Argent Cloud vs Macswarm: remote Simulator vs Pod

What is the difference between Argent Cloud and Macswarm?

Where the agent runs and where the app is built. With Argent Cloud, the agent or CI job runs on your machine or a Linux runner and drives up to 6 iOS Simulators per Runner over the network, on Apple hardware reserved for your project; the app is built on a Mac of yours and uploaded. With Macswarm, the agent runs inside a Pod, a macOS VM where it builds with the Latest Xcode and drives pre-booted Simulators. Free, Pro and Team Pods run on shared Apple silicon; Enterprise reserves Macs. Argent facts are from argent.swmansion.com and its docs, checked 26 Oct 2026.

What does each one rent?

Argent Cloud: Simulators on a reserved Mac. The sim-remote CLI runs on your computer or CI runner, on macOS or Linux, with no local Xcode. Argent assigns your session a hosted Mac, and sim-remote simctl works as a drop-in for xcrun simctl there (how it works, checked 26 Oct 2026). install uploads your .app, up to 1 GiB (sessions and limits), and tunnels let the Simulator reach Metro or a mock API on your machine. Agents connect through the Argent MCP server, which works with Claude Code, Codex CLI, Cursor and other MCP clients (MCP server); tests can use Maestro, Appium or scripts.

Macswarm: a Pod the agent works in. A Pod is a macOS VM with the Latest Xcode, iOS runtimes, three to four pre-booted Simulators, XcodeBuildMCP, Claude Code and Codex. The agent clones your repository, builds in one of two build slots, leases a Simulator, taps through the app, then commits and pushes. One Pod belongs to one developer; two to three of that developer’s agents share it. Claude Team and Enterprise organizations connect through Anthropic’s self-hosted environments, in public beta and off until an Owner enables them (code.claude.com, checked 26 Oct 2026). Claude Pro and Max users start a Pod on Free or Pro and sign in to Claude Code inside it; those steps are not documented yet, so write to info@macswarm.com. Codex connects through an OpenAI Agents API webhook with API access. Inside a Pod lists the image.

How do Argent Cloud and Macswarm compare, row by row?

Argent facts checked 26 Oct 2026 on its pricing page and docs.

The data row compares different things: Argent’s operational machine logs, and Macswarm’s agent logs and screenshots.

When is Argent Cloud the better choice?

• Your agent or CI runs on Linux and the app is already built elsewhere, on a macOS CI job or a developer’s Mac.
• You need several Simulators busy for long stretches and the app is built elsewhere: one fixed Runner fee covers up to 6 at a time. Compare concurrent Simulators, build-host cost and billed Pod runtime, not hours alone.
• Tests run in Maestro or Appium from ubuntu-latest. The actions’ README also documents logging in with GitHub’s OIDC token instead of a stored key (argent-cloud-actions, v2).
• You work in React Native with Metro on your machine, or want Argent’s profiling and network inspection.
• Your agent is not Claude Code or Codex: any MCP client can use the Argent MCP server.

When does a Pod fit better?

• The agent has to build the app too, and nobody on the team has a Mac to build on.
• Sessions start in Claude or Codex in the cloud, so the agent itself needs a Mac with Xcode.
• One developer runs two or three agents that should share a build queue and Simulator pool.
• You want to pay for Pod runtime: $99 for 300 Pod-hours, with 3 free Pod-hours to try.
• The Mac must be in the EU or the US, and Pod-local data should be destroyed after the session. Security has the controls.

Start free with 3 Pod-hours

How do you move an agent-driven Simulator task to a Pod?

This is not a CI migration. Keep scheduled and push-triggered tests in your existing CI; Macswarm has no runner registration or workflow trigger.

1. Connect Claude Code or Codex: see the Claude Code quickstart and the Codex quickstart.
2. Drop the build-and-upload step: the Pod clones the repository and the agent builds in place.
3. Point the agent’s instructions at XcodeBuildMCP instead of sim-remote or the Argent MCP server.
4. Run a mock API or other local server in the Pod, next to the Simulator, without a tunnel.
5. Maestro and Appium are not in the documented Pod image; write to info@macswarm.com if your tests need them.

How do you move a Pod task to Argent Cloud?

