One model. Many languages.
A shared tokenizer and a small neural network read the structure around each line. The model chooses the space between logical groups, without a separate formatting rulebook for every language.
Small model. A little more space.
Give your code
room to breathe.
A tiny model that puts blank lines where your ideas pause.
Same code. Same indentation. A calmer read.
~52 KB shared model
x languages to explore
100% local processing
[ live demos ]
Edit a little. See the space.01 Original / editable
02 Spaced
Code stays in your browser. Spacing & highlighting use WebGPU.
Your code stays yours.
Only blank lines change. Nonblank content and indentation stay intact, with protected multiline regions preserved. This is a spacing tool; it does not wrap lines, sort imports, or replace your formatter.
Zig at the core.
The native CLI uses ggml with GPU acceleration when available. In your browser, WebGPU runs the spacing model while Zig handles code analysis and rendering through WebAssembly. No server, account, or API key required.
[ bring it into your workflow ]
A little space.
A small API.
Initialize once. Format as often as you need.
Read the documentation ↗// npm i gpu-code-spacer
import { createSpacer } from 'gpu-code-spacer';
const spacer = await createSpacer();
const { text } = await spacer.format(source);
// Same code, with room to breathe.