Run OpenAI Codex CLI on Your Phone (iPhone + Android)

You can run OpenAI Codex CLI on your phone today, iPhone or Android. The fastest path: install Cosyra for iOS or Cosyra for Android, sign in, and type codex in the terminal. We pre-install Codex CLI in an Ubuntu 24.04 container in the cloud, so there is no Node install, no Termux ARM headaches, no SSH tunnel to babysit. Add your OpenAI API key, describe what you want, review the diff. 1 hour free on signup, no credit card.

This guide was written by the Cosyra team. We tested Codex CLI on iPhone and Android via Cosyra, plus the two main alternatives, in April 2026. The hands-on numbers below are dated 2026-04-16; we re-verified the install command, the host platforms, and the four distinct surfaces sharing the "Codex" name on 2026-05-19.

What is OpenAI Codex CLI?

OpenAI Codex CLI is OpenAI's open-source command-line coding agent. Source at github.com/openai/codex. You install it with npm install -g @openai/codex (or brew install --cask codex) on Linux or macOS, sign in with your ChatGPT plan or an OpenAI API key, and run codex inside a project. It reads your files, proposes patches, runs commands in a sandboxed working directory, and asks for approval on actions that touch the outside world. It is a binary that expects a real shell, which is why getting it onto a phone takes a bit of work.

Heads-up on the name: OpenAI now ships four products called "Codex" — the CLI this guide covers, the Codex desktop app (codex app), Codex Web at chatgpt.com/codex, and Codex in the ChatGPT mobile app (launched 2026-05-14, which remote-controls Codex running on your Mac). If you came here looking for the new ChatGPT-mobile feature, we cover that paradigm head-to-head in AI pair programming from a phone. This guide is about the CLI binary you actually run inside a Linux shell.

How can you run Codex CLI on a phone?

You can run Codex CLI on a phone three ways: a cloud Ubuntu container with Codex pre-installed reached from a native mobile app (Cosyra), Termux on Android plus Tailscale plus SSH into a laptop or VPS, or Blink Shell on iOS into a VPS you own. Codex CLI needs a real Linux or macOS shell, so every approach puts the actual Codex process somewhere else and gives your phone a good way to talk to it. The three options below all work as of 2026-05-19. A fourth path — remote-controlling Codex on a Mac from the ChatGPT mobile app, launched 2026-05-14 — only works if you already have a Mac to leave awake; if that fits your setup, we walk through it in AI pair programming from a phone.

1. Cosyra: cloud Ubuntu container, Codex pre-installed

This is what we built. A native iOS and Android terminal that connects to a persistent Ubuntu 24.04 container in the cloud. Codex CLI, Claude Code, Gemini CLI, and OpenCode are already installed. You add your OpenAI API key, type codex, and go.

• Works when: you want zero setup and no other machine to manage.
• Breaks when: you have no internet (the container lives in the cloud), or you need to reach a corporate VPN we do not route through. See our honest list of trade-offs in Cosyra vs Termux.
• Cost: 1 hour free on signup, no credit card. Extend with a 10-hour, 7-day trial when you want more. After that, $29.99/month or $300/year. Full details on pricing. You pay OpenAI separately for tokens.

2. Termux + Tailscale + SSH to a laptop or VPS (Android)

Install Termux from F-Droid, install Tailscale on both your phone and a machine that runs Codex, SSH into that machine inside a tmux session, and run codex there. The phone is a dumb terminal, Codex actually runs on the laptop or VPS.

• Works when: you already have an always-on Linux box and are comfortable wiring up SSH, Tailscale, and tmux.
• Breaks when: your MacBook lid closes, your VPS reboots, or Android 12 and later decide your background SSH process is a "phantom" and kill it. We have watched phantom process killer sever a Codex session mid-refactor on a Pixel 8 running Android 14. The fix is a foreground notification hack in Termux:Boot that not everyone wants to maintain.
• Cost: free if you own the hardware, plus your VPS bill if you do not. We think this is the best choice if you already run a home server, and the worst choice if you do not, because you end up debugging networking instead of shipping code.

3. SSH from Blink Shell into your own VPS (iOS only)

On iPhone or iPad, Blink Shell is the gold standard SSH client. Spin up a VPS (Hetzner, Scaleway, DigitalOcean), npm install -g @openai/codex on it, SSH in from Blink inside a tmux session, run codex.

• Works when: you want maximum control, a box that is yours, and an iOS-native keyboard.
• Breaks when: you do not want to be a sysadmin. Patching, hardening, firewall rules, backups, none of that is optional on a box that holds your source code and your OpenAI API key.
• Cost: Blink Shell is $19.99/year, the VPS is $5 to $40/month depending on specs. No Android equivalent. Blink is iOS-only.

How do you set up Codex CLI on iPhone or Android?

You set up Codex CLI on iPhone or Android in about three minutes with Cosyra: install the app, sign in, confirm codex is on the PATH, paste your OpenAI API key into .bashrc, clone a repo, and run codex. No Node install, no SSH, no Termux ARM build issues. We measured 2 minutes 47 seconds on average across five cold starts on an iPhone 15 Pro and a Pixel 8 on 2026-04-16.

We measured this on 2026-04-16 on an iPhone 15 Pro and a Pixel 8: install to first codex prompt took 2 minutes 47 seconds on average across five cold starts.

