jetson-containers
A Docker container build system for local AI on NVIDIA Jetson, with CUDA support and packages for Ollama, llama.cpp, PyTorch and ROS.
jetson-containers is a Docker container build system for developers running local AI and robotics workloads on NVIDIA Jetson hardware. It supplies prebuilt images and lets you combine AI packages into custom containers, reducing the work of assembling compatible GPU software for JetPack/L4T.
Its main distinction is the ability to put tools such as ROS, PyTorch and Transformers in the same container. The compatibility helper selects an image for the device's JetPack/L4T environment, using a local image, downloading one from a registry or building one when needed. Containers run on your hardware; image downloads and package retrieval use external registries and servers.
The package collection covers LLM backends including Ollama, llama.cpp, vLLM and SGLang. For projects beyond text chat, it includes LLaVA and VILA for vision-language tasks, Whisper and Piper for speech, and ComfyUI for image generation. Robotics packages include ROS, LeRobot and OpenVLA, while FAISS and LlamaIndex support document retrieval and vector search.
You can rebuild the stack around different CUDA versions and choose versions of PyTorch, TensorRT and cuDNN. This matters when a workload needs a particular combination of GPU libraries. Package caching reduces repeated downloads and compilation, with an optional local PyPI and APT cache.
The project provides Ubuntu containers with CUDA support for JetPack systems. It also supports ARM SBSA hardware, including NVIDIA GH200 and GB200.
The project documents testing and support for JetPack 6.2 and JetPack 7. Match container images and CUDA dependencies to the JetPack release on your device.
