Morbstack vs. the field.

1. Purpose, method, and an honesty statement

This page compares Morbstack against six things developers actually reach for on macOS: Docker Desktop, OrbStack, Podman Desktop (+ podman machine), Rancher Desktop, Colima (+ Lima), and Apple's open-source container project. It exists to answer, as precisely as the evidence allows, "how does Morbstack actually differ, and where is it actually worse."

Evidence tiers, and what each is worth

• Morbstack's own numbers come from exactly four places in this repository — parity.md, roadmap.md, compat.md, and amd64.md/k8s.md — and nowhere else. No number about Morbstack below was generated for this page. Where a figure is a target rather than a measurement, it is labeled as such every time it appears.
• Local, first-hand inspection covers Docker Desktop, which was installed on the machine this document was written on (bundle timestamps dated December 2024 — an older installed copy, not necessarily the current shipping release). It was inspected read-only: bundle contents, codesign, otool, nm, and a snapshot of already-running processes, nothing started, stopped, or reconfigured. OrbStack, Podman, Rancher Desktop, Colima, and Apple's container are not installed on that machine — every claim about them below is web-sourced, not independently reproduced, and marked accordingly.
• Vendor and third-party claims from the web are quoted and attributed, never presented as Morbstack-verified fact.
• Anywhere research was inconclusive, this document says "could not verify" rather than filling the gap from memory.

2. Architecture

Reading this table honestly: Morbstack, Docker Desktop, OrbStack, Podman machine, Rancher Desktop, and Colima all converge on the same basic shape — one shared Linux VM, Virtualization.framework (or Lima's abstraction over it) as the hypervisor, VirtioFS-family bind mounts. Morbstack's specific bet inside that shape is being the only one of the six with zero modification to the actual container engine verified end-to-end (parity.md #1-#4, #12-#14) — Podman and Apple's container aren't running Docker's engine at all, and OrbStack's engine internals are unverifiable because the product is closed source. Apple's container is architecturally the odd one out entirely: per-container micro-VMs trade away exactly the shared-kernel properties (Compose networks, --network container:x, shared page cache) that architecture.md explains Morbstack deliberately kept a single shared VM to preserve.

Local inspection: what Docker Desktop actually is on this Mac

Commands run, 2026-08-02, entirely read-only, Docker Desktop's own state never touched.

$ du -sh /Applications/Docker.app
1.7G    /Applications/Docker.app

$ du -sh /Applications/Docker.app/Contents/*/
1.1G    /Applications/Docker.app/Contents/Resources/
624M    /Applications/Docker.app/Contents/MacOS/
7.3M    /Applications/Docker.app/Contents/Library/

$ du -sh "/Applications/Docker.app/Contents/MacOS/Docker Desktop.app"
261M    .../Docker Desktop.app          # the Electron UI shell alone

$ du -sh /Applications/Docker.app/Contents/Resources/*/
558M    .../linuxkit/      # boot.img + kernel — the guest VM image
368M    .../cli-plugins/   # buildx, compose, etc., bundled
161M    .../bin/           # docker CLI, kubectl, hub-tool, credential helpers

$ codesign -dv --entitlements - "/Applications/Docker.app/Contents/MacOS/Docker Desktop.app"
Identifier=com.electron.dockerdesktop
...
com.apple.security.cs.allow-jit: true
com.apple.security.cs.allow-unsigned-executable-memory: true

The com.electron.dockerdesktop bundle identifier, plus an on-disk Electron Framework.framework and @bugsnag/plugin-electron-app inside app.asar.unpacked, confirm the management UI is Electron, not native.

$ codesign -dv --entitlements - /Applications/Docker.app/Contents/MacOS/com.docker.backend
com.apple.security.hypervisor: true
com.apple.security.virtualization: true

$ otool -L /Applications/Docker.app/Contents/MacOS/com.docker.backend | grep -i -E "virtualization|hypervisor"
/System/Library/Frameworks/Virtualization.framework/Versions/A/Virtualization ...
/System/Library/Frameworks/Hypervisor.framework/Versions/A/Hypervisor ...

$ nm -g /Applications/Docker.app/Contents/MacOS/com.docker.virtualization | grep -c VZ
307

This settles a question worth settling with evidence rather than reputation: on this machine, Docker Desktop's backend links and uses Apple's Virtualization.framework and Hypervisor.framework — the same family of API Morbstack uses. The 307 VZ-prefixed symbols include VZLinuxRosettaDirectoryShare, confirming Rosetta support at the framework level, consistent with the bundled qemu-system-aarch64/qemu-img also present. Docker Desktop was running at inspection time: 8 docker/com.docker.* processes, combined RSS 89.3 MB — reported here only as an observed data point, not a performance claim about either product. Notably, the running process was launched with --analytics-enabled=true — telemetry is on by default on this installed copy.

