If you live in Oklahoma, you know the drill. Severe weather season rolls in and suddenly every TV station is wall-to-wall coverage with dramatic music, red banners, and worst-case-scenario graphics. The ratings machine kicks in, and it gets hard to tell what's actually dangerous from what's just dramatic.
Meanwhile, the National Weather Service — the people who actually forecast the weather — put out calm, precise, data-driven alerts that tell you exactly what's happening, where, and when. But their raw data isn't easy to read if you're not a meteorologist.
Every warning, watch, and advisory in the United States originates from the National Weather Service. TV stations get their information from the same source. OKStormWatch goes straight to that source — NWS forecast offices, NOAA weather APIs, and official VTEC-coded alerts — and gives you the same data that professional forecasters use.
No middleman. No editorial filter. No incentive to make things scarier than they are.
Raw NWS data is precise but dense. OKStormWatch uses AI to translate forecast discussions, alert text, and weather data into plain English that anyone can understand. Choose the Neighbor voice for quick glances (or reading aloud to kids), Forecaster for everyday use, or Meteorologist if you're a weather enthusiast who wants the full picture.
The AI doesn't make up data. It translates what NOAA already said into language that makes sense.
I built OKStormWatch for my family. My wife wanted to know if she should worry about the weather at the kids' school. I wanted to know if a storm cell was actually heading our way or just passing through. We didn't need hype — we needed honest information, delivered quickly, in a way we could act on.
That's still what this is. A weather station for families who want the truth about what's heading their way.
OKStormWatch has no ads. No tracking pixels. No analytics cookies. No data sales. Your saved locations, notification preferences, and alert history stay on our server and nowhere else. We use a single functional session cookie for authentication. That's it.
Your weather data is not a product. You are not a product.
This isn't a startup. There's no venture capital, no growth targets, no plan to become the next big weather platform. OKStormWatch is a personal project run by one person for a small community of Oklahoma families who care about the weather.
Accounts are personally reviewed and approved. That's intentional. It keeps the service reliable, the community small, and the signal-to-noise ratio high.
Be safe out there. Always pay attention to the weather.
— Josh
The same answer at three depths — pick whichever fits. If you're signed in, this starts on the depth that matches your briefing voice.
OKStormWatch is a private, family-built weather website. It started in Moore, Oklahoma — where tornado sirens are part of the soundtrack of spring — as a hobby project on a spare computer turned home server: a way to stay ahead of dangerous weather without the TV hype. It has one job above all others: if dangerous weather is headed toward your house, make sure you know about it right away.
Think of it as a personal weather watcher that never sleeps. Around the clock, it collects the same warnings, radar, and forecasts that TV meteorologists work from — straight from the National Weather Service and several professional forecast services — and puts everything on one simple page, minus the commercial breaks and the drama. Current conditions, today's outlook, the week ahead, and live radar you can play like a short movie.
When the National Weather Service issues a tornado warning or severe thunderstorm warning for your area, OKStormWatch tells you the moment it's issued — a loud alert on your phone, an email, a message where your family already chats. You don't have to be watching TV or standing within earshot of a siren.
It handles the everyday stuff too: a short morning weather note written in plain language, plus pollen, air quality, sunburn-level UV, burn bans, and even lightning strikes near your home. And during dangerous storms, family members can tap one button that tells everyone else "I'm OK."
There's nothing to install. It works on any phone or computer, it shows weather for your address — not a city twenty miles away — and accounts are approved personally. What began as one spare computer watching one house now watches family homes well beyond Oklahoma. It's still the same idea: someone you trust, keeping an eye on the sky.
OKStormWatch is a self-hosted severe-weather dashboard — a website that pulls from more than a dozen public weather APIs and data feeds, fuses them, and turns the result into one fast page plus aggressive notifications. Here's the tour:
Warnings, two ways. The site keeps a live, always-on connection to NWWS-OI — the NOAA Weather Wire Service, the same real-time feed broadcasters and emergency managers receive — so a tornado warning arrives the moment it's issued. As a backstop, it also polls the Weather Service's public alerts API every 15–60 seconds, and if the live wire ever goes quiet, the poller notices and automatically speeds itself up. Confirmed warnings fan out simultaneously to phone push, Discord, and email.
An early-warning layer on top. Official warnings are the floor, not the ceiling. The site also tracks each storm cell's motion and projects its path against your saved locations — so if a tracked storm is bearing down on your house specifically, you get an emergency-priority heads-up about that, not just "your county has a warning."
