Follow-up: Our guide to goal mode and mobile remote control covers goals for longer tasks, connected computers and surface-specific limits.
If you subscribe to ChatGPT Plus or Pro, take a look at the included Codex access. The name makes many people think of programming first. This guide starts with a small file or document task you can verify yourself.
Codex can also work for people who do not program: save an email draft, sort a table, organize meeting notes or translate documents. What is possible depends on the files, tools and permissions in the chosen environment. An external service such as Notion needs an appropriate connection.
This article is for you if you already know and use ChatGPT but never understood why Codex matters. I will explain it in plain language, without jargon, and walk you through the first five minutes.
What Codex actually is#
Codex is accessible through several surfaces: web, desktop, terminal and editor extensions. Account and plan connect access, but execution location, tools, model choice and memories can differ. This guide starts with a small, verifiable task rather than a fixed number of apps.
The most important surface for you is Codex Web. You reach it directly from ChatGPT through the sidebar or via the URL chatgpt.com/codex. It looks like a second variant of ChatGPT but behaves differently. Instead of answering, Codex Web runs a task in the cloud. You can close the browser, go for a walk, come back later, and Codex will have finished the job.
You can follow supported Codex work from the ChatGPT mobile app. Remote connects to a Mac or Windows PC; access depends on rollout and workspace. The phone can start tasks, show progress and present actions for approval. The connected computer executes remote work and must stay awake and online.
The third is the Codex Desktop App for Mac and Windows. It launched in February 2026 and is aimed at power users who want to manage multiple tasks in parallel. Conceptually it compares well to the Claude Desktop App many people already know from the Anthropic ecosystem.
The fourth and fifth surfaces are a VS Code extension for code editors and the Codex CLI for the terminal. Both are for professionals. In this guide we ignore them.
Why regular ChatGPT users avoid Codex#
I hear the same reasons from clients. They think Codex is only for programmers. They think they need a second subscription. They think they need technical knowledge. They think Codex and ChatGPT do the same thing.
Codex is included in Free, Go, Plus, Pro, Business, Edu and Enterprise, with different limits and possible workspace approvals. Features also depend on surface and account. Check access in your account and current plan documentation before paying for another plan.
Codex is not complicated. Codex Web is as simple as ChatGPT itself. You type a task and click a button. Nothing more happens when you get started. There is no setup beyond connecting your GitHub account once if you want to try real code tasks. For everything that is not code, you do not even need that.
ChatGPT can also edit files, use tools and execute tasks. Codex emphasizes longer work with files, code, commands and verifiable outputs. Either can suit an email draft. What matters is the tools available and the goal: text, a saved file or a change in a working environment.
Your first Codex moment in five minutes#
Sign in with your ChatGPT account and open the Codex or work area offered in your surface. Navigation can change; use the official OpenAI Codex entry when needed. An organization administrator may control access.
Choose an environment that can actually access your material. Local file work needs an authorized folder; cloud work needs uploaded files or a connected service. Describe the output and review criteria. The earlier “Code” and “Ask” buttons do not reliably describe every current surface.
For the first test, use a sanitized email or short sample text. Ask: “Draft a professional reply, at most 120 words, identify the two open questions, and save the draft if file access is available.” Check content and location. Sending is a separate step requiring the appropriate tool and your approval.
That is all you need to understand at first. Everything else is fine tuning.
What you can do today without writing a line of code#
For solo founders, office workers and small teams these are the ten most useful applications I have seen in practice.
You can summarize PDFs, clean tables, format quotes or compare versions. Codex can use files and programs when its environment provides them. ChatGPT also has file capabilities; a product name alone does not make a summary more reliable. Request source locations, compare known numbers and check the result against the original.
You can turn meeting notes into clean to-do lists. You can translate documents with consistency checks, which solves the problem that the same phrase often gets translated differently in each paragraph. You can organize a Notion board if you paste it in as text. You can summarize research with source links. You can generate data visualizations from raw tables. You can prioritize and sort bug reports or support tickets. You can build onboarding trackers for new hires.
You do not need to program these tasks yourself. You do need to provide material, access and the required output format. Simple local file tasks do not need GitHub; some cloud coding workflows need a connected code source.
The iOS app in detail#
In the current ChatGPT iOS app, open the available Codex or Remote area. Choose the connected computer and check an existing task. If missing, check app version, workspace approval and current Remote instructions. Installing the app alone does not guarantee access to every Remote feature.
For example, start revising a concept at your desk and inspect progress on your phone later. The computer remains reachable for this Remote path. A task explicitly started in the cloud has different prerequisites. Check the chosen execution location when starting.
