Manycore Tech Open-Sources Aholo Viewer for Large Scale 3DGS Scenes

Manycore Tech has released Aholo Viewer as an MIT licensed open source project. It’s a TypeScript based 3D Gaussian Splatting rendering engine designed to stream and display city scale 3DGS scenes in a web browser without any client installation.

Aholo Viewer addresses this with a chunk level LOD streaming architecture, prioritized by frustum visibility, fetched in batches, and the first usable view of a scene loads within approximately 10 seconds even for captures at that scale.

Format support covers ply, spz, sog, splat, lcc, and ksplat. The engine ships with three rendering preset profiles, quality-first, performance-first, and extreme-performance, with granular parameter controls for tuning the balance between visual fidelity and frame rate across device classes. GPU aware optimizations are applied throughout loading, scheduling, and the render loop.

Aholo Viewer also includes a physics collision system, based at least in part on PlayCanvas’s Splat-Transform. 3DGS scenes can be converted into queryable voxel colliders, with mesh collision data also accepted as input. The system supports realtime ray, capsule, ground, and wall collision queries, and the same collision data can be shared across walk mode, third-person camera obstacle avoidance, and area limits. That makes Aholo Viewer useful for interactive spatial experiences built on captured environments, not just visualization.

The project is structured as a monorepo with a TypeScript renderer package, an Astro-based documentation site, and a browser playground with interactive examples. A cloud rendering mode, co-rendering 3DGS and high-fidelity OpenUSD meshes in the same frame, with cloud-side rendering keeping lower-spec devices smooth, is listed as coming soon. The viewer is compatible with smartphones, PCs, and VR devices. How much of this is PlayCanvas under the hood is yet to be determined, but it’s exciting nonetheless to see Aholo Viewer released.

Manycore Tech is the Beijing based spatial AI company behind the SpatialLM series of 3D scene understanding models, framed the release as foundational infrastructure for an emerging 3D internet ecosystem. The viewer is available at aholojs.dev, with source code on GitHub.