Finally, turn on push notifications for ChatGPT. You want to know the moment a task finishes so your review loop stays tight. This setup mirrors how MCP connects AI to your tools, giving the agent structured context instead of making it guess.
Is Codex free on ChatGPT mobile?
Yes. As of May 14, 2026, Codex is available on every ChatGPT plan including Free. The preview tier gives you a limited number of Codex runs per day. OpenAI has not published exact quotas for Free users, but early reports suggest roughly five to ten tasks per day on the free plan.
Pro and Team tiers unlock higher throughput and longer sandbox runtimes. If you plan to use Codex as a daily tool, the Pro plan is worth it for the faster queue times alone. The cost math is similar to what we broke down in how to reduce AI API costs without changing your code. Spend where the bottleneck is. For most builders, the bottleneck is review speed, not task count.
Both iOS and Android get the same feature set. There is no difference between platforms. You can start a task on your iPhone and check the result on an Android tablet later, since tasks live in the cloud.
What are the best prompts for Codex on mobile?
Short, specific prompts win on mobile. You are typing on a phone keyboard. You are reviewing on a small screen. Every extra word in your prompt should earn its place.
Here are five prompt patterns that work well:
- Bug fix: "Fix the TypeError in src/utils/date.ts line 42 where parseISO receives null."
- Test generation: "Write unit tests for the createUser function in services/auth.ts. Cover happy path and duplicate email error."
- Doc update: "Update README.md to reflect the new env variable DATABASE_URL and add setup instructions for local dev."
- CI fix: "The GitHub Actions lint step fails on the latest commit. Fix the ESLint config to allow optional chaining."
- Small refactor: "Extract the email validation logic in signup.ts into a shared validator in utils/validate.ts."
The structure is always: task, scope, and one constraint. That is it. Long briefs with background context belong in your AGENTS.md file, not in a mobile prompt. The agent reads AGENTS.md every run, so your prompt can stay lean.
This prompt discipline is useful beyond Codex. The same pattern applies if you use ChatGPT prompts for executive assistants or any other LLM-driven workflow. Tight input, clear output.
How does Codex mobile compare to Codex on desktop?
The agent behind both is the same. Codex GPT-5.4 runs in the same cloud sandbox whether you trigger it from a browser or the ChatGPT app. The differences are all about the review surface.
On desktop, you see a full diff viewer with syntax highlighting, side-by-side file comparisons, and inline comments. On mobile, you get a simplified diff view, one file at a time, with swipe to approve or reject. It works, but it is slower for anything over 50 lines.
Desktop also lets you edit the agent's code directly before committing. Mobile does not. Your options on phone are approve, reject, or send a follow-up prompt asking for changes. This is fine for small tasks. For anything complex, the back-and-forth loop on mobile adds friction.
The practical rule: use mobile to start and approve. Use desktop to debug and iterate. Think of your phone as the dispatch console, not the workbench. If you are comparing AI coding tools more broadly, our Cursor vs GitHub Copilot breakdown covers where each tool fits in a 2026 stack.
What are the pitfalls of running Codex from your phone?
Diff review fatigue. A 200-line diff on a six-inch screen is painful. You will miss things. Scope your tasks so diffs stay under 50 lines. If a task balloons, reject and break it into smaller pieces.
Network drops do not kill tasks. The cloud sandbox keeps running even if you lose signal. But a slow connection makes the review loop frustrating. Download the diff, lose connection, then have to reload. Not a deal-breaker, just annoying.
Security deserves attention. Mobile grants the same repo access as desktop. If your phone is compromised, so is your code. Audit your connected repos quarterly. Remove repos you no longer actively work on. Use GitHub's fine-grained personal access tokens instead of broad org-level access.
Notification noise. If you queue five tasks, you get five pings. Set a focused notification channel or schedule review windows instead of reacting to every alert. This is the same batching principle that makes automated AI morning briefings work. Batch input, process once, move on.
Platform gaps. No GitLab or Bitbucket support yet. No direct file editing on mobile. No terminal access to the sandbox from the app. These are real limits today, not just early-access quirks.
When should you start using Codex on mobile?
Now. The barrier is zero if you already have ChatGPT installed and a GitHub repo. Connect one repo, try one small bug fix, and review the diff on your phone. That first loop will teach you more than any guide.
The real skill is not learning the tool. It is learning to scope tasks for a small screen. Start with green-zone tasks: test generation, doc updates, lint fixes. Build the muscle of writing two-sentence prompts. Graduate to yellow-zone refactors once your AGENTS.md file is dialed in.
Codex on mobile does not replace your IDE. It fills the gaps between desk sessions. The commute, the lunch break, the waiting room. Those dead minutes become shipping minutes.
For more on how AI tools connect to your existing stack, read our full guide on Model Context Protocol: How MCP Connects AI to Your Tools.
If you are building with AI in Southeast Asia or want to meet others doing the same, check out genai.club and the GenAI Summit Asia.
FAQ
Is OpenAI Codex on mobile actually free?
Yes, as of May 2026 Codex is available in preview on the Free tier of ChatGPT for iOS and Android. Free users get a limited number of Codex runs per day. Pro, Plus, and Team plans unlock higher usage caps and priority queue access. You do not need a separate Codex subscription.
What tasks work best with Codex on a phone?
Small, well-scoped tasks shine on mobile: single-file bug fixes, generating unit tests, updating docs, fixing CI configs, and small refactors. Anything that produces a diff under ~50 lines is comfortable to review on a phone screen. Large architectural changes or multi-file rewrites are better handled on desktop where you can navigate the full codebase.
How do I connect my GitHub repo to Codex on mobile?
Open ChatGPT on your phone, go to Settings, find the Codex section, and authenticate with GitHub via OAuth. Select which repos Codex can access and set a default branch. The connection persists across mobile and desktop sessions. You can also add an AGENTS.md file to your repo root to give Codex standing instructions about your project.
Does Codex mobile work offline or need constant connection?
You need an internet connection to submit a task, but the actual code execution happens in OpenAI's cloud sandbox. If you lose signal after submitting, the task continues running. You can review results when you reconnect. However, approving diffs and merging PRs requires an active connection.
What is AGENTS.md and do I need it for mobile Codex?
AGENTS.md is a markdown file you place in your repository root that gives Codex persistent context about your project: coding conventions, test commands, file structure notes, and constraints. It is not required but significantly improves output quality, especially on mobile where you want to type shorter prompts and rely on the agent already knowing your project norms.