vgpu is a TypeScript library engineered for building WebGPU applications with a single, composable API. Designed to support modern agent-driven development workflows, it enables developers to write modular WGSL shaders that function consistently across diverse environments, including interactive web canvases, headless Node.js processes, and automated CI/CD testing pipelines. The library streamlines the complex WebGPU abstraction layer, offering a unified approach to rendering that works for images, videos, and complex real-time graphics.

Functionality centers on providing a streamlined interface for WebGPU, allowing developers to manage GPU contexts, effects, and passes with minimal boilerplate. It handles the lifecycle of shaders from compilation through execution, ensuring that the same code can run in a browser or on a headless server. The platform includes a suite of tools for diagnostic, validation, and documentation, ensuring reliable performance across different hardware targets.

Some of the key features are:

• Modular WGSL: Write reusable shader code as a module graph that resolves dependencies and removes unused declarations at build time.
• Unified Rendering: Use the same shader logic for browser-based interactive canvas rendering, headless Node.js automation, or image and video exports.
• Agent Support: Built-in support for coding agents via a dedicated CLI, MCP connectivity, and comprehensive API diagnostics.
• Performance Optimization: Includes features like render bundles to record draws once and replay them, reducing re-encoding overhead in every frame.
• Compute Interoperability: Seamlessly run machine learning models, such as those using ONNX Runtime Web, alongside vgpu on a shared GPU device.

Operationally, vgpu is designed to be initialized via a central init() call, which establishes the necessary GPU context. Developers then define effects, targets, and frames to manage the rendering pipeline. The library's focus on modularity allows for chaining effects by binding outputs to inputs, enabling advanced rendering techniques like HDR bloom, raymarching, and screen-space reflections. The CLI provides a primary entry point for developers and agents to interact with documentation, validation tools, and runtime repair features.

Some common use cases include:

• Web Visualizations: Creating high-performance, interactive 3D graphics for web applications.
• Automated Testing: Running headless GPU render tests in CI pipelines to verify shader output through snapshot comparison.
• Creative Media: Converting shaders into pre-rendered images, high-frame-rate videos, or dynamic content.
• Machine Learning Integration: Using the GPU to power both ML model inference and real-time visualization of model outputs.