You pick a Delta jet on final approach into Atlanta, click once, and the camera drops into the cockpit. The terrain underneath is real. The altitude readout is real. The plane is real, and it is landing right now.
Then you remember this is running in a browser tab on your laptop, and it cost you nothing.
Quick answer: God’s Eye View is a free, open-source web app that puts live public data — aircraft, ships, satellites, fires, earthquakes, public traffic cameras, radio stations — onto a photorealistic 3D globe you can fly around and talk to. Non-coders can install it in about two clicks using a desktop launcher called Pinokio; developers can clone the repo and run it locally. Most layers work with no account and no API keys, and for personal use it typically costs nothing unless you turn on voice control.
What is God’s Eye View, exactly?
It’s a browser-based globe built by Bilawal Sidhu and Sameh Khamis that fuses public data feeds into one map. The source code is on GitHub under an MIT license, and it runs locally on your own machine rather than on someone else’s server.
The framing is deliberately cinematic — night-vision filters, thermal imaging, a tactical heads-up display — but the underlying signals are ordinary open data. Flight transponders. Ship beacons. Seismographs. City camera APIs.
That gap between the spy-thriller aesthetic and the mundane sourcing is the whole point. It looks classified. It’s assembled entirely from things anyone can download.
Why is everyone suddenly talking about it?
Because the numbers got silly. The repository hit number one on GitHub Trending — daily and weekly — in August 2026, and as of this writing it sits at roughly 36,000 stars and 7,200 forks.
Chris Messina, who popularised the Twitter hashtag, posted it to Product Hunt, where it landed at #8 Product of the Day. Brendan Eich — creator of JavaScript, co-founder of Mozilla and Brave — weighed in publicly as well. There’s a pleasing circularity there: the language Eich designed in 1995 is the one the whole thing is written in.
Do you need to be a developer to run it?
No. There are two paths, and the first one has no terminal in it.
Install Pinokio version 8.2 or later, open the God’s Eye View entry, click Install, then Start. Pinokio is a desktop launcher that handles dependencies and ports for you. It works on Windows, macOS and Linux, finds a free local port, and opens the app.
One caveat worth knowing: if you tried earlier and installation failed, that was a launcher bug. Version 8.2 fixes it — update and try again.
What’s the developer path?
Clone it and run it. You’ll want Node.js 24.14.0 or later (or 26.x), then git clone, npm ci, npm run doctor, npm run dev, and open localhost:4173. The setup doctor warns you off Node 25, which is end-of-life.
The more interesting option is to hand the repository URL to a coding agent running on your machine — Claude Code, Codex, Cursor — and let it do the setup, then start modifying things. The project is explicitly designed for that.
There’s no framework involved. It’s vanilla JavaScript, CesiumJS and Vite, plus Google Photorealistic 3D Tiles for the planet and the OpenAI Realtime API for voice. Plain JavaScript is easy for both humans and language models to read, which is why adding a new layer is a weekend project rather than an architecture review.
What does it cost?
Almost nothing, for personal use. Most layers are free with no signup at all — flights, earthquakes, satellites, city cameras, radio and launches. Another tier is free but needs a signup: ships via AISStream, active fires via NASA FIRMS, live traffic speeds via TomTom.
The photorealistic 3D globe is the one to understand properly. A free Cesium ion token unlocks Google Photorealistic 3D cities and world terrain for eligible personal, non-commercial use, within current quotas. A Google Maps key gives you the same tiles directly, metered, plus in-app place search. Without either, you get 2D satellite imagery and everything still works.
The genuine cost is voice. Realtime audio runs a few cents per active minute, and the app meters it for you: a live spend readout, a $2 warning, and a $5 hard cap that ends the session. That cap is a small design decision that will save a lot of people a bad surprise.
How does it know where every plane is?
This is the part most people find genuinely surprising: nobody in this chain is a government or an airline.
Aircraft broadcast their position continuously over a system called ADS-B — Automatic Dependent Surveillance–Broadcast, essentially a radio shout of “here I am, at this altitude, going this fast.” Hobbyists with cheap receivers pick those broadcasts up and forward them to shared networks.
