1 · Collect
Dedicated collectors fetch forecasts, observations, alerts, text products, and measurements from each provider on a fixed cadence, independently of anyone visiting the site.
FernWX Knowledge Base
How FernWX works, and what every part of it can do.
This page is designed to explain what FernWX is and how to use it, but new changes may arrive before a proper usage guide does. You can always contact the FernWX team to learn more.
How FernWX works
Collect, preserve, present
Most weather apps show you one blended answer and discard yesterday's. FernWX is built the other way around: the record comes first, and every surface is a view over it.
2 · Preserve
Everything lands in the Weather Ledger with normalized history and the raw provider payload kept as evidence. Forecast runs are immutable: a new run is added beside the old one, never over it.
3 · Present
The web app and native clients are read-only consumers of the ledger. They never call providers directly, so what you see is always the recorded evidence, not a fresh unrecorded fetch.
The core invariant
Forecast memory is immutable. Old forecast runs are never overwritten, so questions that are unanswerable elsewhere stay answerable here: which provider changed first, which stayed stable, and which ended up closest to what was observed. AI features can summarize and explain that record, but they never write to it — the line between source data and interpretation is kept visible everywhere.
AI and human judgment
AI can help investigate the record. It does not replace it.
FernWX is a weather data, forecast-memory, and verification platform first. It collects and preserves the record so people can inspect what providers said, when they said it, and how those forecasts compared with what was later observed.
AI can navigate complex account-scoped ledger data, propose guarded read-only SQL, synthesize retrieved evidence, and help reveal patterns or trends. It never becomes weather truth, never overwrites source or forecast history, and never bypasses FernWX's deterministic query guards, execution, source provenance, or verification.
FernWX is designed for human use first. Choosing not to use AI does not diminish the value or core functionality of your forecast memory, ledger views, maps, observations, or verification tools.
When you deliberately choose an external AI provider, the request context and bounded evidence needed for that task — including information about tracked locations — may leave FernWX for that provider and are governed by that provider's terms and privacy practices. FernWX-provided and bring-your-own OpenRouter options are distinct from a ChatGPT account you connect for Make with ChatGPT; FernWX shows that distinction when you choose how to use AI.
Live present and future
Weather
The operational surfaces: what is happening at your locations right now, what each provider says comes next, and the official context around both.
Forecast
Now & Next conditions for each of your locations, with every provider shown side by side. FernWX ranks providers and shows the basis for the ranking instead of silently blending them into one number.
Answers: What is the forecast, which sources agree, and why is one ranked first?
Open ForecastSevere & Alerts
Active NWS alerts, SPC convective outlooks, a storm archive, and nearby upper-air context for your tracked locations. Qualifying severe weather automatically activates Storm Mode collection.
Answers: Is anything active near me, and what did past events look like?
Open Severe & AlertsDiscussions
Area Forecast Discussions — the forecaster-written reasoning behind NWS forecasts — collected and kept alongside the data they explain.
Answers: What are human forecasters actually thinking about this setup?
Open DiscussionsUpper-air soundings
Radiosonde (weather balloon) measurements from the launch sites that matter for your locations: the observed vertical profile of the real atmosphere, not a model of it.
Answers: What did the atmosphere actually look like above the nearest launch site?
Open Upper-air soundingsMaps
Radar, warnings, lightning, surface isobars, and fronts around your tracked locations, plus a replay of retained MADIS station reports that can show either when each reading was observed or when FernWX received it.
Open MapsThe auditability differentiator
Memory
Forecast memory is the core FernWX primitive: every forecast run is preserved and never overwritten. The Memory workspace is where you look backward — its five tabs share one location and query state, so a question carries across them.
Moment
Fix an upcoming, current, or past event window and see what each provider forecast for it at up to four earlier capture times, alongside the forecast discussion available at each. Opening Verification from a moment keeps the same hour and time zone.
Answers: What did each source say about Friday afternoon, as of Thursday morning?
Open MomentReplay
Scrub back to any earlier moment and see the forecast exactly as it stood then — which provider changed first, and which stayed stable as the event approached.
Answers: What did the forecast say on Tuesday about Friday?
Open ReplayVerification
Forecasts scored against what was later observed, so provider performance is measured rather than remembered.
Answers: Who was actually right, and by how much?
Open VerificationProviders
One card per provider: its temperature rank and mean absolute error against the NWS station reference with the sample count behind it, alongside the attribution, cadence, and interpretation limits that qualify how to read that rank.
Answers: Who is ahead, and what caveat governs that claim?
Open ProvidersRecords
The persisted forecast and verification rows behind Memory comparisons, with retrieval times, valid times, references, signed and absolute error, filters by provider, variable, knowledge lead, and valid-time range, and a JSON or CSV export of the whole filtered set.
