Skip to content

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.

901 B
Raw Blame
# Pins that must survive every transitive dependency resolution in this image.
#
# audio-separator brings its own torch and numpy requirements. Left
# unconstrained, pip happily replaces the cu124 torch 2.6.0 wheel installed from
# download.pytorch.org with a CPU-only build from PyPI, and the image then fails
# at inference with "Torch not compiled with CUDA enabled" after a 40-minute
# build. It also upgrades numpy past 1.x, which breaks the compiled extensions
# in av and opencv.
#
# The Dockerfile asserts these versions after installing, so drift fails the
# build instead of the first job.
torch==2.6.0
numpy==1.26.4
transformers==4.41.0
diffusers==0.35.1

# diffusers 0.35.1 requires huggingface-hub>=0.34.0. Nothing in the image imports
# the hub directly, but transformers and diffusers both do, so keep any late
# install from resolving it back under that floor.
huggingface-hub>=0.34.0