Mobile coding is mostly fake. This one isn't.
Most "mobile coding" stories end the same way: a stripped-down editor, a hosted runtime that isn't your environment, and a slow round-trip. You can't run your real toolchain, your cloned repos aren't there, and your AI agent is firewalled off from the projects it knows.
share terminal flips the model. Your laptop stays the host. The CLI streams its actual PTY — same shell, same env, same cloned repos — straight to your phone's browser over an encrypted WebRTC channel. Your phone is just a keyboard and a screen. Claude Code is still running on your laptop, with all its context intact.
A real session, end-to-end
- 01
Park a session on your laptop before you go
Run `share-terminal start --private`. The CLI prints a 6-digit pairing code. Private mode means only your own Google account can claim it — even if someone glimpses the code, they can't use it.
- 02
Open shareterminal.cloud on your phone
No app to download. Sign in with the same Google account, type the code, and the WebRTC handshake completes in a couple of seconds. Your laptop's shell is now in your hand.
- 03
Drive your AI agent normally
Type `claude` or `codex` or whatever you use. The agent runs on the laptop, sees your real filesystem, and streams its output to your phone. Tap to interrupt, type to redirect.
- 04
Walk away when you're done
Close the browser tab and the CLI cleans up — no orphaned daemon listening on the laptop, no recorded log of what you ran. Everything you needed was the laptop's own state.
AI tools that fit this perfectly
Claude Code
Claude Code is a CLI — it lives in a real shell. share terminal hands you that shell from your phone. The agent's file edits, command runs, and confirmation prompts come through the PTY exactly as on your desk.
OpenAI Codex (CLI)
The same story. Codex runs on your laptop with your repo cloned and your auth in place. From your phone, you read its plan, approve diffs, and rerun tests.
aider, opencode, plandex
Any agent or assistant that lives in a terminal works without modification. There is no integration to write — the PTY is the integration.
Plain shells, vim, tmux
Sometimes you just want to read a log on the train. Same pairing flow, no AI involved — it is still your laptop, in your browser.
What you give up — and what you don't
- You don't give up your environment. The PTY is your laptop's real `$SHELL`, with the same env, the same paths, the same dotfiles.
- You don't give up privacy. Output bytes flow directly between your laptop and your phone over a DTLS-encrypted DataChannel. Our servers handle pairing — not terminal data.
- You do give up the desk-grade keyboard. Mobile typing is mobile typing. Pair it with a Bluetooth keyboard if you're doing real work, or lean on the agent to do the typing for you.
- You do give up a closed lid. Your laptop must be awake. Disable sleep on AC, or run inside a tmux session on a small always-on machine you own.
Starter: park a Claude Code session before you leave
Run before stepping away from the laptop. The session stays open until you close the phone tab.
- 01
On the laptop,
cdinto the repo you want to work on. Start a private session — only your own Google account will be able to claim it.cd ~/code/your-repo share-terminal start --private - 02
The CLI prints a 6-digit pairing code. Keep the laptop awake (disable sleep on AC) and head out.
✓ pairing code 482017 · expires in 1:48 - 03
On the phone, open the join page in any browser, sign in with the same Google account, and enter the code. The handshake completes in a couple of seconds.
# in mobile browser: https://shareterminal.cloud → sign in with Google → enter 482017 - 04
You're typing into your laptop's real shell. Run your AI agent and start building.
claude # or codex # or just plain vim, tmux, anything