Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.
SimReady assets (NVIDIA)
NVIDIA's SimReady assets are a few thousand simulation-ready props (industrial, hospital, residential, lighting, OEM equipment) published as OpenUSD under CC BY 4.0. three.ws converts every asset to a web GLB and serves them from the object library and the asset catalog (GET /api/catalog, search_catalog, npx @three-ws/assets add), with the credit line attached to every entry.
Licensing
CC BY 4.0: free for commercial use with attribution. Each entry in the catalog carries license: "CC-BY-4.0" and attribution: "NVIDIA SimReady Assets, CC BY 4.0, https://huggingface.co/datasets/nvidia/simready-assets". Show that line wherever you use the model. The GLBs are modified derivatives (converted from USD, textures downsized, geometry decimated to at most 80,000 triangles); CC BY requires saying so, and the originals are linked from the credit.
How an asset is converted
- scripts/ingest-simready.py lists the dataset through the Hugging Face API and, per asset, downloads only the USD layers and the textures the materials reference (never the
.thumbsorsource_assetscopies, which would triple the transfer). - scripts/simready-to-glb.py opens the root USD with instance proxies, bakes every visible mesh into world space (Z-up to Y-up, meters), triangulates, splits face-varying normals and UVs into glTF vertices, maps each material to glTF PBR (albedo, ORM packed as occlusion + metallic-roughness, tangent-space normal), bakes the
UsdTransform2dtiling into the UVs, and writes 1024px JPEG textures. - scripts/slim-glb.mjs brings the GLB under a triangle budget (default 80,000) with meshopt simplification. The output stays plain glTF 2.0, so a bare
GLTFLoaderor<model-viewer>loads it. - The dataset's own thumbnail becomes a 512px WebP poster.
- The entry lands in
public/objects/simready/catalog.json(same schema as the Poly Haven catalog, plustriangles,materials,size_m,attribution). The run is resumable: finished assets are skipped,--retry-failedretries the rest.
pip install usd-core numpy Pillow pygltflib
python3 scripts/ingest-simready.py --workers 12 # GLBs and posters into ~/simready-out
node scripts/publish-simready.mjs --dry-run # then without --dry-run to upload to R2
node scripts/build-object-library.mjs # publish the manifest the API serves
Physics payloads (PhysX, MuJoCo, Newton variants) are not carried into the GLB; it is a visual asset. Use the original USD from the dataset for simulation.