envd

Why we included this project

Hand-maintaining Dockerfiles and CUDA/Python dependency scripts gets old fast, and envd offers a different route. Instead of writing brittle container instructions, you declare the packages and tooling you want in a build.envd file and run envd up. It builds an isolated, OCI-compatible environment that takes care of conda, Python packages, shells, and Jupyter setup. Because it leans on BuildKit for caching and remote builds, recreating an environment is quicker and wastes less than a fresh Docker build. The same environment runs locally or can be pushed to a registry and pulled into Kubernetes, which helps teams keep setups consistent across laptops and shared infra. This is a developer-experience tool for reproducible ML environments, not a training or serving framework; reach for it when environment drift is the pain point.