What is PhoneCMD?
PhoneCMD is a system with two sides: a host app that runs on your PC and does the work, and a thin client on your phone (or in a browser at /app) that points and taps. The PC hosts the session - real terminal shells, your files, your coding CLIs - and the phone is the window into it.
Once paired, your phone can open persistent terminal sessions, launch coding CLIs like Claude, Codex, Cursor and Gemini and resume past conversations, browse and edit files on the machine, re-run commands from history, and hand an AI agent the keys to operate the PC for you. Each session is a true terminal that stays alive between messages, so interactive tools keep their state exactly like they would at the desk.
The connection goes directly between phone and PC: over your local network when both are on the same WiFi, or directly peer-to-peer over WebRTC when you are remote - the client picks the right path automatically. Your session traffic never passes through our servers either way. It is not a mobile IDE and not a cloud shell; it is a remote control for the machine you already use.
The work does not stop when you leave the desk.
A build finishes while you are out. A coding agent needs one more approval to keep going. A quick fix, a log to check, a file to edit, a command to re-run - all of it waiting on a laptop you are not sitting at. Reaching your own machine from your pocket has always meant fiddly SSH apps, port forwarding, and shells that forget everything the moment you look away.
PhoneCMD was built for that gap. The PC stays in control and keeps doing the real work; the phone becomes a practical, persistent way to watch it, drive it, and pick up long-running coding sessions from anywhere. Pair once and every PC is one tap away, with its shells and scrollback still there.
The goal is simple: keep your whole machine reachable when the laptop is not - without a relay server in the middle, and without giving up end-to-end encryption.
What it does
Real persistent terminals
Each session is a true PTY shell on your PC, not a one-shot command. Interactive tools, REPLs and long-running processes stay alive and keep their state between messages - exactly like a terminal at the desk. A small on-phone terminal emulator renders the raw stream correctly, so overwrites, progress bars and line wraps come out right on a phone-sized screen.
Coding CLIs: launch, separate, resume
Launch the coding CLIs installed on your PC - Claude, Codex, Copilot, Cursor, Gemini, Antigravity, Aider - with a folder picker, extra flags and a first prompt. For CLIs whose sessions are listable, resume a specific past conversation in its own project directory, or browse every project on the PC. Each tab is its own separate conversation, targeting a specific session id rather than "most recent".
One input, two backends
The command bar has a Shell to AI toggle instead of two boxes. In Shell mode the line goes straight into the live terminal. In AI mode it goes to an on-device Claude or GPT agent that interprets it and drives the shell for you - one place to type, whether you want to run a command yourself or ask the machine to.
On-device AI agent
Chat with Claude or GPT using your own API key, stored only in the phone's Keychain and sent straight to the AI provider - never to our servers. The agent runs a tool-calling loop on the phone and operates the PC through the same commands you have: run shell commands, read and list files, create profiles - each executing in a live session you can watch. The agent is always free; it runs on your key, so we never pay for or see your tokens.
Files, sessions, history
Browse, view and edit files directly on the PC. Switch between every live shell and agent session from the Sessions tab, rename any of them, and open new shells or CLIs. The History tab shows the PC-stored command history - tap any entry to re-run it. Multiple PCs sit in a rail, online ones first, each saved for one-tap reconnect.
Answer prompts without a keyboard
When a CLI asks something - a Yes/No, "press enter", a numbered menu - the app detects it and opens a key bar that sends raw bytes to the terminal: Enter, Esc, Y, N, arrows, Tab, Ctrl-C. You can drive interactive prompts from a phone without pulling up the soft keyboard.
Pair in under a minute
Two ways to pair, nothing else to type. Scan the QR code the desktop shows, or paste one PCMD keycode - a single string that carries the LAN endpoint, public key, pairing code and machine name. No host, port or key to enter by hand; the phone decodes it and picks the right path, whether you are local, on WiFi, or remote.
The link survives IP changes and restarts
Remote connections are direct WebRTC peer-to-peer, so there is no relay URL to rot. The keycode carries a stable PC identity (the host's public key, persisted across restarts), and the host re-resolves its LAN IP on every heartbeat - so a DHCP renew, a WiFi reconnect, or a keycode you saved weeks ago still connects. A brief network blip is treated as a hiccup to recover from, not a disconnect: the terminal keeps its output on screen under a small "Reconnecting" banner instead of blanking.
Command-line tool and web client
The host ships a phonecmd CLI that drives the same surface as the phone - run commands, do file ops, list installed CLIs, run a coding CLI or the AI agent, or start a host and print a QR and keycode - handy for scripting and headless servers. There is also a web client at /app with feature parity for use on the same network.
Private by construction
PhoneCMD is end-to-end encrypted, and remote traffic goes directly peer-to-peer, so anyone on the wire sees only ciphertext. The handshake uses X25519 - the PC's public key rides in the QR or keycode out-of-band, which authenticates the host and prevents a machine-in-the-middle. Messages use crypto_box with random nonces, and a session token inside the encrypted channel is required on every command.
Pairing needs your consent: a pairing code alone is not enough - the desktop shows an approval prompt naming the device, and nothing attaches until you approve it. Approved devices are then remembered by their public key, so they reconnect without re-pairing however often either side restarts, while a new device always needs the current code and your approval. You can revoke any device from the desktop and it goes back to pairing from scratch.
Your AI API keys live only in the phone's Keychain and hit the AI provider directly - they never touch our servers. The control-plane API only handles store subscriptions, brokers the brief peer-to-peer signaling handshake, and maps a PC's public key to the endpoint it is currently on. It never relays command traffic and never sees plaintext.
Start free, go unlimited
Everything works on the free plan, capped at 10 terminal commands a day (counted and enforced on the host). The AI agent is always free because it runs on your own API key. Paid plans lift the cap and buy the maintained connection infrastructure - a one-day pass at $1, a weekly plan at $3, or the best-value monthly at $8 with priority support. Payments go through the Play Store and App Store.