warpGet AccessGPU compute,
natively in the browser.
Warp is the TypeScript SDK for WebGPU. Run neural networks, physics simulations, and parallel compute directly on your users' GPUs. Zero servers. Near-zero latency. Ships to Chrome, Firefox, Edge, and Safari.
import { gpu } from '@warp/core'
// Define a compute kernel in WGSL
const vectorAdd = gpu.kernel`
@compute @workgroup_size(64)
fn main(@builtin(global_invocation_id) id: vec3u) {
out[id.x] = a[id.x] + b[id.x];
}
`
// Execute on the user's GPU
const result = await gpu.run(vectorAdd, {
a: new Float32Array([1, 2, 3, 4]),
b: new Float32Array([5, 6, 7, 8]),
})
// → Float32Array [6, 8, 10, 12]How it works
From kernel to production in minutes
Warp abstracts the WebGPU boilerplate so you write compute logic, not pipeline setup.
Write your kernel
Define your compute shader in WGSL using Warp's typed template tag. Full TypeScript autocompletion for bindings and inputs.
Dispatch on the GPU
Pass typed arrays as inputs. Warp handles buffer allocation, bind group setup, and pipeline compilation automatically.
Ship anywhere
Your kernel runs on the end user's GPU in Chrome, Firefox, Edge, or Safari. No GPU servers. No cold starts.
Features
Everything the browser GPU needs
TypeScript-First
Full type safety across kernel inputs, outputs, and buffer bindings. Catch mismatches at compile time.
Zero Infrastructure
Kernels execute on the end user's GPU. No GPU cloud instances, no billing surprises, no cold-start latency.
AI Inference Ready
Load ONNX models, run custom attention kernels, and chain inference pipelines entirely client-side.
WebGPU Native
Direct mapping to the WebGPU compute pipeline. No abstraction overhead, no magic — just clean ergonomics.
Dev Tools Built In
Real-time GPU timing, memory visualization, workgroup heat maps, and a shader debugger in the browser DevTools.
Cross-Browser Support
Automatic feature detection and graceful fallback to WebGL2 or WASM for browsers without WebGPU support.
Use cases
Built for every GPU workload
If it can run on a GPU, it can run in the browser with Warp.
LLM inference without the API bill
Run quantized language models, embeddings, and image diffusion directly in the browser. No API keys, no rate limits, no privacy tradeoffs.
Millions of data points at 60fps
Scientific charts, geospatial heatmaps, and particle systems that would bring a CPU to its knees — rendered in real time.
Simulations that run in the field
Physics engines, fluid dynamics, molecular modeling, and numerical solvers — deployed to any device, offline-capable.
GPU-powered creative tools
Real-time audio processing, procedural generation, image filters, and shader art — running instantly with no install.
Start shipping GPU-powered apps today.
Free to start. No GPU infrastructure required. Your users bring the hardware.