AI agents, in your pocket.

Scan a QR code and control your coding agent from your phone. Approve, continue, and send prompts — without opening SSH or a VPN.

Request beta accessHow it works

No mobile app. Just a browser and a QR code.

You start an agent, step away, and it stops two minutes later — waiting for an approval you can't give from the couch, the train, or lunch. pocketerm is the remote control for exactly that moment.

How it works

  1. 1

    Run your agent under pocketerm

    pocketerm run -- claude

    Your terminal stays fully interactive. A QR code prints below the command.

  2. 2

    Scan the QR with your phone

    The browser opens a live view of the terminal. One QR, one session — the link works only for this run.

  3. 3

    Approve from anywhere

    pocketerm detects when the agent is waiting and notifies you. Tap Approve, navigate its menus, or type a prompt — the agent never sits blocked.

Built for the in-between moments

The approval that can't wait

Long refactor running at your desk, coffee downstairs. The phone buzzes: "awaiting approval". One tap and the agent keeps going.

Watch the long build

pocketerm watch -- npm test shares a read-only view. See the failure the moment it happens, not when you get back.

One more instruction

The agent finished, but you thought of the next step on the way home. Send a prompt from the phone; review the result when you arrive.

Safe by default

End-to-end encrypted

Terminal content is sealed between your machine and your phone (ChaCha20-Poly1305). The relay routes ciphertext; the session key travels only in the QR, never to a server.

Approve-only by default

The phone gets buttons, not a shell: every action passes a policy engine on your machine. Full keyboard input needs an explicit grant on the host — and expires in 60 seconds.

Short-lived and revocable

Sessions expire on their own and die instantly with one keystroke on the host. The first device claims the QR — a link that leaks later admits no one.

pocketerm is not a VPN, not a public web terminal, and not a backdoor — a session exists only while you run a command, and read-only is one flag away.

Private beta

pocketerm is in a closed beta while the rough edges get filed down. Invites include the CLI and a relay token — setup is three lines:

pocketerm config set relay.url https://app.pocketerm.com
pocketerm config set relay.token <your-token>
pocketerm run -- <your-agent>
Request an invite

Free while in beta. Prefer your own infrastructure? The relay is a single container you can self-host.

FAQ

Is this a web terminal?
No — it's a remote control for AI coding agents. The terminal is a live view; the primary controls are approve, continue, prompts, and the agent's menus. (A temporary full-keyboard mode exists when you really need it.)
Do I need to install an app on my phone?
Nothing to install: any modern mobile browser works, straight from the QR.
Can the relay read my terminal?
No. Payloads are end-to-end encrypted and the session secret never reaches the server — the relay sees only connection metadata. You can also self-host it.
What if the session URL leaks?
Sessions are single-QR and seat-limited: after your device connects, the URL admits nobody new. Access expires on idle, is capped in duration, and the host can revoke it at any moment.
Which agents does it work with?
Anything that runs in a terminal — Claude Code, Codex CLI, aider, plain scripts. If it prints and reads, pocketerm can supervise it.