The scale is often overstated. The OpenSky Network began with eight sensors in Switzerland and Germany and has grown to over 7,000 registered receivers worldwide — thousands, not tens of thousands. Each one typically covers 100 to 500 kilometres in line-of-sight conditions.
Stitched together, those volunteers maintain a live model of civil aviation that researchers use for everything from climate analysis to pandemic mobility studies. God’s Eye View pulls from OpenSky and adsb.lol, showing more than 11,000 live aircraft with route history, and renders military ADS-B traffic in amber.
Why does the ocean go quiet in some places?
Fly to Rotterdam and the water is packed with ships. Fly to the Strait of Hormuz and it can look eerily empty. There are two reasons, and most coverage of this tool only gives you one.
The first is physics. Ship transponders use VHF at sea level, not a signal beaming down from 35,000 feet, so you need receivers near shorelines. Terrestrial AIS reaches roughly 50 kilometres from the coast. Satellite AIS extends that to open ocean, but sits behind paid plans. Free feeds are dense in Europe and North America and thin elsewhere.
The second reason is more uncomfortable, and it’s specific to 2026. Ships in that corridor are deliberately switching their transponders off. Fortune reported that AIS now appears to underreport Hormuz traffic by roughly half, with close to a million GPS jamming incidents affecting over 1,100 vessels in a single quarter.
So an empty patch of map is ambiguous evidence. It might mean no receivers. It might mean no ships. It might mean ships that don’t want to be seen. Any tool that shows you absence needs to be read carefully, and this is the single most important habit to develop before you start drawing conclusions from a live globe.
Is the traffic layer showing real cars?
No — and this is the most misleading impression the demos create, so it’s worth stating plainly.
The traffic layer simulates vehicles along real OpenStreetMap roads. With a TomTom key, live flow speeds drive that simulation and set the congestion colours at closer zoom levels. But individual vehicle positions are not live observations.
The colours are real data. The cars are not. If you zoom into an intersection and watch a specific vehicle move, you’re watching a plausible animation constrained by real aggregate speeds — not that car.
That distinction matters more than it sounds. The credibility of any OSINT tool rests on being clear about which pixels are measurements and which are models. The project’s documentation is admirably direct about it, even when the footage is seductive.
What can you actually see through the public cameras?
Around 3,600 public cameras are projected into the 3D scene — not embedded as floating video windows, but mapped into the city geometry itself. Coverage spans Austin, wider Texas via TxDOT, California via Caltrans, London via TfL, plus Ontario, Finland, British Columbia, Estonia, New South Wales and Calgary.
Camera positions are published by the agencies. The orientations are estimated, and you can correct them yourself by dragging a handle on the camera.
On refresh rates, the common claim of one frame every five to ten minutes is too pessimistic. Austin’s still images update roughly once a minute, and TxDOT cameras typically refresh every one to three minutes.
The privacy protection is real, but it works differently than people assume. The City of Austin states that these cameras monitor live traffic conditions only, and that video is not recorded or retained. The safeguard is that nothing is stored — not that the frame rate is crippled. That’s a stronger guarantee, and a more honest one.
What is voice mode good for?
You need an OpenAI key, and then you can talk to the planet. The agent pulls live scene context before answering — coordinates, street names, active layers, view scale — so asking what city you’re over mid-flight actually works.
It handles four broad jobs: directing the camera, annotating the world with real boundary polygons and routes, querying the live layers, and operating the interface hands-free. Asking it to outline Texas draws the actual state boundary rather than a circle. Asking for a walking route between two points produces a street-following path you can then fly along.
Two design details deserve credit. Your OpenAI key never reaches the browser — the client only receives a short-lived session token. And the agent is instructed to confirm only actions that actually succeeded. An AI assistant that cheerfully claims it did something it didn’t is worse than no assistant at all.
Treat its general knowledge with normal caution, though. When it volunteers a fun fact about an airline or an air base, it’s doing what language models do, and the confident ones are the ones to check.
Does it hold up on the details?
Mostly, yes — which is a quiet endorsement of building on primary sources.
Fly to the Texas–Mexico border and you’ll find a base thick with training aircraft. Laughlin Air Force Base, east of Del Rio, is the largest pilot training base in the US Air Force and home to the 47th Flying Training Wing under Air Education and Training Command. Roughly 400 new military pilots earn their wings there each year after a 52-week course. The dense circuit patterns on screen are exactly what a pipeline base should look like.
