vackrooms: An Open-Source Browser Backrooms Game Built with Claude and WebGPU

An Infinite Backrooms Game That Runs in Your Browser

Developer Pablo Stanley has released an open-source project called vackrooms: a Backrooms-style exploration game that runs directly in the browser, generates a new maze on every visit, and requires no download or installation. It's completely free, fully open-source — just open the link and you're in.

Backrooms has become a popular internet horror genre in recent years, defined by endless, unsettling expanses of empty space. vackrooms nails that atmosphere — players wander through Poolrooms, food courts, and streets, and may even encounter computers loading real websites over dial-up. And somewhere behind all those spaces, something is quietly following you.

The Backrooms originated in 2019 from a photo circulated on 4chan — an endless, empty office space with yellow carpet and flickering fluorescent lights. From it grew an entire internet horror mythology: when someone "noclips" out of the boundaries of the real world, they fall into these vacant, still, antiseptic-smelling spaces. The horror doesn't come from monsters, but from the psychological weight of "emptiness that shouldn't exist" and "the feeling that something, somewhere, is watching you." YouTuber Kane Pixels later brought the concept mainstream with a short film series, turning it into one of the defining IPs of contemporary internet creepypasta.

Tech Stack: Three.js and WebGPU Shaders

The project's technology choices are worth noting. The overall rendering is built on Three.js, one of the most mature 3D graphics libraries for the web. The visual effects layer uses vgpu to write WebGPU shaders (WGSL).

The developer specifically highlighted that one of vgpu's advantages is that it's "lightweight, and its documentation is written for AI agents." This means LLMs can directly parse the documentation spec and autonomously generate and validate WGSL shader code — without requiring constant human intervention and debugging.

This hints at a broader trend taking shape: toolchain documentation is starting to target AI rather than just humans. When a library's docs are sufficiently structured and its examples sufficiently clear, its "usability" for AI coding assistants improves dramatically, reducing the overall cognitive overhead of development.

WebGPU is the next-generation browser graphics API following WebGL, which began rolling out in major browsers in 2023. Unlike WebGL's older OpenGL ES-based architecture, WebGPU aligns with modern graphics APIs (Vulkan, Metal, DX12), offering lower driver overhead, compute shader support, and more explicit resource management. WGSL (WebGPU Shading Language) is the shader language designed specifically for WebGPU, with Rust-like syntax and an emphasis on type safety. The WebGPU ecosystem is still in its early stages — community examples and tooling are far less mature than WebGL's — which is precisely why "AI-friendly documentation" has become a key factor in library selection. When human reference material is scarce, whether an AI can independently understand the docs and generate correct code directly determines the ceiling of development efficiency.

Claude's Role in Game Development

The author openly states that Claude is very familiar with Three.js, which was itself a key reason for choosing that tech stack. Throughout vackrooms' development, Claude took on a substantial share of the actual coding work — from setting up Three.js scenes to writing WebGPU shaders using the vgpu documentation.

What's notable is the author's phrasing: Claude could "write and validate WGSL without me babysitting it." This suggests that with the right tools and documentation in place, AI is already capable of completing end-to-end tasks in relatively specialized graphics programming — not just producing scattered code snippets.

This "AI-led coding, human-steered direction" collaboration model is increasingly becoming a practical path for indie developers and small teams to rapidly produce playable demos. One person plus one AI assistant, and you can ship a complete, browser-playable game with procedurally generated levels and atmospheric rendering.

Procedural Generation and Immersive Details

The core mechanic of vackrooms is "generate a new maze on every visit." This procedural generation is a perfect fit for the Backrooms' premise of infinite space — players can never memorize the map, and they can never reach an end.

The project also includes a number of immersive details:

• Varied environments: Poolrooms, food courts, streets, and other distinct areas appear in rotation
• Dial-up computers: Machines that load real websites, weaving retro internet nostalgia into the horror atmosphere
• Pursuit mechanic: "Something following you" maintains a constant sense of tension

None of these elements are complex on their own, but together they form a complete exploration-horror experience — and demonstrate the efficiency gains AI-assisted development offers when rapidly iterating on content-driven gameplay.

Procedural Generation is a broad term in game development for using algorithms rather than manual design to create content. Common techniques include seed-based random maze algorithms (such as recursive backtracking and Prim's algorithm), noise function-driven terrain generation (Perlin/Simplex Noise), and grammar-based (L-System) building and vegetation generation. Its core value lies in producing near-infinite unique content at minimal storage and labor cost — Minecraft, No Man's Sky, and Spelunky all use it as a foundational gameplay pillar. For the Backrooms genre specifically, procedural generation carries extra narrative weight: a space that cannot be memorized or exhausted is itself a source of horror, and the algorithmic unpredictability maps directly onto that theme at a technical level.

Takeaways for Developers

vackrooms reads more like a technical proof-of-concept than a commercial release. Its value lies in demonstrating one viable path for web game development today:

1. Choose mature frameworks that AI knows well (like Three.js) — this significantly improves the quality of AI-generated code
2. Prioritize libraries with AI-friendly documentation (like vgpu) — this enables AI to complete more specialized tasks independently
3. Open-source + browser-playable lowers the barrier to both sharing and experiencing the work

For indie developers looking to experiment with AI-assisted development, this project offers a clear reference point: rather than asking AI to invent wheels from scratch, lay down familiar roads for it — mature frameworks, clear documentation, well-defined goals.

Play the game at vackrooms.vercel.app; the source code is hosted on GitHub (pablostanley/vackrooms) for developers who want to dig into the implementation details.