Open a repo on any of your machines like an IDE — code, git, terminals, Claude and Codex, and your dev server rendered live on the phone. Plus a real terminal, monitoring, and push alerts for every box you own: VPS, Pi, NAS, your Mac.
Pick a folder on any machine you've paired and it opens as a workspace — code, git, terminals, Claude, and the running dev server, all tabs in one space. Here's one edit moving through all four:
A real shell, an IDE-style project space with a live dev-server preview, telemetry, and alerts when something needs you — plus your Claude and Codex sessions — for all the machines you already own.
Full PTY sessions over an encrypted peer-to-peer channel. Run several sessions per host, attach to tmux or screen, and hit Ctrl chords, arrows, and tab-completion from an on-screen key bar built for a terminal, not a notes app.
Open a project folder on any host and get the editor treatment: a file tree, syntax-highlighted editing, git staging and diffs, terminals, agent chats, and a live web preview — all tabs in one space, scoped to that project. Come back tomorrow, on any device, and every tab is exactly where you left it.
PocketShell finds the dev servers already running on your host — Vite, Next.js, Rails, Django — and opens them in the app over the same encrypted peer-to-peer channel. No ngrok, no tunnels, nothing exposed to the internet. Hot reload included: save the file on your laptop, watch the page update on your phone.
CPU, RAM, disk, network, load and temperature, all live, with real-time graphs and a sortable process list. Tap a process to see exactly what's eating the machine.
Set per-host thresholds with cooldowns. Get a push the moment CPU pegs, a disk fills up, or a host goes offline, or when a long-running command (a build, a backup, a coding agent) finishes in one of your terminals.
Every machine you own in a single list with live online / offline status and quick CPU and RAM bars. Open a file browser on any of them to navigate directories, upload, and download.
Browse Claude Code and Codex history on every paired host, then resume a session in a real terminal. Claude Remote Control and Codex Remote provide richer first-party control for their own agents; PocketShell combines both agents with arbitrary shell access, files, git, monitoring, alerts, and dev-server preview across your hosts.
CPU steal on a noisy neighbor. A runaway process pins the box. A deploy quietly OOMs at 3am. You find out hours later, when a user emails or you happen to SSH in. Nothing was watching.
A Claude run on the Mac, a Codex job on the VPS, another on the Pi. Claude Remote Control and Codex Remote are strong first-party options for their own agents, but neither is a general terminal, file browser, monitoring dashboard, and alert system for the whole machine.
A shell on the phone gets you a restart. But reading code across files, reviewing what the agent changed, staging a commit, checking the page actually renders — that's vim over SSH on a six-inch screen, so the fix waits until you're back at a desk.
Either way, PocketShell pushes your phone the moment something needs you — and gives you the whole project, not just a shell, to deal with it.
A vendor's app drives that vendor's agent. Termius is a terminal — and to reach a private host it needs a tunnel like Tailscale or Cloudflare on top, plus SSH keys. PocketShell is one app for the whole machine: terminal, connectivity, monitoring, alerts, files, and your Claude and Codex sessions.
| Capability | PocketShell | First-party agent apps¹ | Happy (OSS) | Termius | Tailscale |
|---|---|---|---|---|---|
| Project space — file tree, code editor, git, terminals in tabs | |||||
| Preview your dev server on the phone — run pocketshell expose once, no tunnel service | |||||
| Real terminal — any shell, tmux, any command | |||||
| Reaches a private host with no inbound ports or port-forwarding | |||||
| Claude and Codex together, on any host | |||||
| Resume a running agent session from your phone | |||||
| Works with whatever Claude plan you already have | |||||
| Live host monitoring — CPU, RAM, disk, network | |||||
| Push alerts — host offline, disk full, CPU pegged | |||||
| Multi-host fleet dashboard with live health | |||||
| File browser & transfer across hosts | |||||
| No vendor relay in the steady-state data path — E2E on P2P/TURN paths | |||||
| Price for the phone features | Free · $4.99 Pro | Claude: eligible plans · Codex: all ChatGPT plans¹ | Free | $10/mo | Free² |
← scroll the table →
¹ Anthropic Remote Control works with eligible Claude subscriptions; its requirements list Pro, Max, Team, and Enterprise, with Owner enablement on Team and Enterprise and no API-key support. Codex Remote is generally available on all ChatGPT plans for paired Mac and Windows hosts. PocketShell does not include or replace either subscription. ² Tailscale is a connectivity layer, not a terminal — in practice you combine it with Termius plus SSH keys just to reach a private host, and still get no monitoring, alerts or agent view. Happy is a great free agent viewer; it has no terminal, monitoring, alerts or files, and its sync relays through its server (end-to-end encrypted). PocketShell's direct and TURN-relayed WebRTC traffic is end-to-end encrypted with DTLS. Terminal and agent traffic can use a server-readable WSS/TLS fallback while WebRTC is unavailable; file signaling has a documented key-exchange fallback. Each tool is rated from the primary sources linked in the detailed guides.
PocketShell prefers authenticated WebRTC between your phone and host. Direct and TURN-relayed traffic is protected by DTLS end to end, while the backend handles sign-in, pairing, presence, alerts, lightweight metric history, and connection setup. Terminal and agent traffic can use a server-readable WSS/TLS fallback while WebRTC is unavailable, and file signaling has a separate caveat. The host agent is open source so these paths can be verified.