The Delta anecdote checks out too, and it’s a good one. The airline’s history begins with Huff Daland Dusters, the world’s first aerial crop-dusting company, incorporated on 2 March 1925 in Macon, Georgia to fight the boll weevil devastating Southern cotton. Delta Air Service was incorporated in 1928 and named after the Mississippi Delta region.
One practical tip that isn’t in the docs: when you arrive somewhere new, wait a few seconds before judging what you see. Terrain tiles stream in progressively, and until they load, aircraft on the ground can appear buried in it.
Can you use it to understand an actual disaster?
This is where the toy stops being a toy.
On 26 August 2026, a catastrophic flood tore down the Trishuli and Lhende Khola river valleys on the Nepal–China border. It was triggered by the partial collapse of the glacier atop Langtang Lirung and extended along roughly 100 kilometres of river. By early September, more than 1,300 people had been confirmed killed, with thousands still missing and damage estimated between $4 billion and $7 billion.
One technical detail is widely reported backwards, and the correction is interesting. The USGS catalogued an M 5.2 event 55 kilometres northwest of Kodari — but classified it as a landslide, not an earthquake. The ground shaking wasn’t the cause. It was the signature of a mountainside moving, picked up by seismographs.
People have been using God’s Eye View to reconstruct events like this: plotting the flood path, pinning geolocated videos along the valley, layering official reports onto real terrain. It’s the same workflow investigative newsrooms use, running on a laptop, for free.
Reporting afterwards noted that the collapse happened in a monitoring blind spot, and that Chinese and Nepali officials had met months earlier to discuss sharing risk data without reaching an agreement. That’s precisely the gap open tools are good at illuminating — and precisely the gap they cannot close on their own.
What does it reveal about the internet itself?
Turn on the infrastructure layers and the map changes character. The bundled static datasets include 4,351 data centres, 712 submarine cables and 704 dams.
Fly to Marseille and you’ll see why cables and compute cluster together. The city’s position in the Mediterranean makes it the gateway for cables linking Europe to Africa, the Middle East and Asia, and that concentration has pulled data centres in behind it. France’s grid helps: nuclear power accounted for roughly 65 to 70 percent of generation across 2024 and 2025, giving relatively stable, low-carbon baseload.
Landing point, then fibre, then power, then buildings full of servers. The physical shape of the internet, drawn on a globe you can spin with a mouse.
Where are the real limits?
Three, and they’re worth being clear-eyed about.
The first is money. The cheap tier is genuinely useful, but the ceiling arrives fast. Satellite AIS, synthetic aperture radar and high-resolution commercial imagery all sit behind enterprise contracts. The project says so itself rather than pretending otherwise.
The second is latency. Free and public imagery of contested areas is often days old by design. That’s a defensible policy choice, and it’s also a real constraint on what citizen analysis can do during a fast-moving event. Reasonable people disagree about where that line should sit, and the disagreement isn’t going away.
The third is scope, and the project draws it deliberately. It models events, assets, infrastructure and systems — aircraft, vessels, fires, cameras, cities. It explicitly does not build features for named-person search, face recognition or tracking individuals, and pull requests crossing that line won’t be merged.
That boundary is a policy, not a technical barrier. The code is MIT-licensed, so anyone can fork it and ignore the rule. What the maintainers can control is what the canonical project normalises — and choosing to make the default version one that watches infrastructure rather than people is a meaningful choice, even if it isn’t an enforceable one.
Should you install it?
If you’re curious about how the world is instrumented, yes. The install is genuinely two clicks, most of it costs nothing, and half an hour with it will change how you read the phrase “open-source intelligence.”
Just carry one habit with you. The project’s own disclaimer says the data may be delayed, incomplete, modelled, inferred or wrong, and that important information should be verified against authoritative sources. Simulated cars. Estimated camera angles. Reconstructed launch trajectories. Absence that could mean three different things.
And if the install still feels like too much, the maintainers have confirmed a hosted version is on the way at Halfpixel — no installation, just a link. The loudest request after launch wasn’t a new feature. It was a URL.
Which is its own small lesson about how far open source travels, and where it still stops.