Forecasts that check each other. Forecast and current-conditions data come from three independent weather APIs, anchored by real observations from National Weather Service stations like KOUN in Norman. A quality-control layer constantly compares every source against that ground truth, flags disagreements, and keeps a running accuracy scorecard — the site literally grades its own data feeds and learns which one to trust for which measurement.
Radar that does more than loop. The radar view is built on NOAA's MRMS mosaic (Multi-Radar Multi-Sensor — the merged national radar picture): reflectivity, rotation tracks, and estimated hail size. Layered on top: a storm-cell tracker that projects each storm's path, a short-term "nowcast" of where the rain is headed next, model-predicted radar up to 18 hours out, and per-storm severe-weather probabilities from NOAA's ProbSevere research product.
The whole environment. Live lightning strikes from the Blitzortung community detection network, satellite cloud cover from the GOES-19 weather satellite, air quality, UV, pollen from two competing sources, drought status, county burn bans, and tropical outlooks for coastal members. And yes — daily local bird sightings from eBird. That one's a hobby crossover, not a weather product: tracking how bird patterns line up with weather patterns. Ignore it freely.
Your morning, in your voice. Every day the station writes members a short briefing from its own data, in a tone you pick — neighbor, forecaster, or meteorologist — so "should the kids take jackets?" and "what's the instability doing?" are both one read away.
Not just one house anymore. Each member's saved locations get their own data pulls, their own alert coverage, and their own timezone handling. It started in Moore; members now check in from well beyond Oklahoma.
It speaks API too. The station exposes its own small API, which is wired into an AI-assistant integration — you can literally ask a chatbot what conditions are like at home, and it answers from this system's live data.
Deliberately boring tech. Under the hood it's small, proven tools on a tiny cloud server, with dozens of scheduled jobs keeping everything fresh. Boring is a feature: at 2 a.m. when the sirens go off, you want the simple thing that works. And one rule is enforced in the code itself: no failure in any optional feature is ever allowed to silence a tornado warning.
OKStormWatch began as a one-house station for Cleveland County, Oklahoma and grew into a small multi-location fusion platform: every member location gets its own ingest fan-out, alert coverage, and timezone handling. Under the hood:
Warning ingest. A persistent NWWS-OI session delivers products at issuance time, with the public CAP API polled every 15–60 seconds as failover — the poller tightens its own cadence when wire health degrades. Dispatch parses storm motion (eventMotionDescription → motionVector → text fallback), and beyond polygon membership there's a pre-emptive threat tier: track-projected and location-specific, delivered at emergency priority when a cell is bearing down on a member's home.
Storm-scale stack. MRMS two-minute mosaics — composite reflectivity, low-level azimuthal-shear rotation tracks with a six-hour accumulation history, and MESH hail size — plus a ~90-second storm-cell tracker with motion-extrapolation nowcasting, NSSL ProbSevere v3 probabilities per object, and HRRR simulated reflectivity out 18 hours. Lightning is Blitzortung over MQTT, geohash-filtered to the region; GOES-19 ABI Clear Sky Mask rides along as a cloud-cover voting peer.
Thermodynamics. Model soundings are integrated through the EL on a 10-level profile up to 200 hPa for CAPE — an earlier 500 hPa truncation was under-reading CAPE three- to five-fold until the QC layer exposed it against HRRR — plus CIN with honest no-LFC handling, 0–500 m SRH, and SHERBS3 coverage for the high-shear/low-CAPE end of the spectrum.
Observations and trust. KOUN anchors (KOKC, KCHK, KTIK secondary), supplemented by IEM ASOS — which routinely beats api.weather.gov's publication lag by an hour or more. Every multi-source display field goes through a resolver: median voting, freshness windows, station-before-model priority. Derived quantities are pair-consistent by rule — heat index only ever computes from a same-source, time-aligned temperature/humidity pair, never station temperature × model RH. A QC suite snapshots all sources hourly, runs daily forecast verification against observed outcomes, and keeps per-source accuracy scorecards. Disagreements get flagged, not averaged away.
Tropical. NHC CurrentStorms plus the Atlantic Tropical Weather Outlook, region-gated to coastal members.
The invariant. Warning dispatch depends on none of the above. Radar, soundings, QC, lightning, satellite — any of it can fail without silencing a tornado warning. That rule is enforced in code, and it's the part of the design that matters most.