Codex on your phone: monitor your runs remotely
Most "answer your coding agent from your phone" tools are Claude-only, so if you run OpenAI Codex CLI you're usually told to just keep a terminal open. You don't have to. This page covers why Codex runs stall in the first place, the real ways to watch and answer them from a phone, and the honest trade-offs — including when the free options beat ours. (Disclosure up front: DevFob, one of the three options below, is our product.)
Why a Codex run stops and waits for you
Codex CLI is deliberately cautious. It works inside a sandbox, and depending on your approval mode it will pause the whole run and wait for a human whenever it wants to step outside what it's been pre-approved to do. In practice a long Codex run stops for:
- Command approvals. Running a command outside the sandbox or the workspace directory — installing a dependency, touching the network, running a script with side effects — triggers an approval prompt in most modes.
- Escalation requests. A sandboxed command fails and Codex asks to retry it with more access. Until you answer, nothing moves.
- Questions and plan checks. Codex asks which of several approaches you want, or presents what it intends to do before a bigger change.
- Being finished. The least obvious one: the run completed twenty minutes ago and the follow-up prompt it needs from you is the only thing between you and the next task.
You can flip Codex to full access and never see a prompt again — but then it can do anything on your machine unattended, which is exactly why most people don't. The realistic setup is "cautious mode + a way to answer from wherever you are." Here are the three honest ways to get the second half.
Option 1: ssh / mosh + tmux — free and manual
Run Codex inside tmux on the Mac, then ssh (or better, mosh) in from a phone terminal app like Blink or Termius and attach to the session. This is the classic answer, it costs nothing, and no third party sees your session content.
The rough parts are real, though: nothing notifies you — you have to keep checking a terminal to discover Codex has been waiting; approval prompts and multi-choice questions are painful to drive from a phone keyboard; you need reachable ssh from wherever you are (usually a VPN or Tailscale); and the Mac must stay awake on its own. If you live in a terminal and check in on your own schedule anyway, start here — it may be all you need.
Option 2: Codex cloud + the ChatGPT app — let OpenAI run it
OpenAI's own remote story is different in kind: instead of mirroring the Codex running on your machine, you delegate tasks to Codex in OpenAI's cloud sandboxes and follow them from ChatGPT on the web or the ChatGPT mobile app, reviewing diffs and PRs from your phone. If your workflow fits that shape, it's genuinely good — it's first-party, it's built into a subscription you may already pay for, and the phone experience is the whole point.
What it doesn't do is help with the local case: a cloud task runs against a repo in
OpenAI's environment, not the checkout, branch state, credentials, and tools on your Mac.
If the run you care about is codex in your own terminal — local services,
private code that shouldn't leave the building, your exact environment — the cloud path
isn't monitoring that run at all; it's a different way of working.
Option 3: DevFob — mirror your local Codex to your phone (that's us)
DevFob is a Mac menu-bar app that mirrors the agent sessions running on your Mac — Codex, Claude Code, and Grok Build — to free companion apps on iPhone, Apple Watch, Android, and Wear OS. For a Codex run you get it on glass: which session is going, what it's doing, how long it's been at it, the last reply, and its rate-limit windows — so from the couch you know whether it's working, waiting, or done, without squinting at a scrollback. Answering questions and approving plans from your phone is Claude Code today (multi-choice questions as tappable forms, permission gates one tap, full plan text on approvals); for Codex, DevFob is a monitor and you still answer at the Mac.
The architecture is the opposite of a relay: phone and Mac talk directly over your LAN or the Mac's hotspot — Curve25519 pairing, every frame sealed with ChaCha20-Poly1305, no cloud, no account, zero app analytics, and the license check happens offline. Nomad mode keeps the Mac awake and reachable lid-closed in a bag. The same app shows per-account rate-limit meters and can auto-switch Claude accounts before a run stalls (auto-switch is Claude-specific today; Codex sessions get the mirroring and usage). $15 once for the Mac app, 14-day full trial, free updates; the phone and watch apps are free (Android is currently in closed beta). macOS 14+.
| ssh + tmux | Codex cloud (ChatGPT app) | DevFob | |
|---|---|---|---|
| Where Codex runs | Your Mac — your checkout, env, and tools. | OpenAI's cloud sandboxes — a repo copy in their environment. | Your Mac — DevFob mirrors the local CLI session. |
| Told when it needs you | No — you poll by reattaching. | Yes — ChatGPT app notifications for cloud tasks. | Yes — push to phone/watch the moment an approval or question lands. |
| Answering approvals | Typing y/n and arrows on a phone keyboard. | Cloud-task reviews in the app; not your local CLI's prompts. | Monitor for Codex (answer at the Mac); tappable answers are Claude Code today. |
| Works offsite | Anywhere you can reach ssh (VPN/Tailscale). | Anywhere with internet. | LAN, the Mac's hotspot, or a relay you self-host — no vendor cloud by design. |
| Privacy | No third party (beyond your VPN, if any). | Session and repo content live in OpenAI's cloud. | Local-only, end-to-end encrypted; session content never leaves your devices. |
| Watch app | No. | No. | Yes — Apple Watch and Wear OS; read the plan, approve from the wrist. |
| Cost | Free. | Included with eligible ChatGPT plans. | $15 once (launch price; $29 from Aug 14), 14-day trial, free updates; companion apps free. |
Where the others are genuinely the better pick: if you're happy delegating work to Codex in OpenAI's cloud, the ChatGPT app already gives you a polished phone experience for free — you don't need us for that. And if you live in a terminal, check runs on your own schedule, and don't mind typing at prompts on glass, tmux + mosh costs nothing and has no moving parts. DevFob also requires the phone and Mac to share a network (LAN, hotspot, or a relay you self-host) — if you want zero-thought access from any café Wi-Fi with no setup, a cloud path will always be more convenient than a local-only one.
Where DevFob is the better pick: the Codex run you care about is on your Mac against code that shouldn't transit anyone's cloud; you want to be pushed the moment it stalls instead of polling; you want approvals and questions as tappable prompts (or on your wrist) rather than a phone terminal; or you run Codex and Claude Code side by side and want one mirror for both.
What "monitoring Codex from your phone" should actually mean
A raw terminal view technically shows you everything and practically tells you nothing — at phone size you're squinting at a scrollback to figure out whether the agent is working, waiting, or done. What you actually want on a phone is state, not scrollback: which sessions are running, what each one is doing right now, how long it's been at it, and a prompt you can answer in one tap when it blocks. That's the bar we built DevFob to, and it's a fair bar to hold any option on this page to — including ours.
Codex CLI behavior (sandboxing, approval modes) and Codex cloud details as of July 2026 — OpenAI iterates quickly, so check their docs for current specifics. Spotted something outdated or unfair here? Tell us and we'll fix it. Also useful: every way to run Claude Code from your phone and our comparison with Omnara.