Experimental In-Browser Sandboxes
Developer Playgrounds
Interactive sandbox environments to evaluate emerging client-side technologies, benchmark on-device machine learning models, and test hardware execution limits with zero server compute.
Flagship ToolWebGPU Accelerated
Local WebGPU LLM Benchmark
Run quantized open language models (SmolLM2, Qwen2.5-Coder, Llama 3.2) entirely in your browser. Measure pure decode tokens/second and time-to-first-token directly on your GPU without sending a single prompt to any backend server.
SmolLM2 360M(376 MB VRAM)Qwen2.5-Coder 1.5B(1.6 GB VRAM)Llama 3.2 1B(880 MB VRAM)
Privacy
100% In-Browser
Telemetry
Live Tok/s & TTFT
Runtimes
WebLLM & TF.js
STANDARDIZED TELEMETRYLIVE GPU PROBE
Decode Speed45 - 90 tok/s
Time to First Token120 - 350 ms
Standard Suite3 Prompts x 3 Runs
Community RanksPublic Leaderboard
Hardware acceleration via WebGPU shaders
Launch Playground
Zero Server UploadWebGPU Accelerated
Private In-Browser Vision AI
Remove backgrounds, upscale images 4x, and generate depth maps directly in your browser. Pixel data never leaves your device — verifiable in your browser network tab.
Background Removal(RMBG-1.4 · ~45 MB)Upscale 4x(Swin2SR · ~90 MB Tiled)Depth Map(Depth Anything V2 · ~50 MB)
Privacy
100% In-Browser
Cutouts
Full-Res Masking
Upscaling
Seam-Blended Tiles
NETWORK VERIFICATIONZERO UPLOADS
Image Transmission0 bytes sent
Weights SourceHuggingFace CDN
Execution EngineWebGPU / ONNX
Offline ModeSupported When Cached
Client-side computer vision via Transformers.js
Launch Vision Tools
