God's Eye View: what it does and what you need to run it

God's Eye View is an open-source web app that draws live aircraft, ships, satellites, earthquakes, wildfires and public traffic cameras on a 3D globe, and lets you steer it by voice. Since version 0.1.0 on 31 August 2026 it starts without a single API key. Most of what it shows is free. Voice control is the one feature that costs real money, and the app meters that for you.

This is what you can do with it today, what you need to run it, and what it costs. We read the repository, the licence, the release notes and the providers' price lists on 16 September 2026, and then installed it ourselves on a Windows 11 laptop. The first globe, with live aircraft, was on screen within a minute of cloning the code. The screenshots below are our own.

What God's Eye View is

The project began as a video series by Bilawal Sidhu, a former Google Maps product manager, first under the name WorldView. The videos showed a console that looked like a spy satellite feed but was built from public data. On 24 August 2026 Sidhu put the code on GitHub under the MIT licence. In his Map the World newsletter he wrote that readers had kept asking for it, so he released it. The repository lists two maintainers: Sidhu and Sameh Khamis of Halfpixel.

Technically it is a browser app: plain JavaScript on top of CesiumJS, served by a small Node server on your own machine. Nothing runs in the cloud on your behalf. If you add API keys, they live in a local file, and the server only answers requests from your own computer unless you deliberately open it to your network.

What you can do with it today

The README lists fifteen data layers, thirteen of which work without any key. This is what is on the globe and what each layer needs.

Three bundled datasets ship with the code: 4,351 data centres, 704 dams and 712 submarine cables. The cameras are projected into the 3D scene rather than shown as flat video, and cover Austin, Texas, California, London, Ontario, Finland, British Columbia, Estonia, New South Wales and Calgary.

The features are what make it feel like a film set:

• Click any aircraft and the camera locks on and draws its trail.
• Press Cockpit and you ride inside a real flight over real terrain.
• Number keys switch the whole planet through sensor looks such as CRT, night vision and thermal.
• A detection overlay draws bounding boxes and IDs on everything in view.
• Share links carry your camera position, layers and even a tracked target into a URL.

Voice needs an OpenAI key. With it, you can say things like "Take me to Tokyo", "How many flights are over Texas right now?" or "Outline the state of Texas", and the agent moves the camera, answers from the live layers, or draws on the globe. The README counts 28 voice tools.

Two honesty notes from the makers, which readers should know before they get excited. Street traffic is simulated on real roads, not live vehicle positions. Camera poses and rocket launch paths are estimates, and the app labels them as such.

What you need to run it

There are two ways in. The one-click route is Pinokio, a desktop launcher, version 8.2 or later, on Windows, macOS or Linux. The terminal route needs Node.js 24.14 or newer, or Node 26, and then four commands:

• Clone the repository from GitHub.
• Run npm ci to install the locked dependencies.
• Run npm run doctor to check your setup.
• Run npm run dev and open localhost:4173 in a browser.

The four terminal windows above are from our own run on Windows: the clone, npm ci reporting 123 packages, the setup doctor ending with "Ready. Run npm run dev", and Vite announcing the local address.

Without any key you get Esri satellite imagery on a 2D-style globe with keyless terrain, plus every keyless layer. If Esri is unreachable the app falls back to OpenStreetMap.

Keys are added inside the app, through a panel called POWER UP, not by editing files. The README describes them as a ladder:

Alongside those, free sign-ups at AISStream, NASA FIRMS and TomTom switch on ships, fires and live traffic speeds. An OpenSky account raises the flight polling allowance. An OpenAI key switches on voice.

We installed it: here is what happened

We took the terminal route on 16 September 2026, on a Windows 11 laptop with an Intel Core Ultra X7 358H, an Intel Arc B390 GPU and 31 GB of RAM, running Node.js 24.19. Every number below is measured, not estimated.

Two snags, both small and both caused by our own machine. First, it had no npm on its PATH, only Node. We worked around it by running npm through pnpm, which most people will never need to do. Second, the setup doctor treats missing npm as a blocker and says the app cannot start. It can: the dev server started normally. Once we put npm back on the PATH and ran everything a second time, the doctor reported npm 11.19.1 and ended with "Ready", npm ci finished in 4 seconds from a warm cache, and Vite was ready in under half a second. If npm is installed the way Node's own installer does it, neither snag exists.

One thing npm did that surprised us: version 11 skips packages' install scripts by default, so the bundled test browser was never downloaded. The app does not need it. The test suite does.

