v1.0 — now in private preview

The agentic meteorology platform for the AI‑native era.

Relaweather gives your systems a sense for the sky. Autonomous agents ingest petabytes of atmospheric signal, reason across models, and act — so your products stay ahead of every front, gust, and gradient.

1.7B
signals / hour
240 ms
tail latency
94.3%
12h skill score
Wireframe topography of an atmospheric pressure field
live · north atlantic · 850 hPa
Hyperlocal nowcastingCyclogenesis trackingRenewable yield forecastingWildfire risk synthesisMaritime routingInsurance peril modelingAviation turbulenceAgricultural advisoriesHyperlocal nowcastingCyclogenesis trackingRenewable yield forecastingWildfire risk synthesisMaritime routingInsurance peril modelingAviation turbulenceAgricultural advisoriesHyperlocal nowcastingCyclogenesis trackingRenewable yield forecastingWildfire risk synthesisMaritime routingInsurance peril modelingAviation turbulenceAgricultural advisories
01 — thesis

The atmosphere is the largest unstructured dataset humanity has ever measured. Legacy meteorology stops at the chart. We start where the decision begins — with agents that read, reason, and respond to weather on your behalf.

02 — capability

A swarm of specialists, not a single oracle.

Each agent owns a sliver of the atmosphere — pressure, moisture, radiation, momentum — and negotiates with the others. The result is a continuous, queryable picture of what the sky is about to do.

NOWCAST-α

Sub-kilometer nowcasting, refreshed every 90 seconds.

Convective bursts, microbursts, precipitation onset — resolved at a scale your operations actually feel. Stream as MQTT, gRPC, or natural language.

1km grid90s cadenceglobal
SYNOPTIC-β

14-day ensemble reasoning.

Probabilistic narratives across 51 ensemble members, distilled into one decision-ready brief.

PERIL-γ

Real-time peril attribution.

Quantify wildfire, flood, and wind exposure across your asset book before the event hits.

API
> rela agent ask cyclo-7 \
    "should we ground flights at SFO 14:00 UTC?"

yes — confidence 0.81
├─ low-level shear: 38 kt
├─ cig: 400 ft, descending
└─ recommend: defer 90m
fabric

Bring your own model.

Wire in your private observations or downstream physics-ML. Relaweather fuses them with public NWP into one coherent state.

03 — loop

Sense.
Reason.
Act.

  1. /01
    Sense
    Satellites, radars, METARs, soundings, IoT, social — 1.7B signals/hr normalized to a shared ontology.
  2. /02
    Reason
    Agents debate across NWP, neural, and analog models — arriving at a calibrated posterior, not a single forecast.
  3. /03
    Act
    Push to your stack: routing engines, alerting, autonomous fleets, customer comms — with the audit trail to prove it.
04 — fabric

Every model worth running, in one fabric.

We don't pick a model. We orchestrate them — weighting each by skill on the specific phenomenon, region, and horizon you care about.

01
ERA5 + IFS HRES
ECMWF physics, hourly
live
02
GFS 0.25°
NOAA operational
live
03
GraphCast 2.0
DeepMind neural NWP
live
04
Pangu-Weather
Huawei 3D transformer
live
05
FourCastNet v3
NVIDIA fourier neural
live
06
AURORA-L
Relaweather foundation model
preview
private preview · q3 2026

Give your systems a sense for the sky.

We're onboarding a small cohort of teams in energy, logistics, aviation, and insurance. Tell us what you're building.

~48h response · no marketing