vgpu
WebGPU library for agents: one typed API for browser canvases, headless Node, and serverless, with CLI discovery, MCP, and copy-ready examples.
What's interesting
- One typed API across browser canvas, headless Node, and tests
- CLI and MCP paths for agent-driven discovery
- Small bundles: fullscreen effects around 25 KB gzipped
Use in your project
Install
pnpm add vgpu- Skim examples with `npx vgpu examples`.
- Port the closest example to your scene.
- Target ~25 KB gzipped if this ships on a marketing page.
Prompt your agent
From AI UX Playground: https://aiuxplayground.com/playground/vgpu vgpu by Vercel Labs From Vercel Labs. A small TypeScript WebGPU stack with typed WGSL imports and a single Gpu handle—same code in the browser, headless Node (Dawn), and a deterministic mock for tests. Bundles stay tiny; a fullscreen effect ships around 25 KB gzipped. Agents can load agents.md and llms.txt, browse examples with `npx vgpu examples`, or connect to the hosted MCP at /api/mcp. Open source at vercel-labs/vgpu; install with `pnpm add vgpu`. Use vgpu for our [hero | loader | background] effect. Port the closest example to our React app. Browser canvas first. Name the WGSL files to touch and how you will test. Keep bundle size in view.
