Forecasts your agents
can verify.
An agent can request a Weatherkind forecast and the trailing 30-day error record for the same coordinates. That gives it measured accuracy data to include with the forecast instead of an unsupported claim.
A small tool definition
The API accepts typed JSON with named sections and variables, ISO-8601 dates, and stable field names. A basic forecast tool definition is short:
{
"name": "get_weather",
"description": "Forecast for a coordinate. Accuracy data is available from POST /v1/accuracy for the same coordinates.",
"url": "https://api.weatherkind.com/v1/forecast",
"method": "POST",
"body": {
"location": { "latitude": 39.19, "longitude": -106.82 },
"units": "us",
"current": { "variables": ["temperature", "condition", "wind_speed"] }
}
}The endpoints
The same endpoint list is available in /llms.txt. Add it to an agent's context, point a crawler at it, or give it to a coding agent.
- POST /v1/forecast { location, units?, current?, quarter_hourly?, hourly?, daily? }
Named forecast sections with flat values: current, quarter_hourly, hourly, and daily.
- POST /v1/nowcast { location, timezone?, units? }
Pro+ (nowcast:read). Fixed two-hour precipitation nowcast in one-minute steps, including amount, rate, probability, confidence, and hazard signals when available.
- POST /v1/accuracy { scope, days?, units?, minutely?, hourly?, daily?, compare_to?, include? }
Verified error vs real station observations for a locality, weather region, or the whole network: error and bias by lead, precipitation event skill, probability calibration, and an optional persistence skill score.
- POST /v1/historical { location, timezone?, units?, start, end, variables, include? }
Field-selectable daily observations for completed local dates: maximum/minimum temperature, precipitation, rain, snow, and AQI, with optional station and source provenance.
- POST /v1/alerts { location, include? }
CAP-aligned active public alerts for an exact point.
- POST /v1/outlook { location, units?, start, end, variables }
Daily climate outlook for days 15 through 365.
- POST /v1/elevation { location, units?, include: [terrain_profile]? }
Ground DEM elevation and an optional valley / mid / peak terrain profile.
- GET /v1/geocode q
Place search: name → coordinates.
What agents can check
An agent can call POST /v1/accuracy to read
the verified error record for a set of coordinates. Set
compare_to: ["persistence"] to compare the
forecast with a naive baseline.
An agent can call POST /v1/historical for
selected daily observations over completed local dates, then inspect
optional station and source provenance. It should use
/v1/accuracy instead when the question is
how a forecast performed.
A single forecast response can include current, quarter-hourly, hourly, and daily sections, so the agent does not have to combine several endpoint responses.
Weatherkind ingests raw NOAA and ECMWF model output and runs its own engine. The forecast does not come from another commercial weather API.
Paths are versioned, changes within a version are additive, and field names and units stay stable.
Run the accuracy check.
Request the accuracy record for any coordinates as JSON before you choose the API.
curl https://api.weatherkind.com/v1/accuracy \
-H "Authorization: Bearer $WEATHERKIND_API_TOKEN" \
-H "Content-Type: application/json" \
-d '{
"scope": { "type": "local",
"location": { "latitude": 39.19, "longitude": -106.82 } },
"hourly": { "variables": ["temperature"] },
"compare_to": ["persistence"]
}'