3. Licensing and cost

This is Morbstack's most precise, least hype-dependent differentiator, so it gets stated exactly rather than rhetorically.

Morbstack, Podman Desktop, Rancher Desktop, Colima, and Apple's container are all genuinely free-forever open source with no commercial trigger — on licensing terms alone, Morbstack is tied with four established projects, not uniquely differentiated from all of them. Its real differentiation is narrower: it is the only one of the six simultaneously (a) free with no commercial-use asterisk, (b) built around unmodified, off-the-shelf upstream dockerd/containerd — no downstream engine patch, published ports made reachable by a small Morbstack userland proxy that dockerd execs through its own --userland-proxy-path flag — rather than a different or closed engine, and (c) aimed at drop-in Docker Desktop replacement rather than a CLI-only tool or a different API surface entirely. Whether it achieves that combination today is a feature-completeness and quality question, covered honestly in §5 and §6.

4. Feature matrix

Morbstack rows are marked directly from the 29-check parity.md audit and the compatibility contract — PASS / PARTIAL / FAIL where parity.md tested it explicitly, not yet built where roadmap.md places it at M1 or later, and not planned where compat.md states it as an explicit non-goal. Competitor cells are sourced from vendor documentation, cited inline; where a claim could not be confirmed it says so.

5. Performance: claims vs. measured reality

Kept in three explicitly separate buckets, because blurring them is the single easiest way to lose a skeptical reader's trust.

(a) Vendor claims — quoted and attributed, not verified here

• OrbStack states, on its own comparison page, "fast startup," "low CPU usage," "low power usage," and "memory on demand" relative to Docker Desktop — presented as a checkmark comparison, not first-party numbers on that page itself (OrbStack vs. Docker Desktop). Third-party writeups report ~2s OrbStack startup vs. 20-30s for Docker Desktop, and roughly 200MB vs. 4GB idle memory — third-party numbers, not independently reproduced here (sliplane.io comparison).
• Apple's container claims its per-container VMs achieve "sub-second startup times through careful kernel optimization," per third-party technical summaries — could not independently verify the underlying Apple source.
• No performance claims from Podman Desktop, Rancher Desktop, or Colima were found stated as vendor marketing in this research pass.

(b) Morbstack's own measured numbers — with conditions, not marketing

Real, timed, end-to-end runs on one development machine, Apple silicon — from architecture.md, parity.md, amd64.md, and k8s.md — not the CI-gated morb bench numbers for M2+.

• Cold, socket-activated docker run --rm hello-world: 2.09s wall clock, guest MRB0 control socket ready 1.69-2.01s after VM bring-up.
• morb stop: 0.25-0.40s. A subsequent cold boot to a responsive docker ps: 1.69-1.73s, repeatable with zero ERROR/WARN log lines.
• First-boot disk formatting (mkfs.ext4): 22ms; mount 56ms after that; a real alpine registry pull: 1.95s.
• Idle-auto-suspend cycle: guest powered off in 384ms after the idle timer fired; next socket-activated command woke a fresh VM in 2.13s. This is a graceful stop, not a true suspend-to-zero snapshot/restore — Virtualization.framework's restoreMachineStateFrom does not work for Morbstack's direct-kernel-boot guest on this development host, a documented API limitation, not a Morbstack bug.
• amd64-via-Rosetta slowdown, same guest and session: no measurable container-start penalty; general-purpose compute ~1.8-2x slower; SHA-256 at 16KB blocks ~5.1x slower; RSA-2048 sign ~3.8x slower, verify ~2.3x slower.
• Kubernetes (morb k8s): cold enable-to-Ready 8.3s with a full payload transfer; ~4s on a subsequent enable with the payload already present. Idle cost with the cluster on and nothing deployed: guest memory rose from ~137MB to ~617MB, host VM-process CPU rose from ~0.7% to ~20-23%.

(c) Morbstack's published targets — not yet measured, do not read as results

docs/roadmap.md's public performance target table commits Morbstack to publishing accurate status against these from M2 onward via morb bench — goals the project is answerable to, explicitly not claims about the current build.

Note the overlap and the gap between (b) and (c): the cold boot measurements in (b) (2.09s, 1.69-1.73s repeat boots) already land close to the ≤2.0s target in (c) — but "resume ≤500ms" is a target with no real suspend-to-zero implementation to measure yet, and idle CPU/wakeups/RSS have no reported measurement at all as of this writing. Treat the closeness of the boot numbers as encouraging, not as evidence the harder targets are anywhere near proven.

