Introduction

Kiln is a WebGPU-native, out-of-core volume renderer for large volumetric datasets. It streams multi-gigabyte volumes over HTTP into a bounded GPU residency/atlas cache, resolving them through virtual-texture indirection.

Kiln streams OME-Zarr (NGFF v0.4/v0.5) datasets directly — single-channel and multichannel, up to 4 channels — as well as a compressed sharded binary format. Input can be uint8, uint16, or float32; uint16 and float32 are converted to r16float for GPU storage.

Where to next

• Quick start — get a viewer rendering a streamed volume in a few lines.
• Loading data — remote OME-Zarr, local files, and multichannel datasets.
• URL parameters — share and bookmark specific views.
• Gallery — live example datasets, each opening directly in the viewer.
• Architecture — how the virtual texturing and streaming systems work.