Step 1: Install Cosyra and sign in

Download from the App Store or Google Play. Sign in with Apple, Google, or email. On first launch, we provision a fresh Ubuntu 24.04 container.

Step 2: Confirm Codex CLI is there

No install step, Codex CLI is baked into the image. Verify it in one command.

Codex CLI ships fast — the capture above is from 2026-04-16; the latest release on github.com/openai/codex/releases is 0.131.0 (2026-05-18). The install flow does not change, and we re-bake the image weekly so the version inside the container tracks upstream.

Step 3: Add your OpenAI API key

Get a key from platform.openai.com/api-keys. In the Cosyra terminal, set it as an environment variable and persist it in .bashrc so future sessions pick it up.

You pay OpenAI directly for tokens at their published rates. We do not proxy, meter, or see the responses.

Step 4: Clone a repo

Step 5: Run your first Codex prompt

Codex reads the relevant files, proposes a patch, and waits for you to approve. You see the diff on your phone screen before anything is written to disk.

What can you actually do with Codex on your phone?

You can refactor a module on the commute, review a teammate's PR from the couch, ship a 2 AM hotfix when you are on call, upgrade a dependency in a waiting room, and pick up the same session on an iPad when you switch devices. The container persists, so each of these slots into a 10 to 25-minute gap without any "get my laptop out" friction. Five real sessions we run below, with the prompts and diffs we typed.

Abstract claims about "mobile coding" are useless without concrete scenarios. Five micro-workflows we actually run from a phone.

Refactor a module on the commute

Subway, 14 minutes to your stop, one seat. Open Cosyra, cd into the service, run codex, type "rename the UserService.fetch method to getUser and update all callers, including tests." Codex lists the 15 files it plans to touch, you approve, it applies the patch, runs the tests, shows the green output. Commit and push before you surface.

Review a teammate's PR from the couch

Saturday morning, coffee, couch. A PR notification lands. You do not want to open a laptop. Fetch the branch in Cosyra, point Codex at it: "compare this branch against main, focus on security and missing error handling, write a short review." Paste the output into the GitHub PR comment from the mobile browser.

Ship a hotfix when you are on call

Pager goes off at 2 AM. You are in bed. Open Cosyra, pull, run codex with the stack trace, tell it to reproduce and fix. Review the diff on your phone, approve, push, watch the deploy pipeline go green. Go back to sleep. We did this twice in March 2026 and logged under 9 minutes from pager to green.

Upgrade a dependency in a waiting room

Dentist is running 25 minutes late. You have been putting off bumping an annoying dependency. Open Cosyra: "upgrade axios from 0.27 to 1.x, update any broken call sites, make sure tests still pass." Codex handles the migration one file at a time, shows you each diff, and you approve or reject as you go.

Test from an iPad without installing anything

iPad with a keyboard case is a surprisingly good Codex machine via Cosyra. Same container state as your phone. You pick up mid-session when you switch devices. The state persists because the container persists, not because we sync some client cache.

What are the real limits of running Codex CLI on a phone?

The real limits are no offline mode, OpenAI token spend adds up fast, phone keyboards slow long prompts, Codex defaults to looser permissions than Claude Code, and Docker-in-Docker is not guaranteed inside the container. We do not think Codex on a phone replaces a laptop for everything, but knowing where it stops helps you match the tool to the task instead of fighting the tool.

We do not think Codex on a phone replaces a laptop for everything. Here are the real constraints.

• No offline mode. The container lives in the cloud. No internet, no terminal. If you code on planes without wifi, a local setup like Termux (despite its own quirks) or a laptop wins.
• Your OpenAI tokens cost money. Codex is not free to run. A chunky refactor can burn a couple of dollars in tokens. Budget accordingly, or cap spending in the OpenAI dashboard.
• Phone keyboards are slow for long prompts. Voice dictation helps a lot. "Rewrite the auth middleware to support refresh tokens and add integration tests" is faster dictated than typed with thumbs. Bluetooth keyboard is faster still.
• Codex has looser default permissions than Claude Code. This is a Codex design decision, not a Cosyra one. Claude Code stops for approval on most shell commands. Codex will happily rm, curl, and pip install inside the workspace in its default mode. For phone sessions we recommend codex --ask-for-approval always, especially when you are one mis-tap away from approving something you did not read carefully. Some Codex power users will tell you this slows them down; on a phone, we think that slowdown is worth it.
• No Docker-in-Docker or Kubernetes guarantees inside the container. If your project needs docker to run its tests, verify that works for your specific case before committing to a phone-only workflow.

How does Cosyra compare to SSH and Termux for Codex CLI?

Cosyra wins for zero-setup and dual-platform (iOS and Android) phone use, Termux plus SSH wins if you already run a home server and are an Android user, and Blink Shell plus a VPS wins if you are an iOS user who wants a box you own. None of them are strictly best; each maps to a specific trade-off. The table below lines them up on seven attributes, re-verified 2026-05-19.

For the broader picture of all four AI agents on mobile (Claude Code, Codex CLI, OpenCode, Gemini CLI), see our pillar guide on AI coding agents on mobile. If you want the same walkthrough for Anthropic's tool, see Claude Code on phone. If you'd rather use Google's free-tier agent, see Gemini CLI on phone; for the open-source alternative that runs against any provider you have a key for, see OpenCode on phone. If you mostly drive Codex by describing the change in plain English instead of typing every edit, that's the workflow we cover in vibe coding on your phone.

Frequently asked questions