Codex CLI remote access

Codex CLI runs on the machine that has your code. TermLink relays the terminal it is running in, so you can read it and answer it from any browser or phone — without SSH, a VPN, or an open port.

How do I reach a Codex CLI session on another machine?

Run TermLink on the machine, then open that session from a browser. Codex keeps running where it already was; what travels is the terminal.

termlink            # on the machine with your code
                    # then open it from any browser

The machine dials out to the TermLink relay and never accepts an inbound connection, so there is no port to expose and no key to distribute. That is the same for Codex as for anything else — how a remote terminal works without SSH or a VPN.

Does TermLink actually understand Codex, or just show the terminal?

Both, and the second part is what makes the session list useful. TermLink recognises Codex as its own agent rather than treating every session as anonymous text.

Codex marks its transcript differently from Claude Code — it opens a paragraph with • and echoes your prompt behind ›, where Claude Code uses ● and >. TermLink reads those marks, so a session card shows Codex's most recent message as a line you can actually read, not a block of monospace.

It also handles the way Codex writes: Codex renders inline code and drops the backticks, while leaving ## headings and - list markers standing. TermLink pulls the headings and lists back out of that text.

Can I stop Codex asking for approval on every step?

Turn on Auto-Yes for that session. TermLink watches for Codex's approval prompt and confirms it, so a long run keeps going while you are away from the keyboard.

Codex uses a numbered menu, the same shape Claude Code uses, so Auto-Yes confirms with Enter. Cursor Agent asks differently and gets its own key. TermLink keeps one per agent rather than sending the same keystroke everywhere and hoping.

It is off by default and scoped to one session. It is a convenience rather than a safety feature — how Auto-Yes works, and when to leave it off.

Can I run Codex and Claude Code at the same time?

Yes, and they do not have to be on the same machine. Each session is a terminal, so one can be Codex and the next can be Claude Code, on a laptop and a workstation, in the same list.

Each card carries that session's latest line, read with the right marks for whichever agent is in it. You read a few lines and go to the one that needs you — managing several sessions at once.

Sessions that are not running an agent at all belong there too: a build, a log you are tailing, a shell you left mid-thought. See the session manager for AI coding agents.

Does it work from a phone?

Yes. TermLink runs in a mobile browser and adds a control-key row — Esc, Tab, Ctrl+C, arrows — so a terminal UI stays usable on touch.

You can read what Codex is doing, answer its prompts, and flip Auto-Yes without going back to your desk. More on that in using a coding agent from your phone.

What does it cost?

There is a free plan, and TermLink does not charge by agent. Free and Pro differ in one number: how many sessions run at the same time — two on Free, eight on Pro at $9 a month.

The cap counts sessions, not viewers or machines. What each plan includes is on the pricing page.

What we have and have not checked

Codex support is in daily use here and inline rendering was fixed and verified on an idle session — 28 samples over 26.5 seconds with no drift. What has not been measured yet is the screen while Codex is actively redrawing, so we are not going to tell you it is flawless.

Claude Code remains the agent we support most closely and test the most. If you see Codex redraw badly, tell us what you saw — bert@n2computing.com, or troubleshooting for the problems we already know about.

Last checked: August 25, 2026.

Get started Open app