Terminal

The Terminal tab attaches to any pane on any connected host and shows the real PTY — ANSI colors, TUI redraws, agent output — streamed live and end-to-end encrypted.

Attaching

Pick a host, then a pane, and attach. The view opens from a clean snapshot of the pane's current screen, then follows it live. Detach any time; the pane on the host is completely unaffected — you are watching and typing into it, not moving it.

Opening a terminal sits behind the same biometric gate as the cockpit.

Interacting

• Type through the on-screen keyboard; a shortcut strip covers what phone keyboards lack — Esc, Tab, Ctrl combinations, and arrows.
• Scroll back through history (up to 2,000 lines per request).
• Resize — the pane adopts your phone's measured character grid while attached, so TUIs re-flow to the screen you're actually reading.

Two modes

• Stream (Linux and macOS hosts) — frames arrive as an authenticated, ordered sequence, coalesced to ~20 fps. If the sequence ever gaps — a dropped frame, a reconnect — the app discards the partial view and reopens from a fresh snapshot rather than showing a plausible-but-wrong screen.
• Snapshot (Windows hosts) — the view refreshes as full screen captures instead of a continuous stream.

The host's mode is advertised as a capability and picked automatically.

Limits

Terminal input is capped at 16 KiB per send, the grid at 400×240 characters. Everything in between phone and bridge is sealed per-message — the hub never sees a frame. Input expires within seconds if undeliverable, like every other command.

TIP — > The terminal is the ground truth for permission decisions: before allowing something consequential, attach and read exactly what the agent is asking.