A GPU UI framework for Zig with zero dependencies — and a CLAUDE.md in the repo

Gooey promises GPU-rendered native UI on Metal, Vulkan and WebGPU with no external Zig packages. The thread spotted the CLAUDE.md, did the math on 200,000 lines in three months, and stopped being excited.

A zero-dependency GPU UI toolkit for Zig

Gooey GitHub repository card
Gooey GitHub repository card

Gooey is a hybrid immediate/retained-mode UI framework written in Zig, pitched as a way to build fast, GPU-rendered apps across macOS, Linux and the browser. It hit the front page on the strength of that promise, and the thread opens enthusiastic before the comments do some forensics on how it was built.

Metal, Vulkan, WebGPU and "zero dependencies"

The README's claims are concrete: native graphics per platform — Metal on macOS, direct Vulkan with GLSL shaders on Linux, WebGPU on the web — with GPU-accelerated text, SVG and image atlasing plus MSAA anti-aliasing. It offers declarative, flexbox-style layout, an animation system with easing and animateOn triggers, retained widgets like TextInput and scrollable containers, native file dialogs, clipboard, drag-and-drop, IME, and accessibility hooks for VoiceOver, Orca and ARIA. The selling point is "zero dependencies" — no external Zig packages required, though system libraries like FreeType and HarfBuzz are still needed. It's explicitly early: the API is evolving, WASM builds are temporarily deferred over a Zig 0.16 incompatibility, and there's a single global font for now.

The CLAUDE.md tell, and the lines-per-day math

The correct image of the repo got assembled by a commenter who read past the feature list:

"It's not particularly revelatory to point out that this project has been generated largely by LLM (claude most likely, given the CLAUDE.md in the repo) … I fear that I'm in for a period of lamentation as we get wave after wave of promising sounding developments, but where the reality is low quality, LLM generated crap." — a commenter who inspected the repo's docs

Another did the arithmetic and questioned whether anyone can actually own code produced at that rate:

"200000 lines of code addition in 3 months, basically 2000 lines per day. I don't think any human brain is capable to handle such cognitive complexity, unless the code is all plain data get/setters." — a skeptic running the numbers

The sharpest reversal is who showed up to redirect the audience. Rather than dunk, he simply named the human-built option:

"Another Zig GUI project that people might be interested in is DVUI" — the VP of community at the Zig Software Foundation

There was even a naming collision: the author of a different Gooey, a Go/WebAssembly GUI framework, turned up with a "Sadface :-(".

Excitement is cheap; an audit is not

The Zig ecosystem genuinely wants more native UI, and Gooey's feature checklist reads like exactly what people ask for. But a GPU framework is the kind of thing you have to trust at the memory-safety layer, and Zig offers no automatic guarantees there. When the provenance is 2,000 machine-written lines a day and the headline artifact is a CLAUDE.md, the thread's instinct was the conservative one: point newcomers at the project a person actually maintained.

Get breakdowns like this every day

The top Hacker News stories, unpacked — by email, Telegram, or RSS.

Next story · 4 of 19 Read more
🚀Tech

Google's JPEG XL victory lap, complete with an AI-reconstructed whiteboard scene

Google traces JPEG XL back through a decade of open-source experiments — WebP, Guetzli, Brunsli, PIK, FUIF. The thread mostly wanted to talk about the AI-generated photo of two engineers at a whiteboard, and the company that once tried to kill the format writing its origin story.

★ 88 · 💬 47 · opensource.googleblog.com · 4 min