This diagram shows the preferred WebRTC path. When it is unavailable, terminal and agent frames can transit PocketShell over server-readable WSS/TLS; file signaling normally adds an E2E envelope but can fail over to WSS/TLS if key exchange fails.
The agent makes one outbound connection. Nothing on your machine listens for the internet. No port forwarding, no VPN, no firewall holes.
A phone is trusted only after you scan a QR printed by the host and approve it, gated behind Face ID / Touch ID on the device.
Direct and TURN-relayed WebRTC traffic is end-to-end encrypted with DTLS. Terminal and agent traffic can use a server-readable WSS/TLS fallback while WebRTC is unavailable; file signaling has a documented key-exchange fallback.
Ed25519 keypairs are pinned per device and bind the WebRTC certificate fingerprint. The backend cannot substitute a peer on the signed WebRTC path.
PocketShell does not intentionally persist terminal scrollback, agent conversations, or file contents. Some payloads can transit the server-readable fallback paths documented on the security page.
One command, pocketshell uninstall, removes the agent and its keys. No leftover daemons, no leftover trust.
The host agent is open source under Apache-2.0. It's the same binary that runs on your machine. Review the install script and the source before you run anything. Build it yourself if you'd rather not trust our release.
On any machine you want to reach: your VPS, your Pi, your Mac.
curl -fsSL https://pocketshell.app/install.sh | bashrecommendednpm i -g pocketshellIt's open source. Review it before you run it:source on GitHub·install.sh
The install script shows a QR automatically. Installed via npm? Run this to bring it up.
pocketshell pairOpen PocketShell on your phone and point it at the QR. Approve with Face ID and you're paired.
Runs onLinuxmacOSWindowsRaspberry PiNASVPS
Start free with up to two hosts. Upgrade when your fleet grows, or buy it outright once and keep it.
Purchases happen inside the iOS / Android app. Nothing is sold on this page.
Still unsure? Read the source, or email [email protected].
Those apps drive one vendor's agent — Claude's app controls Claude, the ChatGPT app controls Codex. Their updates route through the vendor's service. PocketShell prefers authenticated WebRTC and keeps P2P/TURN traffic end-to-end encrypted; terminal and agent traffic can use a server-readable WSS/TLS fallback while WebRTC is unavailable. PocketShell sits a layer above: a real terminal, live monitoring, and alerts for the whole machine, with your Claude and Codex sessions together in one place, across every host. It's the control panel for the machine, not a remote for one agent.
PocketShell doesn't include, replace, or resell any AI subscription. It's a window into the Claude Code and Codex CLIs already installed and signed in on your machines — whatever plan you already use (Pro, Max, or an API key) keeps working exactly as it does today. Anthropic Remote Control separately requires an eligible Claude subscription and does not support API-key authentication; Codex Remote is generally available on all ChatGPT plans. Viewing and resuming local history in PocketShell adds no AI plan requirement of its own. And if you don't use coding agents at all, everything else — terminal, monitoring, alerts, files — works with no AI subscription.
Run Claude Code or Codex on your machine as you normally would. PocketShell shows the sessions running on each host, lets you browse past ones, and resumes any of them in a real terminal on your phone. You get a push the moment a run finishes.
The backend handles sign-in, pairing, WebRTC coordination, presence, alerts, lightweight metric history, and fallback relay. Direct and TURN-relayed WebRTC payloads are end-to-end encrypted with DTLS. Terminal and agent frames can be server-readable on WSS/TLS fallback; file signaling normally adds an E2E envelope but has a documented key-exchange fallback. See the security page for the full data map.
Yes, Apache-2.0. The agent you install is the code on GitHub. Read it, audit it, or build it from source yourself.
A host only trusts devices you've explicitly paired. Revoke the lost device from Settings → Security on another paired phone. New backend-mediated setup is blocked once revocation is committed; an already-open direct channel closes when the host next synchronizes its trust list (five minutes by default). Pairing is also biometric-gated.
No. The agent makes a single outbound connection. Nothing on your machine listens for inbound traffic, so there's no port forwarding and no VPN to set up.
A project space is a workspace pinned to one folder on one of your hosts. Inside it you open tabs — files with syntax highlighting, a git panel with staging and diffs, terminals, Claude or Codex sessions, and a live preview of the project's dev server. The layout is saved as you go: kill the app, switch phones, come back a week later, and the same tabs open to the same places. Free accounts get two project spaces; Pro is unlimited.
No. Preview traffic is end-to-end encrypted on the same P2P/TURN paths as the terminal — your dev server keeps listening on localhost, and nothing is published to the internet or routed through a public URL. Each port has to be explicitly allowed on the host, grants made from the phone are temporary (they expire when the agent restarts), and you can revoke any of them at any time.
Not on our servers. Terminal contents are never stored by PocketShell, scrollback lives on your own devices.
Run pocketshell uninstall on the host. It removes the agent and its keys, with no leftover daemons or trust.
Linux, macOS, Windows, Raspberry Pi, and NAS boxes, effectively any VPS or machine you run.
