Fly to the Strait of Hormuz in God’s Eye View, switch on the vessel layer, and the water goes blank. One of the busiest oil chokepoints on the planet, and the screen shows almost nothing moving. Rotterdam, an ocean away, is so dense with ship icons you can barely see the harbour underneath them.
Nothing is broken. You are looking at the shape of a data feed rather than the shape of the world, and understanding that difference is the single most useful thing you can bring to this software.
The short version: God’s Eye View is a free, open-source 3D globe application from Bilawal Sidhu and Sameh Khamis at Halfpixel, published on GitHub under the MIT licence. It runs on your own computer and drapes public data feeds — aircraft transponders, ship beacons, orbital elements, earthquake reports, government traffic cameras — over a photorealistic Earth. There are two ways in: a one-click Pinokio launcher for non-developers, or a terminal clone for everyone else. It starts with no account and no API keys, most layers stay free, and the only component that reliably costs real money is the voice assistant.
What it is, and what it isn’t
The project describes itself as a spy-satellite simulator where the sources are public and the data is real, which is an unusually honest piece of marketing. There is no satellite overhead feeding you live video. The photorealistic globe underneath everything is Google’s Photorealistic 3D Tiles — the same textured 3D mesh that powers Google Earth, built from aerial and satellite imagery captured over months and years, not minutes.
What moves on top of that fixed globe is live-ish telemetry from fifteen separate feeds. The distinction matters in practice: when you zoom into an airport and see aircraft on the taxiway, the aircraft positions are current and the tarmac they’re parked on is a photograph from some point in the past.
Sidhu is a former Google product manager who spent six years on augmented reality and 3D mapping before moving into independent technology commentary; the project began life as a weekend build called WorldView. The launch landed harder than anyone expected. The repository reached number one on GitHub Trending, daily and weekly, in August 2026. Chris Messina, who proposed the Twitter hashtag, posted it to Product Hunt, where it took eighth Product of the Day. Brendan Eich — creator of JavaScript and co-founder of both Mozilla and Brave — turned up in the replies with a two-word endorsement. As of 20 September 2026, the repository carries roughly 38,800 stars and 7,800 forks.
Installing it without touching a terminal
The non-developer route runs through Pinokio, a desktop launcher that handles dependency installation for AI and web projects. Install or update Pinokio to version 8.2 or later, open the God’s Eye View entry, click Install, then Start. It works on Windows, macOS and Linux.
The version requirement is not optional trivia. Earlier Pinokio releases failed to install this particular launcher, and 8.2 is the release that fixed it. If you tried this in the first week after launch and gave up, that is probably why.
One caution the README spells out and the video doesn’t: do not type API keys into Pinokio 8.0.40’s native Configure panel, which mishandles the app’s nested config file and logs submitted values. Keys belong in the app’s own POWER UP panel instead.
Installing it from the terminal
The developer route is a standard Node project. You need Node.js 24.14.0 or later on the 24.x line, or 26.x — the project’s setup checker warns against Node 25, which has reached end of life.
git clone https://github.com/bilawalsidhu/gods-eye-view.git
cd gods-eye-view
npm ci
npm run doctor
npm run dev
Then open http://localhost:4173. Note npm ci rather than npm install: it installs the exact locked dependency versions, which is what you want for a project moving this fast. The doctor step reports Node readiness and which providers it can find, without printing your credentials back at you.
Pointing an AI coding agent at the repository works too, and is how most non-developers in the project’s community appear to have got started. The server binds to localhost on both paths, so nobody else can reach it unless you deliberately open it up.
What’s actually on the globe
Fifteen layers ship with the app, and thirteen of them work with no key at all.
| Layer | Source | What you get |
|---|---|---|
| Flights | OpenSky Network + adsb.lol | 11,000+ live aircraft, plus route history |
| Military flights | adsb.lol | ADS-B military traffic, rendered in amber |
| Vessels | AISStream | Thousands of ships, weighted heavily to coastlines |
| Satellites | CelesTrak | An 838-object catalogue, propagated with SGP4 |
| Earthquakes | USGS | Global seismic activity, trailing 24 hours |
| Active fires | NASA FIRMS | Thermal hotspot detections, trailing 24 hours |
| Public cameras | City and state transport APIs | ~3,600 cameras projected into the 3D scene |
| Traffic | TomTom + OpenStreetMap | Simulated vehicles, real congestion colouring |
| Radio | Radio Browser | Up to 750 geolocated stations with an analogue tuner |
| Transit, bikeshare, directions | Operator GTFS-RT, GBFS, OSRM | Live vehicles, dock availability, street routing |
Bundled alongside those are static infrastructure datasets: 4,351 datacenters, 704 dams and 712 submarine cables. Those don’t change hour to hour, which is precisely why they’re shipped in the box rather than fetched.