1. Build a Simulator .app on a Mac: a macOS CI job or a developer’s machine.
2. Install sim-remote where the agent or job runs, export SIM_REMOTE_API_KEY and run sim-remote login.
3. For an agent, run npx @swmansion/argent@latest init to register the Argent MCP server.
4. In GitHub Actions, build on macOS and test on ubuntu-latest. The sketch below is an untested launch smoke test: it fails if the app does not install or launch, and saves a screenshot. Set RUNTIME and DEVICE to a pair the hosted Mac lists, at or above your deployment target.

jobs:
  build:
    runs-on: macos-latest
    steps:
      - uses: actions/checkout@v5
      - run: xcodebuild -scheme MyApp -destination 'generic/platform=iOS Simulator' -configuration Debug -derivedDataPath build build
      - run: tar -czf app.tgz -C build/Build/Products/Debug-iphonesimulator MyApp.app
      - uses: actions/upload-artifact@v4
        with:
          name: app
          path: app.tgz
  smoke:
    needs: build
    runs-on: ubuntu-latest
    env:
      RUNTIME: com.apple.CoreSimulator.SimRuntime.iOS-26-0  # example; use one the hosted Mac lists
      DEVICE: iPhone 17                                     # example; use one the hosted Mac lists
    steps:
      - uses: actions/download-artifact@v4
        with:
          name: app
      - run: tar -xzf app.tgz
      - uses: software-mansion-labs/argent-cloud-actions/install@v2
      - uses: software-mansion-labs/argent-cloud-actions/acquire@v2
        with:
          api-key: ${{ secrets.SIM_REMOTE_API_KEY }}
      - run: |
          UDID=$(sim-remote simctl list devices available --json | jq -r --arg rt "$RUNTIME" --arg d "$DEVICE" '.devices[$rt][] | select(.name == $d) | .udid' | head -n 1)
          test -n "$UDID"
          sim-remote simctl bootstatus "$UDID" -b
          sim-remote simctl install "$UDID" MyApp.app
          sim-remote simctl launch "$UDID" com.example.MyApp
          sleep 5
          sim-remote simctl io "$UDID" screenshot launch.png
      - uses: actions/upload-artifact@v4
        with:
          name: launch-screenshot
          path: launch.png

The v2 tag and README are in the argent-cloud-actions repository; Argent’s docs page still shows v1 (both checked 26 Oct 2026). acquire releases the machine when the job ends, even if it fails.

Does Argent Cloud build the iOS app?

No. Argent’s GitHub Actions documentation says an ubuntu-latest job can install an app and drive a hosted Simulator with no macOS runner and no Xcode, but the build of the .app still needs a Mac. The sim-remote CLI uploads the .app, up to 1 GiB per file (docs.swmansion.com, checked 26 Oct 2026). On a Macswarm Pod the agent builds on the Pod itself.

Is Argent Cloud cheaper than Macswarm?

It depends on how many Simulators you need at once and where the app is built; the two do not sell equivalent hours. Argent Cloud is $299 a month per Runner whatever the hours, with up to 6 Simulators in parallel, plus a Mac to build the app: a 10-minute build a day on GitHub’s standard macOS runner is $12.40 over 20 days at $0.062 a minute, gross before included minutes (argent.swmansion.com, docs.github.com, checked 26 Oct 2026). Macswarm Pro is $99 a month for 300 Pod-hours, build included, then $0.40 a Pod-hour. One developer working three active hours a day on 20 days bills about 63.3 Pod-hours with the ten-minute tails: $99 on Pro, model usage extra.

Mac in the cloud pricing puts the same month next to rented Macs and per-minute hosts.

What should you read next?

• iOS Simulator for coding agents: the testing loop on a Pod and other cloud Simulator options.
• Running Claude Code on a cloud Mac: how a session reaches a Pod.
• Limits and queues: agents, build slots and Simulators per Pod.

Where do these numbers come from?

• Argent Cloud price, Runner contents, self-serve limit and delivery time: pricing and overview. How it works, uploads, tunnels, limits and data: getting started, how it works, sessions and limits, security and data. Checked 26 October 2026.
• CI and the build requirement: GitHub Actions docs and argent-cloud-actions; MCP clients: MCP server; Claude eligibility: self-hosted environments; GitHub's rate: Actions runner pricing. Checked 26 October 2026.
• Macswarm: pricing, Limits and queues, Security and the privacy policy. Monthly figures are our arithmetic from list prices and the stated scenario.