That is why the iOS app becomes the main entry point into Codex for many users during the day, not the browser at their desk.
Codex Web or Desktop App, which one fits you#
Both paths lead to the same result, but they speak to different users.
A cloud task suits material available there when you do not want to use a local computer for execution. Upload files or connect the supported source and check available tools. Web and desktop are not universally equivalent: local files, programs, configuration and memory can differ. Cloud tasks still need appropriate access and output review.
Desktop is useful for local folders, programs and several work streams. Local execution does not mean all data stays on your computer: model calls can transmit relevant content to OpenAI. Check data controls, plan and material approvals.
For the first month I recommend most users stick with Codex Web. Whoever is still interested after that and wants more, installs the Desktop App on top.
What Codex costs#
Codex usage is included in supported ChatGPT plans. Allowance depends on plan, model, task size and execution path. Local and cloud usage can share limits. Check the current usage window and official pricing rather than assuming a fixed number of sessions.
Published limits estimate different task types, not a guaranteed weekly number of long sessions. Track actual usage on representative tasks. When capped, check reset time, available models and possible additional credits in your account. Prices and taxes vary by country and payment channel.
Business, Enterprise and Edu do not use workspace data for training by default. Personal accounts, including Plus and Pro, have a training setting. This does not replace reviewing retention, access and approved data. A plan label is not a blanket privacy guarantee.
If you do not know which plan fits, stay on Plus and reassess after three months whether you are hitting limits.
Pitfalls I see in practice#
If the entry is missing, check account, workspace permission, app version and official documentation for your surface. April navigation need not remain the same in September.
If an environment requires a code source, check whether it fits your task. A desktop environment with an authorized folder often suits local document work better than a cloud repository task. Choose the correct execution path rather than skipping necessary permissions.
A smaller available model can help with limits, but neither fixed savings nor unchanged quality is guaranteed. Select from the actual choices offered and compare usage and output on your task.
Codex has native memories. Local clients use a separate store under ~/.codex/memories, off by default. Settings and /memories control its use; it is not automatically identical to ChatGPT memory on the web or Work. Use files such as AGENTS.md for authoritative project rules too. An external MCP store is needed only if your workflow requires additional shared storage. Our memory guide explains the distinctions.
The ramp in five realistic phases#
Phase one is the first encounter. Five minutes, one non-coding task, the aha moment.
Phase two is the first real coding moment, if you want to try code tasks. Connect GitHub once, open an empty repo, give a simple instruction like "generate a README file", accept the result as a pull request.
Phase three is mobile control: connect a computer, start a small task and inspect progress, approvals and output in the app.
Phase four is the Desktop App if you want more. Install in ten minutes, log in, create a workspace folder, try parallel sessions.
Phase five is the memory layer if you need one. That is optional, that is a power-user step, that is not a prerequisite for getting value out of Codex.
In practice most users find real value in phases one and two and then decide for themselves whether phases three and four are worth it.
What is still coming in 2026#
Work surfaces develop quickly. An announced integration or roadmap does not mean a specific feature is available in your account. Base setup on current official instructions.
The useful skills remain: specify an output and review criteria, limit access appropriately and inspect delivered files or changes. These transfer between surfaces.
Do not plan work around an unavailable product promise. Test the path your account actually offers today.
When Codex is not for you#
To keep this text honest. If you only use ChatGPT for short knowledge questions, if you do not have recurring tasks an assistant could run, if you do not need file operations but only answers, you do not need Codex. You are paying for it already but you lose nothing by leaving it alone.
For recurring table, document and file tasks, Codex can take on part of the work. A test including preparation and review shows actual time saved. A feature list cannot promise to halve your working time.
What to do next#
If you have never been in Codex, go to chatgpt.com/codex now. Enter a task you had to do anyway today. Watch what happens. That is all it takes for a first impression.
If using the mobile app, check the available Codex or Remote area and connect a supported computer following the official guide. Start by testing a small task and its approvals.
If you have tested it for a while and want to decide whether the Desktop App makes sense for you, go to openai.com/codex and download the installer for Mac or Windows. Log in with your ChatGPT account. In ten minutes you will have a workspace and a first multi-agent session running.
And if you are wondering how to integrate all of this into a bigger workflow, take a look at our AI systems. At StudioMeyer we build similar setups for SMB teams who were in the exact situation you are in today. Paid tools that go unused, routine tasks that could disappear into AI, a desk that could get less crowded.
Sources and product status#
Codex access and data terms / OpenAI usage and pricing / Local Codex memories / OpenAI data controls
