Heidelberg AICurriculum

Codex - Full Course for Beginners

Tech With Tim ·2026-07-14 ·2 min read

Summary written by us from the video's transcript. The video, and everything in it, is Tech With Tim's work.

Learn how to set up and use the Codex desktop app, configure models and plugins, and automate coding tasks and computer actions through practical examples.

Takeaways

  • The desktop app provides the richest set of features, including model selection, voice dictation, and built‑in browser/terminal toggles.
  • Use plugins like Computer and Chrome to let Codex control local applications and web pages, enabling automation beyond code generation.
  • Select appropriate models and reasoning levels: GPT 5.5 medium for most tasks; switch to mini models when usage limits are a concern.
  • Organize work with projects and branches, leveraging the GitHub plugin for automatic repository creation and version control.
  • Slash commands and annotations let you issue precise actions—such as editing specific UI elements—in the integrated browser.

What is Codex and How to Access It

Codex is OpenAI’s tool for code generation, computer automation, and browser control. It runs as a standalone desktop application for Windows or macOS; Linux support isn’t available yet. You sign in with your ChatGPT subscription (as low as $20/month) or an API key (not recommended due to cost).

Other access points include an IDE extension for editors like VS Code, a CLI for server use, and a cloud version that runs in a browser. The desktop app offers the most features, so the tutorial focuses on it.

Navigating the Desktop Interface

The main chat window includes a model selector, voice dictation, approval modes (ask for approval, approved, full access), and a slash‑command menu. Right‑hand toggles reveal a controllable browser and terminal; a left sidebar provides additional views and settings.

Projects let you group chats by folder, and the profile menu gives access to settings, usage, and billing information.

Choosing Models and Reasoning Levels

Model options are GPT 5.5 (default for most tasks), GPT 5.4, GPT 5.4 mini, and 5.3 Codec Spark. Reasoning levels—light, medium, high, extra‑high—trade speed for response quality; higher reasoning can take up to ten times longer. For typical use keep GPT 5.5 with medium reasoning, switching to faster mini models only when credits run low.

Installing and Using Plugins

Plugins extend Codex’s abilities. Key ones demonstrated are the Computer plugin (enables OS control), Chrome extension (browser automation), and GitHub integration (auto‑creates repositories). Install each from the plugins pane, grant required permissions, and they appear in the slash‑command list.

Example: using “@computer” to open Notepad, write text, or launch Spotify via Chrome shows how Codex can manipulate both local apps and web pages.

Creating a Project and Generating Code

Start a new project by selecting a folder or creating one from scratch; the tutorial creates a “Chess Game” project with a main branch. After setting full‑access mode, a prompt describing a multiplayer chess MVP triggers Codex to generate multiple files, set up WebSockets, and launch a local server.

The generated code can be opened in an IDE (e.g., Cursor), inspected via the preview pane, or run directly from the integrated terminal.

Interactive Browser Annotations

Within the built‑in browser view you can take screenshots to the clipboard, annotate HTML elements, and issue targeted changes. For instance, annotating the board title or colors prompts Codex to modify the page accordingly, demonstrating fine‑grained UI editing.