apple-container

Apple container

Reference scope: this imported guide mixes release-era and later branch documentation; its 1.0.0+ annotations are not verified introduction versions. Match the installed container --version and subcommand --help to the same upstream release tag before using flags or defaults. Current branch documentation can describe unreleased behavior. No container runtime was executed for this static review.

When to Use

• Use when building, running, or managing OCI/Linux containers on Apple-silicon macOS with Apple's open-source container CLI
• Use when you want lightweight per-container VMs instead of a Docker daemon
• Use when translating Docker-style workflows (build, run, exec, logs, networking) to Apple's container tooling

Apple's container is an open-source CLI for building, running, and managing OCI/Linux containers on Apple-silicon Macs. Each container runs inside its own lightweight virtual machine (backed by the Containerization framework and the Virtualization API). A shared per-user API server and helpers run via launchd, with a separate runtime helper per container. Images are standard OCI artifacts, so they interoperate with Docker registries and other OCI tooling. The CLI is deliberately Docker-like (container run, container build, and image ops under container image push/pull), but it is a distinct tool: do not assume Docker command paths, flags, defaults, or daemon behavior carry over (e.g. there is no container images/push/pull top-level command — image verbs live under container image).

Safety Gate

Container installation, service startup, image pulls, builds, runs, registry login, pushes, and resource cleanup change local or remote state. Keep the exact registry, mounts, ports, privileges and persistence impact within existing task authorization. Clarify missing destructive scope, sensitive mounts, credentials or external exposure before acting; do not ask again for actions already authorized.

Requirements

• Apple silicon only (M1 or later). Intel Macs are not supported.
• macOS 26 (Tahoe) is the officially supported target. The maintainers do not support older macOS and typically will not fix issues that can't be reproduced on 26. The binary still runs on macOS 15 (Sequoia) but with reduced networking: only the single default subnet is available, and the container network group and --network flag error out. macOS-26-gated features are called out throughout the reference files.
• Version: this skill documents the 1.0.0 release (the historical reference snapshot). The machine group, container cp, container export, container prune, container image prune, container registry list, and container system version were added in 1.0.0 (not in 0.7.1) — features that postdate 0.7.1 are flagged (1.0.0+) in the reference files. Run container --version and container <group> --help to see what your installed build supports.
• Install by downloading the signed .pkg installer from the project's GitHub releases (apple/container) and running it. See references/concepts.md for the full requirements/compatibility matrix and how the VM-per-container model works.

Setup

Install the signed package, then start the background services once:

1. Download the latest signed installer .pkg from the GitHub releases page.
2. Double-click the downloaded package and follow the prompts, entering your admin password so it can place files under /usr/local. For automated package installation, verify macOS installer syntax and package signature rather than inventing a container subcommand.
3. Start the services and confirm they are healthy:

# Start the container services (container-apiserver + helpers via launchd). On first run it
# offers to install the default Linux kernel — accept it, or start non-interactively with
# `--disable-kernel-install` and add a kernel later via `container system kernel set`.
container system start

# Verify services are healthy
container system status

container system start must have run before any container/image/build command works — a connection/XPC error can indicate stopped services, incompatible components, permissions or launch failures. Inspect container system status, installed versions and logs before deciding whether startup or repair is needed. Stop and deregister the launchd services with container system stop (which takes only -p/--prefix). The startup flags for container system start (-a/--app-root, --install-root, --log-root, --enable-kernel-install/--disable-kernel-install, --timeout) are in references/configuration.md.

Upgrade / downgrade / uninstall use helper scripts in /usr/local/bin (stop first with container system stop): update-container.sh (add -v <version> to pin a version), and uninstall-container.sh -d to remove user data or -k to keep it. Full recipes in references/workflows.md.

Command groups at a glance

Invoke everything as container <group> <subcommand>. Container-lifecycle verbs (run, create, start, stop, exec, logs, inspect, list/ls, delete/rm, kill, stats) and build are top-level; image operations like push, pull, and tag live under container image. Run container <group> --help for exact flags, or read references/commands.md for the exhaustive matrix.

