Compare
Whose computer is your code on while the agent works?
That architectural choice decides which toolchain runs, which local state exists, and who must hold a copy of the repository.
Cloud runner
repository→managed environment→result
PRCHD
your phone→your Mac→your branch
The short, honest table
| Cloud agent runner | PRCHD | |
|---|---|---|
| Where the working copy lives | Provider-managed environment | Your Mac |
| Whose toolchain runs | Configured remote environment | The one already on your Mac |
| Local uncommitted state | Not present unless uploaded | Already present beside the work |
| Agent access | The service’s configured agent | Your installed Claude Code, Codex, or Grok CLI |
| What closing the phone does | Task continues remotely | Task continues on your Mac |
| Review surface | Browser or product UI | Phone views rendered from your Mac |
Where cloud runners are better
No local setup is a real advantage.
- Start from a device without preparing a Mac
- Work on a repository you do not have checked out
- Keep running when your own computer is unavailable
- Hand the remote task to a teammate in the same system
What your own machine gives you
The environment you actually debug.
- Your installed toolchain and exact versions
- Your local services and configuration
- Your existing uncommitted state
- Your agent CLI, provider account, and rate limits
- No third party holding the working copy for PRCHD
Where PRCHD sits
A remote control, not a place to run the agent.
PRCHD gives the sessions already on your Mac a phone-sized surface for direction, diff review, commands, terminals, and git. Your phone reaches the Mac directly over your tailnet. The public PRCHD service is not in that app-traffic path and never receives source code.
Keep the code where the toolchain is.
Start with the Mac host and pair your phone over Tailscale.