Google has released the open-source Three.Live library to convert AI-generated code into 3D scenes and augmented reality experiences in fractions of a millisecond within a secure sandboxed environment.
Launching the open-source Three.Live library and addressing the security and speed dilemma
Web security sandboxing and the embedded Rust compiler for smartphones
Instant binary format conversion and the XR.Live hand-tracking package
A project written entirely using Google Antigravity for smart coding
An analytical review of the strategic and economic dimensions of the event
Frequently asked questions
Launching the open-source Three.Live library and addressing the security and speed dilemma
Google has launched an open-source software library named “Three.Live,” which converts Three.js code generated by artificial intelligence models into interactive 3D content, augmented reality experiences, and extended reality applications for the web, Android, and iOS systems. The company announced the project on Tuesday, November 3, 2026, releasing version 1.0.0 as a developer preview under the Apache 2.0 license, with Google’s open-source blog clarifying that the library is “not an officially supported Google product.”
The project aims to solve a tangible practical problem: large language models have become adept at writing procedural Three.js code that describes shapes, materials, lighting, and animation, yet this generated code is unsafe to run and requires a secure sandboxed environment. Furthermore, mobile phones render 3D graphics faster through their native pipelines compared to running them inside an embedded browser, as explained by Jonathan Bingham from Google’s AI Foundations team.
Web security sandboxing and the embedded Rust compiler for smartphones
Three.Live handles a declarative subset of the Three.js API as a portable language for describing 3D scenes. On web browsers, untrusted scripts run inside a completely isolated Web Worker, where network connection functions such as fetch commands and web sockets are removed before executing any user code.
On Android, iOS, macOS, and the command line, scripts run inside an embedded compiler written in Rust and built on the Boa JavaScript engine. Both environments enforce strict limits on allowed execution time and geometric shapes to prevent resource exhaustion.
Instant binary format conversion and the XR.Live hand-tracking package
The compiler converts each 3D scene in memory into binary glTF 2.0 files and Apple’s USDZ format, a process Google states typically takes between 2 and 5 milliseconds. Subsequently, each platform renders the output directly through its native engines: a lightweight WebGL 2 viewer is used on the web, Google Filament and Scene Viewer on Android, and RealityKit and AR Quick Look on Apple devices.
Google states that the standard web package featuring sandboxed scripting and augmented reality weighs approximately 45 kilobytes after compression. An accompanying package named “XR.Live” adds hand-tracking gestures and 3D object generation via voice or text prompts within extended reality sessions.
A project written entirely using Google Antigravity for smart coding
Google revealed a striking detail: the Three.Live library was “written entirely by Google Antigravity,” the company’s AI programming tool, which worked with a single former software engineer currently serving as a product manager. The company describes it as an unofficial companion to the broader XR Blocks framework designed for Android XR.
The project’s GitHub repository describes the library as an “alpha developer preview,” providing ready-to-use web packages, frameworks for Apple and Android devices, and the command line. Open Source For You noted that the project’s primary goal is to enable 3D content development without requiring complex modeling skills.
An analytical review of the strategic and economic dimensions of the event
Google’s open-source Three.Live library provides an innovative engineering solution to one of the most prominent gaps between the ability of large language models to write graphic code and running that code safely and efficiently on user devices. By handling a declarative subset of the Three.js API and executing generated code inside a sandboxed Web Worker stripped of network permissions or via an embedded Rust compiler on smartphones, the library neutralizes the security risks of untrusted code while converting scenes in fractions of a millisecond into native formats that run at maximum speed across Android, Apple, and web engines.
The development story of this library is no less compelling than its technical function. Written entirely by the Google Antigravity AI programming tool in collaboration with just a single product manager, it offers tangible, practical proof of AI programming tools’ ability to complete complex, multi-platform software projects at a compact size of just 45 kilobytes. With the addition of the XR.Live package to support hand tracking and prompt-based 3D generation, this initiative opens the door for developers to build augmented reality experiences and 3D content without needing complex traditional modeling expertise.
Frequently asked questions
Question: What is the function of the Three.Live library launched by Google?
Answer: It is an open-source library that converts AI-generated Three.js code into interactive 3D scenes and augmented reality experiences that run securely and quickly on the web, Android, and iOS.
Question: How does Three.Live ensure security and speed when running generated code?
Answer: It sandboxes code on the web while removing network connection functions, and utilizes an embedded Rust compiler on mobile devices that converts scenes in 2 to 5 milliseconds into native formats while enforcing execution time limits.
Question: How was the source code for the Three.Live library built and written?
Answer: Google explained that the library was written entirely by the Google Antigravity AI programming tool in collaboration with just a single product manager.
Question: What does the XR.Live companion package add for developers?
Answer: It adds support for hand-tracking gestures and the ability to generate 3D objects directly from voice or text prompts within extended reality sessions.