Every container runtime on macOS runs a Linux VM. Linux doesn't have Metal support, so you can't use MLX, MPS, or other Metal-accelerated frameworks inside a container. MetalBox solves this by running workloads as native macOS processes with Docker-like resource management: * Hard memory limits (RSS watchdog that kills and restarts processes when limits are exceeded) * Metal/MLX GPU memory caps (automatically injected via a wrapper) * Filesystem and network sandboxing using macOS sandbox-exec
Hey PH! I built MetalBox because I needed to run ML inference pipelines on a Mac Mini with Apple Silicon. Docker can't access Metal GPUs, and running natively means giving up resource limits.
MetalBox provides container-like management for native macOS processes. It's a single Go binary that handles process lifecycle, memory enforcement, GPU memory caps, health checks, and isolation, with a web dashboard for monitoring.
It's open source (MIT), installs from PyPI, and works with any workload, not just ML. If you run anything on macOS that needs resource limits or lifecycle management, give it a try.
Report
No reviews yetBe the first to leave a review for MetalBox