Run your coding agents from your phone

The problem is the pane, not the SSH connection

A coding agent on your own machine is already reachable over SSH. The failure starts after you are attached. Claude Code, Codex, and similar tools hold the focused pane. Anything you type there is that program's stdin. On a laptop you might chord a tmux prefix and get away with it. On a phone, prefix chords are fragile, and a foreground agent will happily swallow tmux next-window as a prompt fragment.

SSHHIP's contract is structural: multiplexer actions never go through the interactive PTY. They run on a separate SSH exec channel against the session you attached to. The agent keeps its pane. You still move windows, focus a neighbor, split, zoom, or detach. That mechanism is the product, not a convenience wrapper around typed commands. The Claude Code guide walks the setup and the two failure modes that show up only after a naive attach "works."

Anthropic's Remote Control is the official default for a Claude-app check-in: QR into the Claude iOS app, chat UI, subscription login. That is the right tool for that job. The honest split, including API keys and the roughly ten-minute unreachable timeout, is Claude Code Remote Control vs SSH.

How a check-in actually goes

Save the host once. Set startup behavior to attach tmux or herdr with the session name you already use. Connect. You land inside that session, not a fresh empty shell. The Command Dial is the whole control surface: keys, pane focus, multiplexer actions, snippets, voice, and the software keyboard. The keyboard appears only when you pick Keyboard. Most of a check-in never needs it. Opening a session, reconnecting, or dismissing a sheet will not flash it.

Directional pane focus is a first-class dial control, not a prefix sequence. A boundary edge is a calm no-op with haptic feedback. Destructive closes ask for confirmation. Detach, on tmux, targets SSHHIP's own client by resolved TTY so a second terminal on the same session is not kicked off. herdr has no socket-exposed detach; that leaf is hidden rather than faked. The tmux hub and herdr hub cover the native differences. The verified mapping is the command-level source of truth.

Send Image, without a third-party host

When the agent needs a screenshot, SSHHIP uploads the image over the live SFTP channel into ~/.sshhip-uploads/ on the same host and inserts the remote path on the input line. It does not press Return. The agent reads a file on the machine it already runs on. There is no second login, no image CDN, and no confirmation step: pasting or picking an image is the intent. This is not an SFTP browser; there is no Files provider and no general file tree. Send Image is the scoped upload. Other clients that offer a full file browser are listed honestly on the comparison matrix.

What SSHHIP deliberately does not do

Saying the limits is part of the product. SSHHIP does not ship an agent kanban, a diff viewer, push approvals while the app is closed, or a vendor relay in the notification path. App Store builds are SSH-only: a network change means reconnect-and-reattach. With a multiplexer that is lossless for the agent, not seamless Mosh roaming. If you need a vendor-hosted agent surface, a different client may fit better - start with the Moshi comparison. If you need port forwarding, jump hosts, or an SFTP browser, start with Termius or Secure ShellFish.

Credentials stay in Keychain / Secure Enclave. There is no SSHHIP account and no SSHHIP backend. Optional private iCloud sync covers saved-host metadata only, defaults off, and never includes passwords or keys. Access is a local one-month trial, then a $29.99 lifetime unlock - the same full app either way. Details live on the download page and privacy.

Where to go next

• Guides hub - setup paths and workflow comparisons in one place
• Flagship guide: Claude Code from iPhone
• Remote Control vs SSH
• Codex from iPhone
• Reach your Mac with Tailscale
• Check that the host is phone-ready
• tmux hub · herdr hub
• Honest comparison matrix