temperature_2m": "°C", "wind_speed_10m": "km/h", "precipitation": "mm", ...},
 "model": "best_match",
 "current": {"time": "2026-10-30T20:00", "temperature_2m": 8.6, "apparent_temperature": 6.8, "weather_code": 51,
             "weather": "Light drizzle", "wind_speed_10m": 8.0, "wind_from": "WNW", "wind_gusts_10m": 18.0,
             "relative_humidity_2m": 95, "precipitation": 0.1, "cloud_cover": 100, "pressure_msl": 1032.0, "is_day": 0, ...},
 "daily": [{"date": "2026-10-30", "weather": "Light rain", "temperature_2m_min": 7.0, "temperature_2m_max": 9.0,
            "precipitation_sum": 0.0, "precipitation_probability_max": 0, "wind_speed_10m_max": 10.0, "wind_from": "WNW",
            "apparent_temperature_min": 4.0, "apparent_temperature_max": 7.0, "wind_gusts_10m_max": 30.0,
            "uv_index_max": 2.1, "sunrise": "2026-10-30T06:58", "sunset": "2026-10-30T16:40",
            "daylight_duration": 35000.0, ...}],
 "hourly": [{"time": "2026-10-30T21:00", "temperature_2m": 8.4, "precipitation_probability": 10, "weather": "...", ...}],
 "attribution": "Weather data by Open-Meteo.com (CC BY 4.0)"}
// history
{"kind": "history", "place", "units", "from": "2024-07-01", "to": "2024-07-31",
 "summary": {"days": 31, "days_with_data": 31, "mean_temperature", "mean_max", "mean_min",
             "hottest": {"date", "temperature_2m_max"}, "coldest": {"date", "temperature_2m_min"},
             "precipitation_total", "wet_days"},
 "daily": [{"date", "weather", "temperature_2m_min", "temperature_2m_max", "temperature_2m_mean",
            "precipitation_sum", "snowfall_sum", "wind_speed_10m_max", "sunshine_hours"}],
 "hourly": [{"time", "temperature_2m", "apparent_temperature", "precipitation", "weather", ...}],   // with --hourly
 "monthly": [{"month": "2024-07", "mean_max", "mean_min", "precipitation_total", "wet_days"}],   // ranges over 62 days
 "note": "the last days are preliminary ...",                                                  // ranges ending in the last 5 days
 "dataset", "attribution"}
// air
{"kind": "air", "place", "units",
 "current": {"time", "european_aqi": 56, "european_aqi_level": "moderate", "us_aqi": 62, "us_aqi_level": "moderate",
             "pm2_5", "pm10", "ozone", "nitrogen_dioxide", "sulphur_dioxide", "carbon_monoxide", "uv_index"},
 "pollen": {"birch": 3.0, "grass": 12.0} | null,
 "daily": [{"date", "european_aqi_max", "european_aqi_level", "us_aqi_max", "us_aqi_level", "pm2_5_max"}],
 "model", "attribution"}

All times are the place's local time. Units follow --units and are listed in units. A wet day has at least 1 mm (0.04 inch). In the table, the sun column says "polar night" or "polar day" when the sun does not rise or set.

Workflow for the agent

1. Use the place the user means. If the note: lists alternatives, or the header names a different country than the conversation suggests, run again with "Name, Region" or --country. Mention the resolved place in your answer ("Paris, Texas").
2. Answer the question, not the table.
   • "Umbrella?": the precipitation chance and amount for the hours that matter. Use --hours 48 when the question is about tomorrow.
   • "What to wear?": the feels-like temperatures (the daily "feels" column), the gusts and the rain.
   • Give numbers with units, in the user's convention: --units imperial for US users.
3. Forecasts are model output. Beyond about 7 days they are rough trends. Say so when giving days 8–16. best_match blends national models by region; for a second opinion, use --model ecmwf_ifs025 or --model gfs_seamless.
4. History is reanalysis, not a station record. ERA5 is a 25 km grid (ERA5-Land 0.1°, about 11 km). It smooths extremes and local effects such as cities, coasts and valleys. For official records, point to the national weather service. The last 5 days are preliminary.
5. Air quality is a CAMS model forecast, not a sensor reading. Pollen is only available in Europe, in season.
6. Credit Open-Meteo ("weather data by Open-Meteo.com") when you publish the numbers. The free tier is for non-commercial use; a commercial product needs an Open-Meteo API key (OPEN_METEO_API_KEY).
7. On exit 4, stop. The free-tier limits are in the compatibility line above. Don't loop over many places or dates: one history call covers a whole range.

Exit codes

One error: line on stderr with a hint after —.

Limitations

• No warnings or alerts. Severe-weather warnings come from national services (DWD, Météo-France, NWS…), not from these models.
• The model grid is coarse. Point forecasts in mountains, on coasts and in cities can be off by several degrees.
• History before 1950 is less certain. ERA5 assimilates fewer observations for that period.
• Pollen covers Europe only. Air-quality resolution is about 45 km globally and 11 km in Europe.
• Weather descriptions are English. They come from the WMO weather code; --lang changes only place names.
• Licences. Open-Meteo data is CC BY 4.0; OpenStreetMap place results are ODbL. The output credits both.

Chaining

• weather history + pdf-typeset → a one-page climate summary for a trip or report.
• wiki-lookup entity CITY --props P625 gives coordinates for a place whose name is ambiguous.