macOS production¶
Apple Silicon (and Intel) Macs are a supported deployment shape for edge or small-scale production — e.g. Mac Mini / Studio with Metal / Neural Engine for inference-heavy workloads.
Hardware acceleration¶
| Backend | When used | Config |
|---|---|---|
| Metal | GPU-accelerated inference, embeddings, compute-heavy paths | ASTRA_USE_METAL=true |
| Neural Engine (ANE) | When exposed via Core ML or stable framework APIs | Detected automatically; explicit fallback to Metal or CPU |
| CPU | Default fallback when accelerators unavailable or disabled | Always available |
Detection via runtime.GOOS == "darwin" or build tags. Backend selection is explicit; fallback always to a working path. ASTRA_USE_CUDA is ignored on macOS.
Native binaries built for darwin/arm64 and darwin/amd64 — no emulation on Apple Silicon.
Operations¶
Services can run under launchd, a process supervisor, or your org’s standard macOS server practice. Infrastructure (Postgres, Redis, etc.) may still run in containers on the same host.
Note
Binary build commands, plist paths, and compose files are not published on this wiki — use the private repo.