Welcome to Field·4

Field·4 is an experimental, live-coding environment for new media art, VR, music and data visualization.

If you want to get your hands on it, install Field and make a first moving drawing. If you'd rather look around first, take the Tour, or follow an idea from one medium into another in the Long Way Round. Looking for a particular thing? Start with what's already in the box. Been here before? What's new lists the features and pages added since, newest first. The source code, and the place to report a bug, is Field·4 on GitHub.

Field·4 runs on macOS, Windows and Linux. It draws desktop graphics with Vulkan, so you can write shaders and render meshes and particles from a box. For VR, you can drive a headset attached to your computer through OpenXR, where the platform supports it, or send a scene to a headset or phone browser through WebXR. There is also a small native front end for the Apple Vision Pro.

Field's IDE can also run inside a headset, through OpenXR. On Quest 3 with WiVRn, edit and run code on the live canvas and in Monaco using a keyboard and mouse connected to the headset. Controllers point, click, scroll, position and resize the browser panel. The same document remains available to another connected browser.

The primary programming language for this version of Field is Kotlin on the desktop, which we've gone to some effort to bend into a language and runtime that is both a joy to experiment in and fast enough to make real things with. Kotlin, together with Vulkan GLSL, is the primary combination of languages used daily in Field. In addition, there are bindings for other languages, including Clojure (as a JVM language), Python (via the JEP project), Lua (via LuaJIT), JavaScript that executes remotely in the browser or on mobile devices for Three.js-based WebXR, and a great many other languages. These extensions usually have autocomplete, AI assistance, live preview and some kind of interoperability with the rest of Field's supported languages.

Field·4 has, and will continue to have, enthusiastic support for AI-forward coding practices, offering chat grounded in documentation, source code, the current document, and individual code selections, as well as tools that let an agent drive Field the way you do: create and connect boxes, run their code, read the errors and output, and take screenshots of the result. If that isn't your cup of tea, then this project might not be the coziest spot for you to curl up in. The very first version of Field (then called Fluid, in 2001) was itself built to build, interact with, train and play with software agents.

Field supports "bring your own key" OpenAI / Anthropic integration, as well as local models. The agent transcript gallery shows what that tooling can make: documents that agents have built with limited supervision from a blank canvas, each one openable in Field. Long Way Round walks through agent-generated projects that connect several of Field's tools, from geometry and simulation to rendered images.

Although Field began as a research project at MIT's Media Lab in 2001, this new version is currently in "early preview"—your help in developing it is deeply appreciated. It's all new code, incubated during UChicago's winter 2026 VR Production class after a successful 2025 trial in the Augmented Reality in the Garden and Remaking Movies classes.

Starting places

In a hurry? Field opens on a Welcome page; the first drawing walks you from its Start here: a moving drawing button to a drawing you change, run, stop and reopen.

If you aren't reading this documentation inside Field·4 itself, then you'll want to read the Installation instructions or, if you are window-shopping, perhaps take a tour.

Alternatively, the Survival Guide is a good cheat sheet.

If you do have Field open on your machine while reading these docs, many of the read-only examples will open directly in Field when you click the "Open in Field" button.

Field's code is AGPL-3.0-or-later; your code, AI and local history explains the local services and optional external providers.

Agents wrote much of this site's text, using a harness that is part of Field. This site is part of Field's build process: almost every one of the 2000 screenshots here is captured by running the code beside it — either the code fence visible above it or hidden in the source for the website. Even the video behind this page is automatically generated (and regenerated) too. Field has lots of other tests and build infrastructure, but this exercise in agent-driven, literate programming is a large part of its test suite.