Skip to main content

​
pbctl

pbctl is a fast, reproducible package manager for NVIDIA Jetson, built on the conda ecosystem. It provides workspace management, dependency resolution, task running, and environment isolation — purpose-built for edge AI and embedded Linux development on Jetson hardware.
pbctl currently supports Jetson Orin Nano with JetPack 6. Support for additional Jetson modules and JetPack versions is coming soon.

​
Highlights

Jetson-Aware

Detects your Jetson hardware, L4T version, and CUDA capabilities automatically. Resolves packages compatible with your device.

Reproducibility

Isolated environments with lockfiles built-in. Share identical setups across your Jetson fleet.

Tasks

Built-in cross-platform task runner with dependency support. Automate build, test, and deploy pipelines.

Conda Ecosystem

Access over 30,000 packages from conda-forge and other conda channels, including CUDA-accelerated libraries.

Multiple Environments

Compose multiple environments in one manifest — separate inference, training, and development dependencies.

Global Tools

Install global CLI tools in isolated environments. A modern replacement for apt on your Jetson.

​
Quick Demo

Detect your Jetson hardware: 
pbctl probe
Set up a workspace for ML development: 
pbctl init my-jetson-app
cd my-jetson-app
pbctl add python pytorch cuda-toolkit opencv
pbctl run python -c "import torch; print(f'CUDA available: {torch.cuda.is_available()}')"
Install tools globally: 
pbctl global install gh ripgrep btop

​
Available Software

pbctl defaults to conda-forge, the largest community-driven conda package repository with over 30,000 packages. Key packages for Jetson development:
  • ML/AI: pytorch, tensorflow, onnxruntime, scikit-learn, numpy, pandas
  • Computer Vision: opencv, pillow, torchvision
  • CUDA: cuda-toolkit, cudnn, cutensor
  • Python: python, jupyterlab, ipython
  • C/C++: cmake, ninja, gcc, boost
  • CLI Tools: git, gh, ripgrep, make, curl
Browse thousands more on conda-forge.org.