A realtime all-sky radar in your browser. Live aircraft, satellites, sun & moon — projected onto a polar sky dome around a fixed observer, scored for anomalies by a Rust/WASM engine, with no build step, no API keys, and no backend.

▶ Live demo: ruvnet.github.io/skygraph

The dome reads like a star chart of everything moving above you: the centre is straight up (zenith), the outer ring is the horizon, north is up. Every dot is a real aircraft from public ADS-B feeds, gliding at 60 fps; every diamond is a satellite propagated live with SGP4; the gold ✦ ones are sunlit against a dark sky — step outside and you can see them with the naked eye.

• Live air traffic — polls airplanes.live (fallback adsb.lol) every 5 s for aircraft within 40 NM, with smoothed dead reckoning between polls so dots glide instead of snapping. Type, registration, squawk, wake category, climb/descend arrows; emergency squawks (7500/7600/7700) get a red double ring and a header alert.
• Satellites — CelesTrak TLEs (visual ≈ the 150 brightest, stations, or starlink) propagated per frame with SGP4 compiled to WebAssembly. A 24 h pass predictor lists upcoming naked-eye passes (rise–set, max elevation, direction) and can fire a browser notification 5 minutes before one.
• Sun & moon — drawn on the dome; the sun's position also drives the satellite illumination test (cylinder shadow model) that decides which passes are actually visible.
• Anomaly scoring — every live track is embedded into a 32-dim vector (in Rust/WASM) and scored against a rolling memory of past traffic persisted in IndexedDB: the dashboard literally remembers your sky across sessions. Scores follow the §15 composite formula with mandatory human-readable reasons; dots and rows take the band color (green normal → red strong anomaly).
• Behavior badges — geometric detectors flag holding patterns [HOLD], survey grids [GRID], go-arounds [GO-AROUND], and formation pairs [FORM].
• Conflict prediction — pairwise closest-point-of-approach in observer coordinates; predicted separation < 1 km horizontal and < 300 m vertical within 90 s draws a dashed alert line between the pair (visible in the screenshot header: "⚠ CPA alert").
• Weather & space weather — Open-Meteo current conditions (temperature, wind + gusts, humidity, pressure, cloud, precipitation) and the NOAA SWPC planetary Kp index with an aurora hint when it spikes.
• Routes — click an aircraft and its callsign resolves to airline + origin → destination via adsbdb.
• Recorded replay — the last hour of real traffic streams into an IndexedDB ring buffer; the footer ⏪ scrubber replays it. There is no synthetic data anywhere.
• Offline grace — block every API and the dome stays up with a retrying status line. Zero page errors, verified headless.

Everything is vanilla JavaScript ES modules + a single Canvas (an experimental WebGPU instanced path for Starlink-scale satellite counts hides behind a ⚙ toggle). Every file stays under 500 lines.

The demo is fully static — any file server works (ES modules need http://, not file://):

git clone https://github.com/ruvnet/skygraph
cd skygraph/docs
python3 -m http.server 8000
# open http://localhost:8000/

The prebuilt WASM engine is committed at docs/pkg/, so this works with no toolchain at all.

Top-right gear: toggle layers (aircraft / satellites / sun & moon / trails / labels / conflict alerts), pick the satellite TLE group, set trail length, try the WebGPU satellite renderer, and arm pass notifications. Settings persist in localStorage.

docs/                  the dashboard (GitHub Pages serves this)
  sky.js               conductor: render loop, selection, layer wiring
  project.js, astro.js projection math (WGS-84→az/el/range), sun/moon, sat illumination
  live-feed.js         ADS-B + weather polling, rolling tracks, table sync
  sat-feed.js          CelesTrak TLE fetch + 6 h cache
  novelty.js           IndexedDB embedding store → calibrated vector novelty
  score-live.js        §15 scoring bridge into the WASM AnomalyScorer
  behavior.js          holding / grid / go-around / formation detectors (pure fns)
  conflict.js          pairwise CPA + predicted-path cones (pure fns)
  passes.js            24 h visible-pass timeline + notifications
  record.js            recorded-replay ring buffer
  gpu-sats.js          experimental WebGPU instanced satellite sprites
  draw.js, panels.js, settings.js, route-info.js, space-wx.js
  pkg/                 prebuilt sky-monitor-wasm (committed for the demo)

core/                  Rust: the SkyGraph appliance core (ADR-199)
  coords, track, embedding, anomaly, …   pure subset → compiles to wasm32
  indexer, skygraph, pipeline            native-only (RuVector VectorDB + GraphDB)

wasm/                  Rust → WASM bindings used by the dashboard
  SkyProjector         batched WGS-84 → az/el/range projection
  SatPropagator        SGP4 from TLEs + predict_passes (24 h timeline)
  embed_track/novelty  32-dim track embeddings + indexer-calibrated novelty
  AnomalyScorer        the exact §15 composite scorer, native-parity tested

The projection, embedding, and scoring code paths in the browser are the same Rust as the native appliance — compiled to WebAssembly, with parity tests asserting identical output. A plain-JS mirror of the projection keeps the page functional if docs/pkg/ is missing (satellites, novelty, and pass prediction need the WASM engine).

# rustup target add wasm32-unknown-unknown && cargo install wasm-pack
wasm-pack build wasm --target web --out-dir ../docs/pkg

The full pipeline — RuVector VectorDB similarity search, the SkyGraph property graph with citeable explain(), daily sky briefs — runs natively:

cargo run -p sky-monitor --release     # synthetic-day demo + SkyGraph + brief
cargo test -p sky-monitor              # acceptance tests (ADR-199 §31)
cargo test -p sky-monitor-wasm         # wasm crate (native-parity, SGP4, screen mapping)
node --test docs/test/                 # behavior + CPA detectors

The default observer is the reference Oakville node (43.4675 N, −79.6877 W). Edit OBSERVER in docs/sky.js to move it to your location.

SkyGraph is the presentation plane of ADR-199 ("Sky Monitor / SkyGraph appliance") from the RuVector project — see the sky, remember the sky, explain the sky — where the same core drives a native appliance with vector memory and a graph store. Built with Claude Code multi-agent swarms.

MIT © 2026 rUv