Answers: Where did this value come from, and when was it collected?
Open RecordsQuery and analyze the record
Intelligence
Analytical surfaces built on top of the ledger. AI has a bounded role here: it can query, summarize, and explain the record, but it never writes weather truth and its output is always labeled as interpretation.
Smart Card Library
Saved, re-runnable analysis cards. Cards can be saved straight from report sections, either stay on a fixed location or follow the one you are viewing, re-run on current data, and export as presentation-ready PNGs with run provenance stamped on them. Card titles and captions can follow the location each result ran for.
Open Smart Card LibraryObservatory
Curated analytical dashboards over the ledger: provider trust scores, forecast volatility, lead-time skill, nowcast-versus-reality, and data freshness — the standing questions, pre-built.
Open ObservatoryResearch Lab
Ask questions of your ledger in plain language. FernWX translates them to SQL over your account-scoped data, shows the query it ran, and charts the result — the question surface for anything the Observatory does not already answer. It opens on Explorer. Its Agentic Reports tab goes further: an agent plans a sequence of queries, runs them, and assembles the findings into a structured report whose sections can be saved to the Smart Card Library — either as a one-off report or as a saved Runbook you can run again.
Answers: Ad-hoc questions like: how did overnight low forecasts drift last month?
Open Research LabResearch instruments
Lab
FWX Lab is an optional research view beside the main app, with one instrument per kind of evidence: upper-air soundings, model transitions, provider evidence, forecast moments, ledger queries, lightning, and the three below. Lab reads the same record; nothing in it changes that record.
Tropical Cyclone Desk
Active Atlantic, East Pacific, and Central Pacific tropical cyclones from retained NHC advisories: observed tracks, the newest official forecast, and how earlier advisories’ forecasts moved.
Answers: How has the official track for this storm shifted since it formed?
Open Tropical Cyclone DeskHydrology Lab
River and stream gauge height and discharge for a linked or searched USGS gauge, with history read directly from USGS. Same-stream Measurements compares connected gauges upstream and downstream on the same river network.
Answers: Is the river rising here, and how do gauges upstream and downstream compare?
Open Hydrology LabProvenance Repository
The raw provider payloads behind the record, with per-fetch hashes and byte counts. Open any payload as stored, download it, and check its SHA-256 hash in the browser.
Answers: What exactly did the provider send, and is this the same file?
Open Provenance RepositoryStorm Mode and collections
FernWX collects differently when severe weather matters.
FernWX normally collects weather on its standard cadence so the platform can preserve forecast history, observations, and the context around changing conditions without overwhelming the ledger.
When severe weather reaches the threshold FernWX cares about, Storm Mode turns on automatically. That switches the platform into a Rapid collection mode so warning periods are preserved with tighter timing and better event context.
Storm Mode is about evidence. It helps answer questions like what warning was active, what nearby upper-air profile was available, what the forecast outlook looked like, and what the weather sources were saying while the event was unfolding.
Normal collection
Used for day-to-day forecast memory, provider comparison, and account weather history.
Storm Mode
Used during qualifying severe-weather periods so FernWX can preserve the warning, the nearby environment, and the faster-changing source data together.
Why this exists
FernWX is built for people who want to inspect what happened, not just glance at a single weather value and move on.
Where the data comes from
Data sources
Every location you add is collected from all applicable providers by default — you never have to pick sources to get full coverage, and you can opt individual ones out per account.
- Apple WeatherKit — Forecasts and current conditions
- National Weather Service — Forecasts, alerts, discussions, radiosonde and SPC products
- Open-Meteo — Forecasts, historical context, and air quality
- Tomorrow.io — Forecasts and current conditions
- NOAA MADIS — Account-selected CWOP / Integrated Mesonet surface stations
- NOAA National Hurricane Center — Tropical cyclone advisories, tracks, and forecasts
- USGS — River and stream gauge height and discharge
See the full list of data sources and their attribution requirements.
Accounts and keys
Accounts, signup, and AI keys
FernWX currently accepts public-beta account creation without an invite code. An account carries your tracked locations, preferences, and account-scoped ledger history; signed-out visitors get a public forecast lookup that never touches account data.
Bring your own key. AI features can run on your own OpenRouter key (and an OpenAI key for speech) instead of a FernWX-provided key. Stored keys are encrypted, never shown back to any client after saving, and never echoed into error messages.
For agents and scripts
The Agent API
Everything the app can read, an authorized agent can read too. The Agent API exposes the ledger through an OpenAPI contract with account-scoped, expiring API keys, and the FernWX MCP server at fernwx.com/api/mcp serves the same reads as tools to Claude, ChatGPT, Cursor and other MCP hosts, with OAuth sign-in. Your own tools and AI agents can query the same record this site renders.