6. Where Morbstack is worse — the honest section

This section is deliberately at least as detailed as §§2-5 combined are flattering. Everything here is either a documented FAIL/PARTIAL from the real 29-check audit, an explicit not yet built item from the roadmap, an explicit non-goal, or a structural maturity gap that is simply true of a one-maintainer, pre-alpha project regardless of what its docs say.

Things that are broken, absent, or still lack release-grade live evidence

Things competitors have that Morbstack has chosen never to build

Per the explicit non-goals in docs/compat.md — stated up front so readers don't mistake them for gaps that will eventually close:

Maturity gaps that are just true, regardless of documentation quality

Anyone reading this page and concluding "Morbstack is strictly better" has not read it carefully enough — that conclusion isn't supported by the project's own documentation, and this section exists so it doesn't need to be argued with.

7. Who should use what

A recommendation section that doesn't send every reader to Morbstack is the only kind worth writing.

Use Docker Desktop if:

you need Windows container support, GPU passthrough that's actually documented and supported, Docker Desktop Extensions, or you're inside an organization already paying for it and the $9-24/user/month is a non-issue. It is also, simply, the thing with the largest install base and the least chance of a workflow-specific surprise.

Use OrbStack if:

you're under the $10k/yr or personal-use threshold, want a closed-source but reportedly fast and well-integrated native experience today, and don't need the source to be inspectable or the product free at any company size.

Use Podman Desktop + podman machine if:

you specifically want a daemonless/rootless model, need Kubernetes-native tooling, or your organization has policy reasons to avoid a Docker-Engine-shaped daemon — accepting you're not running the actual Docker Engine, only something CLI-compatible with it.

Use Rancher Desktop if:

Kubernetes is the primary workload, not an afterthought, and you want first-class containerd/nerdctl alongside Moby with a choice between them.

Use Colima (+ Lima) if:

you don't want or need a GUI at all, you're comfortable at the command line, and you want the lightest possible free and open-source Docker-API-compatible VM with maximum backend flexibility.

Use Apple's container if:

per-container VM-grade isolation matters more to you than Docker ecosystem compatibility, you're on macOS 26 with Apple silicon, and you're comfortable with a CLI that doesn't speak the Docker API and has no Compose story today.

Use Morbstack if:

you want a free-forever, Apache-2.0, source-visible Docker Desktop replacement built around an unmodified upstream Docker Engine (published ports served through dockerd's own stock --userland-proxy-path hook, not an engine patch) — and you are comfortable being an early adopter of pre-alpha software with one maintainer, no security audit, missing host.docker.internal/zero-config discovery/buildx-out-of-the-box, and the specific /tmp bind-mount and inotify gotchas documented above. If any of those gaps would break your actual workflow today, use one of the other six until Morbstack's roadmap closes them.

8. Sources

All accessed 2026-08-02 unless noted.

Morbstack (this repository, git HEAD ba9afc6)

• docs/parity.md — the 29-check drop-in parity audit
• docs/roadmap.md — milestones and the public performance target table
• docs/compat.md — the compatibility contract and explicit non-goals
• docs/architecture.md — system design and VM-lifecycle measurements
• docs/amd64.md — Rosetta translation, measured slowdowns
• docs/k8s.md — morb k8s, measured enable/idle-cost numbers
• docs/sharing.md — VirtioFS bind mounts, ownership model, inotify gap
• README.md — status, what works, what doesn't yet

Local, first-hand inspection (read-only, 2026-08-02)

/Applications/Docker.app via du, codesign -dv --entitlements -, otool -L, nm -g, ps aux; ls /Applications, which colima podman nerdctl rdctl container, brew list --cask for competitor presence.

Docker Desktop / Docker Inc.

• Docker Subscription Service Agreement — free-use thresholds (§4.2)
• Docker pricing — plan prices
• Docker Desktop release notes — current version context (4.83.0, July 20 2026)

OrbStack

• Licensing
• Architecture
• Pricing
• Terms of Service
• Privacy policy
• OrbStack vs. Docker Desktop
• sliplane.io — OrbStack vs Docker Desktop in 2026 (third-party benchmark summary, not independently reproduced)

Podman / Podman Desktop

• podman-desktop LICENSE
• Podman machine providers (DeepWiki summary)
• Podman Desktop — Rosetta support
• podman-desktop/telemetry

Rancher Desktop

• Rancher Desktop LICENSE
• Rancher Desktop docs
• GPU/CUDA limitation discussion
• Rancher (Manager) telemetry FAQ — concerns Rancher Manager, not confirmed identical to Rancher Desktop's own telemetry

Colima / Lima

• abiosoft/colima
• Colima FAQ

Apple container

• apple/container
• apple/container technical overview