ADS-B is worth a one-line gloss, because two layers depend on it: it’s a system most commercial aircraft use to broadcast their own position, altitude and identity on an open radio frequency, which anyone with a cheap receiver can pick up. The crowdsourced networks that aggregate those receptions are the reason live flight data is free at all. The video credits “tens of thousands of volunteers” — the OpenSky Network’s own reporting puts it at more than 6,000 registered sensors, heavily concentrated in Europe and North America. Impressive, but an order of magnitude smaller, and the concentration explains a lot about what you can and can’t see.
The traffic layer is a simulation
This is the correction that matters most, because the demo is genuinely convincing and the conclusion most viewers draw from it is wrong.
Watching an Austin highway turn solid red at 6:30pm, then diving into a camera at that intersection to confirm bumper-to-bumper traffic, feels like a verification loop. It isn’t quite. The project’s documentation states plainly that the traffic layer renders simulated vehicles on OpenStreetMap roads, and that with a TomTom key, live flow speeds drive the simulation and the congestion colours below roughly 8 km. Individual vehicle positions are not live observations.
So the congestion is real, sourced from TomTom’s aggregated flow data. The cars are a visualisation of that aggregate, not a picture of specific vehicles. You cannot pick a car out of that stream and follow it, and the software is not pretending you can.
What “real time” means here
The app renders motion smoothly, which creates an impression of continuous live tracking that the underlying data doesn’t support. Feeds arrive every 15 to 30 seconds. God’s Eye View deliberately renders one interval behind real time and interpolates between known fixes, using dead reckoning to fill the gaps. A plane gliding across your screen is partly measured and partly predicted.
Public camera feeds refresh far more slowly still — the operators publishing them generally supply high-resolution stills at low frame rates rather than video, which limits what anyone can do with them. Rocket launch replays in the Space Missions layer are labelled RECONSTRUCTED ESTIMATE in the interface, because that’s what they are: a modelled ascent, not telemetry.
None of this makes the tool less interesting. It does mean “live” is doing different work in different layers, and the app is more careful about saying so than most of its coverage.
Why the ocean goes quiet
Back to the Strait of Hormuz. Ships broadcast their position using AIS, a maritime equivalent of ADS-B, and AISStream aggregates those broadcasts into a free feed. The catch is physics. Aircraft transmit from tens of thousands of feet, so a single ground receiver covers an enormous area. Ships transmit at sea level, so a receiver only hears them within roughly line of sight from shore.
Where there are lots of volunteer receivers near busy coastlines — northern Europe, the US — coverage is excellent. Where there aren’t, the map goes dark. Closing that gap requires satellites carrying radio-frequency sensors, and that data is sold commercially, sometimes at a premium over exactly the regions people most want to watch.
The practical lesson generalises well beyond this one app: on any open-source intelligence map, an empty area means nobody is listening there, not nothing is happening there. Treating absence as evidence is the most common way people misread tools like this.
What it costs
The headline answer is that a curious individual can run the whole thing for nothing. The detail is worth having, because “free” splits into four different things.
Keyless and free. Esri satellite imagery, OpenSky anonymous flight polling, adsb.lol, CelesTrak, USGS, public cameras, radio, transit, bikeshare and OSRM routing. No signup at all.
Free, with a signup. AISStream for ships, NASA FIRMS for fires, TomTom for traffic flow, and an OpenSky account for higher polling allowances. Provider quotas apply.
Free but conditional. The photorealistic 3D globe is the interesting case. A Cesium ion Community account is free for individuals and explicitly limited to personal, non-commercial use, with a quota of 1,000 Google Photorealistic 3D Tiles root tiles per month. Going direct through Google instead means enabling billing: Google’s published pricing gives 1,000 free Photorealistic 3D Tiles requests per month, then charges $6.00 per thousand in the first paid tier. One root-tile request covers a long stretch of rendering, so a solo user exploring occasionally can realistically stay inside either free allowance — but the key should be restricted and a budget alert set regardless.
Actually metered. Voice. The assistant runs on OpenAI’s Realtime API, priced at $32 per million audio input tokens and $64 per million output tokens for the flagship realtime model. Per-minute cost depends on how much the agent talks and how well the session caches; independent estimates for production voice agents generally land somewhere between a few cents and a few tens of cents per minute, with the mini model materially cheaper. God’s Eye View meters this for you: a live session-spend readout, a model toggle, a $2 warning and a $5 cap that ends the session.
The trap is the non-commercial clause. Cesium’s Community tier and several bundled datasets are licensed for personal use only, and many providers treat any publicly deployed service as commercial even when users pay nothing. Running it on your own laptop is fine. Putting it on a public URL is a licensing question, not just a hosting one.
Talking to the globe
Voice is the feature that makes the demos travel, and it’s more capable than a voice remote. The agent has twenty-eight tools and pulls live scene context — coordinates, street names, active layers, view scale — before answering, so “what city is this?” mid-flight resolves correctly.
It does four broad jobs: camera direction (“orbit this area slowly”), annotation (“outline the state of Texas”, which draws the actual enclosing boundary rather than a circle), querying the live layers (“how many flights are over Texas right now?”), and operating the interface hands-free. Routes are drawn on real streets and the camera can fly them.
