The supervision problem
An agent like Claude Code can work productively for long stretches — then it hits a permission prompt, finishes a task, or goes somewhere you didn't intend. If checking on it means walking back to the desk, you either hover (wasting the autonomy) or let it idle (wasting the run). The fix is making the agent's window visible and controllable from wherever you are.
Setup 1: stream the terminal
Run your agent in a terminal on the Mac, and stream that terminal window to your iPhone with Apperture. Terminal text stays legible because the window is sized to your phone, not shrunk from a desktop. When the agent asks for approval, tap and type your answer from the couch — more on this workflow here.
Setup 2: stream what the agent is building
For iOS work, the interesting window often isn't the transcript — it's the app. With the Apperture agent skills installed, your agent can build and launch the app in the iOS Simulator, and Apperture streams the Simulator to your real iPhone with native touch and rotation. Each iteration lands in your hand seconds after the agent builds it.
Setup 3: leave the house
Both setups extend beyond your Wi-Fi with a private VPN. Tailscale (free) makes your Mac reachable anywhere while keeping the stream peer-to-peer — no cloud relay, so your code and your agent's output never touch third-party servers.
Practical notes
- Keep the Mac awake for the duration of long runs.
- Stream one window per session and switch as needed — the point is legibility.
- Pairing requires physical QR-code proximity once, and the Mac owner's approval; see the security overview.
- Watching Xcode builds and tests works the same way, agent or no agent.