That was the whole installation. From an empty folder to a live globe took under a minute of actual waiting.

The keyless globe, ten seconds in

The first thing you see is a first-run panel with four missions. We chose Live Contacts. About ten seconds later the camera was over an airfield with a tracked aircraft, and the Contacts panel listed 233 live flights and 9 military flights within 250 km, all without a key. Ships showed "feed unavailable", which is correct: that layer needs the free AISStream sign-up. A chip at the top briefly read LOAD FAILED while one feed returned errors, then cleared on its own.

One detail we only noticed in the screenshots: on the first load the panel credited OpenSky Network as the flight source, and after a Reset Globe it credited adsb.lol. The app appears to switch between the two anonymous sources on its own. That is useful to know, because the anonymous OpenSky quota is small, and it explains why flights kept flowing.

Reset Globe pulled the camera back to the whole planet with call signs and flight levels still attached. Pressing a number key switched the view to night vision, with sliders for gain, bloom and scanlines. It all ran smoothly in a headless Chrome at 1600 by 900 pixels, which is a harder test than a normal browser tab.

What we did not test: Pinokio, anything that needs a key, and macOS or Linux.

What it costs in practice

We checked the providers' own price pages on 16 September 2026. Prices change, so check again before you rely on these.

For 3D cities, a hobbyist exploring a few evenings a week stays inside Google's free allowance, but billing must be switched on, so set a budget alert first.

Voice is the real cost, and it is billed per token, not per minute. The app uses gpt-realtime-2 as its standard model and gpt-realtime-2.1-mini as the cheaper option. OpenAI's list prices per one million tokens:

To make that concrete, we worked through one ten-minute session. Assume you speak for four minutes, the assistant speaks for three, and there are twenty turns, each resending about 3,000 text tokens of context with most of it cached. We use the common rule of thumb that a minute of speech is roughly 600 input tokens or 1,200 output tokens.

At that pace the app's built-in $5 hard cap lasts around two hours on standard. This is an estimate, not a measurement: the tokens-per-minute figures are not on OpenAI's current price page and come from its original per-minute pricing.

The app shows a live spend readout next to the microphone, warns at $2 and closes the session at $5. That cap is a safety net inside the app, not a billing limit at OpenAI, so set a usage limit in your OpenAI account too.

What it is not

It is not a people tracker. The project says it models events, assets, infrastructure and systems, will not add named-person search, face recognition or individual tracking, and will not merge pull requests that cross that line.

It is not an authoritative source. The README warns that data may be delayed, incomplete, modelled, inferred or wrong, and says not to use it for navigation, emergency response or other safety-critical decisions.

It is not a hardened service. If you open the server to your local network, anyone who can reach it can spend through your configured keys. The key-entry panel switches itself off when the server is shared.

It is not entirely MIT. The code is, but the data is not. The bundled submarine cable map from TeleGeography is licensed for non-commercial use only and must be removed for commercial deployments. OpenSky's flight data is non-commercial too, and Cesium ion's free plan is personal use. Anyone planning a commercial product on top of this has licensing homework.

Our verdict

God's Eye View is the cheapest way we know to get a feel for geospatial and open-source intelligence work without talking to a sales team, and after installing it we can add: it is also one of the least fussy. Start keyless, add the free Cesium ion token for the 3D cities, and only add an OpenAI key if you actually want to talk to it.

Two corrections to what you may have read elsewhere. Coverage from late August, including an early hands-on test by Firmatic, said a Google key was mandatory and the globe stayed black without one. That was true at the time. Version 0.1.0 changed it, and our own install confirms it: satellite imagery and live aircraft with zero keys. And the project's GitHub Trending and view-count badges are the maker's own claims. Our own reading on 16 September 2026 put the repository at 35,450 stars, which is a lot, whatever the rankings said.

Next we want to add the free Cesium ion token and see what the photorealistic 3D cities do to an ordinary laptop.

How we checked

Repository, LICENSE, package.json, .env.example, DATA_SOURCES.md and the v0.1.0 release notes on GitHub; Bilawal Sidhu's announcement on Map the World; Firmatic's 26 August test; OpenAI, Google Maps Platform and Cesium ion pricing pages, all read on 16 September 2026. Star count from the GitHub API on the same day. Installation and screenshots on 16 September 2026 from commit 531ef52, on the machine described above. Illustrations marked as AI-generated were made by us; screenshots are unedited apart from resizing.