Two design decisions deserve credit. Your OpenAI key never reaches the browser — the client receives only a short-lived session token. And at street level, where the agent reads a screenshot to identify signage, it’s instructed not to invent labels it can’t see. Voice is also entirely optional; without a key the mic button simply reports that voice is unavailable and everything else works.
The line the project draws
A tool with this aesthetic invites surveillance framing, and the maintainers have pre-empted it in writing. The stated scope is events, assets, infrastructure and systems — aircraft, vessels, satellites, fires, cameras, cities. The project does not build features for named-person search, face recognition or individual tracking, and says pull requests crossing that line won’t be merged.
The community ALPR layer, merged since the walkthrough, is a good test of whether that holds. It maps the locations of automated licence plate readers using OpenStreetMap nodes tagged man_made=surveillance and surveillance:type=ALPR — the same crowdsourced tagging behind the DeFlock project. It shows where the cameras are. It does not touch plate records, camera video, or operating status, and the documentation says so explicitly. Mapping surveillance infrastructure and conducting surveillance are different activities, and this layer is firmly the first.
That said, the honest version of the privacy point is that combining public datasets reveals more than any one of them does alone. Every individual feed here is published deliberately by someone. Fusing them onto one globe is a meaningful change in accessibility, even when nothing new is disclosed.
What’s changed since the walkthrough
The video ends with Sidhu asking viewers whether they’d want a hosted version — a link you click instead of software you install. That question has been answered. The README now confirms a hosted God’s Eye View is being built at Halfpixel, with the repository staying open source. No date or pricing has been published.
The ALPR layer has landed. Issue counts and open pull requests suggest a fast-moving contribution queue, and the maintainers have been explicit that experimental layers can be pulled in by an AI agent before they’re merged into main.
One claim from the video is worth flagging as unresolved. Sidhu notes that free Sentinel satellite imagery over conflict zones runs about three days behind, and reads that as governments restricting access. A simpler explanation fits the published specifications: the Sentinel-2 constellation revisits a given point roughly every five days at the equator and every two to three days at mid-latitudes, and ESA states that Level-2A products are available within about eight hours of sensing. What you’re usually waiting on is the next overpass, not a censor.
What it’s actually for
The strongest use case isn’t watching planes, however good the cockpit mode is. It’s reconstruction — taking a real event, pulling public reports and footage, and plotting them in three dimensions to understand how something unfolded across terrain. Several people have already done this for recent disasters, and the architecture is built for it: every layer is a separate module, and pointing a coding agent at the repo to add your own feed is the documented workflow rather than a hack.
The ceiling is real and the project names it. Terrestrial AIS goes quiet mid-ocean. Premium imagery, synthetic aperture radar and the deeper commercial feeds sit behind enterprise contracts. And the maintainers’ own warning about the long game is the most interesting technical note in the README: the present is the cheap part, and tiling, serving and scrubbing what changed over time at any real resolution is where the costs turn brutal.
The project’s own disclaimer is the right place to end. God’s Eye View is an exploratory visualisation of public and third-party data, which may be delayed, incomplete, modelled, inferred or simply wrong. It is not built for flight or maritime navigation, emergency response, or any operational decision. Read as a window onto which signals about the world happen to be public, it is one of the more interesting things you can install this year. Read as an authoritative picture of what’s out there right now, it will mislead you — and mostly by what it leaves out.
Repository state, pricing and layer details checked 20 September 2026.
9. FAQ
Is God’s Eye View free? The software is free under the MIT licence, and thirteen of its fifteen layers work with no account. The photorealistic 3D globe is free through a Cesium ion Community account for personal, non-commercial use within a 1,000-root-tile monthly quota, or metered through a Google Maps key with 1,000 free requests per month. Voice is the one component that reliably costs money.
Do I need to be a developer to run it? No. Install Pinokio 8.2 or later, open the God’s Eye View entry and click Install, then Start. The terminal route exists for people who want to modify the code.
Is the data real or simulated? Both, depending on the layer. Flights, ships, satellites, earthquakes, fires and camera feeds are real public data. The traffic layer renders simulated vehicles on real roads, coloured by real TomTom congestion data. Launch replays are labelled as reconstructed estimates.
Why are there no ships in some regions? The free AIS feed depends on shore-based receivers, and ships transmit at sea level, so coverage is strong near well-equipped coastlines and thin elsewhere. Empty water means no receiver is listening, not that no ships are present.
Can I use it to track a specific person or vehicle? No, and the project has ruled it out by design. There is no named-person search, no face recognition and no individual tracking, and pull requests adding them are stated to be unmergeable. The ALPR layer maps where plate-reader cameras are, using OpenStreetMap tags — it holds no plate records.
Can I put it on a public website? Not without changing the data providers. Several of the free tiers, including Cesium ion Community, are licensed for personal, non-commercial use, and many providers treat a free public service as commercial use.
Is there a hosted version? The maintainers have confirmed one is being built at Halfpixel, with the open-source repository continuing alongside it. No launch date or pricing has been announced.