Exact subcommand names, aliases, arguments, and flags for each group live in references/commands.md — consult it before running an unfamiliar command rather than guessing Docker-equivalent syntax.

Navigating this skill

Read the reference file that matches the task; do not guess flags or behavior.

• references/commands.md — exhaustive CLI reference: every command group, subcommand, alias, argument, and flag. Read this to construct any concrete container ... invocation, or to confirm a flag exists before using it.
• references/concepts.md — architecture (VM-per-container, Containerization framework), system requirements and macOS 15 vs 26 differences, networking model, per-container IPs, security model, and a Docker-vs-container comparison. Read this to explain how or why something works, or when a Docker mental model gives the wrong answer.
• references/configuration.md — the system service, config.toml / property model, default kernel, DNS domains, default registry, builder resources, and machine settings. Read this to change defaults, tune CPU/memory, point at a private registry, or manage the kernel.
• references/workflows.md — copy-pasteable task recipes (run an image, build & push, wire up local DNS, mount a volume, expose ports) and troubleshooting for common failures. Read this first when the user wants to accomplish a concrete end-to-end task.

Key rules

• This is not Docker. The CLI resembles Docker, but flags, defaults, and daemon behavior differ. Verify syntax in references/commands.md instead of assuming Docker equivalence.
• Check service state when runtime operations need it. Inspect container system status; start stopped services when the task authorizes it. Diagnose other failures rather than repeatedly starting or reinstalling.
• Images are standard OCI artifacts and interoperate with Docker registries and other OCI tools. Image references that omit a registry default to docker.io (configurable via the registry.domain property — see references/configuration.md).
• Each container gets its own IP address on its network (one lightweight VM per container). There is no shared Docker bridge; reach a container directly by its IP, or set up a local DNS domain (container system dns create ..., admin required) for name-based access.
• container network requires macOS 26. On macOS 15 only the single default subnet is available and the network command group is unavailable — see references/concepts.md.
• Use fully-qualified image references when precision matters (e.g. docker.io/library/alpine rather than bare alpine) to avoid ambiguity about the source registry.

Limitations

• Apple Container requires Apple silicon and has materially different support and networking behavior across macOS releases; verify the installed CLI version before relying on a flag.
• OCI images and registry content are third-party inputs. Inspect and trust the image source before pulling or running it.
• This skill does not make container workloads safe by default: mounts, published ports, privileged settings, registry credentials, and cleanup can expose or destroy data.
• Stop before uninstalling, pruning, deleting containers, volumes, or images, and require task authorization covering the exact destructive scope; obtain clarification when that scope is missing.

Operational boundaries

Use trusted pinned images and verify custom kernels, init images and plugin executables before use. Builds execute Dockerfile instructions; build args and printed environments can expose secrets. Use supported secret mounts rather than credential build args. Disable shell tracing when passing credentials, and prefer explicit HTTPS: historical auto scheme behavior may choose plaintext even on private networks. HTTP is only for an explicitly authorized disposable local test registry without reusable secrets.

Container machines weaken ordinary container defaults: the image controls init, expanded Linux capabilities, and potentially host home and SSH-agent access. Default examples disable the home mount; verify that flag on the installed release. Read-only home mounts still disclose keys and dotfiles. Agent forwarding allows guest processes to authenticate as the host user; do not forward it for untrusted workloads. Ordinary mounts, socket publishing and --ssh need the same explicit scope. VM isolation is not a guarantee against host compromise.

Prefer explicit loopback bindings and inspect actual listeners; do not assume an omitted host address is loopback-only. Before cleanup, inventory exact resources and preserve needed data. Stopping services interrupts other user workloads; builder deletion discards cache. Backup data and consult version-specific migration guidance before downgrading; keeping data with -k does not guarantee backward compatibility. Performance and journaling choices require workload/durability measurements, not fastest/safest guarantees.