Step-by-Step Beginner’s Guide to Getting Started with Codex

Why users are switching to Codex

Three factors drive the migration from Claude Code to Codex: frequent account blocks on Claude Code, a rapid series of capability upgrades in Codex (computer control, browser automation, plugin ecosystem), and a stronger underlying GPT model that narrows the performance gap.

Codex product forms

Codex App – a desktop graphical interface that can chat, read and edit local files, control the computer, and connect to external tools.

Codex CLI – a terminal‑only tool that runs without a UI and is optimized for project‑level automation.

The App acts as a universal assistant; the CLI serves as a project partner. They can be used together.

Prerequisite: Access ladder

In China Codex requires a proxy (referred to as the “ladder”) for every operation—download, registration, login, and usage. Obtain one before proceeding.

Account setup

Register an OpenAI/ChatGPT account. Existing ChatGPT accounts work; new registrations may need an overseas phone number.

Choose a tier:

Free – limited quota.

Plus – $20 / month, sufficient for most daily tasks.

Pro – $200 / month, for heavy usage.

Installation

Codex App

Download from https://openai.com/codex. On macOS drag the app into “Applications”; on Windows run the installer. Open the app, log in with the ChatGPT account, and keep the ladder active throughout.

Codex CLI

Install Node.js ≥ 18, then run: npm install -g @openai/codex If the default registry is slow, use the Alibaba mirror:

npm install -g @openai/codex --registry=https://registry.npmmirror.com

After installation, start the CLI with codex and follow the browser‑based login flow (ladder required).

Basic operations in the App

Left panel – navigation bar (new chat, projects, plugins, automation, settings).

Middle panel – conversation area where you type prompts.

Right panel – result area showing generated files, previews, and diffs.

Two entry points for work:

New conversation – pure chat, no file access.

New project – binds a local folder; Codex can read and modify files inside.

When creating a project, select a permission level. Higher permissions give Codex more autonomy but increase risk; start with the medium level and raise it only after understanding the prompts.

Practical examples

1. Generate a Word research report

Prompt example:

帮我写一份《2026年新能源汽车行业调研报告》，要求：
1. 自己搜索最新的行业数据
2. 包含市场规模、头部品牌、技术趋势、政策环境
3. 输出为 Word 文档

Codex opens its built‑in browser, searches for the latest data, compiles market size, top brands, trends, and policy information, and returns a formatted 4,200‑word Word document within minutes.

2. Create a PPT

Use the built‑in Presentations plugin. For more polished slides, combine the plugin with Skills and MCP (not detailed here).

3. Computer control – CNKI paper search

Prompt example:

帮我去知网搜索"大模型"相关的论文，筛选最近一年的，按照引用量排序，
把前10篇的标题、作者、摘要整理成一个表格给我

Codex launches the internal browser, performs the search, applies the time filter, extracts titles, authors, and abstracts, and returns a table. The whole process took about 7 minutes, whereas manual copying would require half an hour.

4. Automation scenarios

Schedule daily email summaries.

Generate weekly project reports.

Monitor website changes.

5. Plugins

Available built‑in plugins:

Spreadsheets – table handling.

Presentations – PPT generation.

Browser Use – web‑search and data extraction.

Computer Use – UI interaction with desktop applications.

Gmail / GitHub / Google Drive – external account integration.

Plugins are capability packs; Skills describe workflows; MCP connects external tools.

CLI usage

The CLI shares sessions with the App, so actions performed in one view appear in the other.

Example: build a Snake game

codex

Then prompt:

帮我用 HTML + JavaScript 写一个贪吃蛇小游戏，要求：
- 用方向键控制
- 有计分功能
- 页面好看一点

Codex creates the HTML and JavaScript files; opening the HTML file in a browser runs the game.

CLI permission modes

Suggest – only provides suggestions, never edits files.

Auto Edit – edits files but asks for confirmation before each command.

Full Auto – fully autonomous editing and command execution.

Start with Auto Edit if unsure; switch to Full Auto for end‑to‑end automation.

Choosing between App and CLI

Use the App for report writing, document generation, and computer control.

Use the CLI for project‑level coding, tool development, and deeper integration.

Both can run simultaneously without interference.

Comparison of three Codex modes

ChatGPT – web/app chat assistant; best for Q&A, copywriting, image generation.

Codex App – desktop AI workstation; includes all ChatGPT capabilities plus local file access and computer control.

Cloud Codex – server‑side AI agent; runs tasks when the computer is off and is not limited by local resources.

Pitfalls and tips

Never approve permission dialogs blindly; ask Codex “what will this permission do?” before confirming.

Iterate within the same conversation to refine results instead of starting a new chat.

Install only the built‑in plugins at first; add others when needed.

If the CLI throws an error, paste the error message back to Codex for automatic diagnosis and fix.

Keep the access ladder active throughout download, login, and usage; loss of the ladder breaks the connection.

Additional practical tips

Use @ in the dialog to select a specific file in the project directory.

Check quota by clicking the usage indicator; a Plus account covers typical daily usage.

Select the latest model (GPT 5.5) and the “regular” speed; “fast” mode is 1.5× speed but consumes more quota.

App and CLI sessions are synchronized; a conversation started in the CLI appears in the App UI.