Codex 101 — OpenAI Codex Guide

A bilingual guide to using OpenAI Codex in real projects
CLI · Desktop App · IDE Extension · Web/Mobile

Last updated: August 3, 2026
Live Site · Korean README

---

About

Codex 101 is a practical, bilingual guide for people learning or rolling out OpenAI Codex. It covers the main Codex surfaces, setup decisions, safety controls, configuration, MCP, workflow automation, and prompt patterns.

The guide is written from the official OpenAI Codex documentation first, then manually reviewed before publishing. Because Codex changes quickly, treat the live guide as a learning companion and always confirm critical model, pricing, security, and availability details in the official docs.

Start Here

• First-time users: read sections 04-06 for setup and first run, then 10 for sandbox/approval basics, and 14 for OpenAI Docs MCP.
• Professional users: read sections 12-14 for AGENTS.md, config.toml, and MCP, then 15-17 for sessions, automation, and prompt contracts.
• Example-driven readers: open the live site and the use-case gallery to start from real workflows instead of feature lists.

August 3, 2026 Refresh Notes

• Rechecked developers.openai.com/codex/models first. Recommended Codex models remain the GPT-5.6 family: gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna; gpt-5.5 remains previous-generation, gpt-5.2 / gpt-5.3-codex remain deprecated for ChatGPT-authenticated Codex, and gpt-5.4 / gpt-5.4-mini now have an August 31, 2026 ChatGPT sign-in Codex retirement notice.
• Tightened the Max/Ultra note from the current model and subagents docs: Max gives one selected model more reasoning time for a single hard task, while Ultra uses subagents for divisible work and may require enabling Ultra in model picker slider under Settings > Configuration.
• Rechecked pricing, Speed/Fast mode, remote connections, app/features, app/automations, config, MCP, hooks, plugins, skills, Sites, OpenAI API latest-model/changelog, and current Platform/API surfaces. Fast mode still includes GPT-5.6, GPT-5.5, and GPT-5.4 while GPT-5.4 is available; GPT-5.6/GPT-5.5 use 2.5x Standard credits, GPT-5.4 uses 2x, and API-key workflows stay on API token pricing. Saved ChatGPT-authenticated Codex configs should replace gpt-5.4 with gpt-5.6-terra and gpt-5.4-mini with gpt-5.6-luna before August 31, 2026.
• Rechecked the latest Codex changelog and added CLI 0.145.0: paginated thread history, expanded /import, Bedrock endpoint/auth updates, audio inputs/tool outputs, realtime V3, multi-agent V2 stabilization, inline visualization links, and Windows/MCP reliability fixes.
• Added the July 23 ChatGPT desktop app 26.715 update: ChatGPT Voice can coordinate Chat, Work, and Codex tasks in the desktop app and through Remote on iOS; local projects can include multiple folders with one primary folder for Git, new chats, AGENTS.md, skills, and config.toml.
• Reflected OpenAI API GPT-5.6 guidance separately from Codex product guidance: gpt-5.6 routes to gpt-5.6-sol; GPT-5.6 adds Programmatic Tool Calling, Multi-agent beta, explicit prompt caching, persisted reasoning, max reasoning effort, Pro mode, and original/auto image detail preservation. API hard spend limits and the July 29 Terraform provider remain Platform/API lane items. The July 30 price-performance post lowers GPT-5.6 Luna API pricing by 80% and Terra by 20%, says those lower Luna/Terra costs are reflected in Codex and ChatGPT Work usage counting, and introduces API Fast mode for Sol at up to 2.5x faster Standard speed at twice the price. The July 30 ARC-AGI-3 post reinforces the Platform/API lane: use Responses API, retained reasoning, and compaction for long agent runs.
• Re-audited the official use-case catalog: official individual pages 101 after excluding collection pages, local cards 101, promptEn gaps 0, promptKo gaps 0, official Starter prompt mismatches 0, line/token structure regressions 0, and no new or changed Starter prompts.
• Rechecked Remote connections against current official docs. iOS/Android can control awake/online macOS or Windows hosts, and supported Mac or Windows desktop apps can continue work on another host from Settings > Connections > Control other devices. Host files, credentials, permissions, plugins, browser setup, Computer Use, and local tools remain on the connected host.
• Rechecked openai.com/index and OpenAI Developers/API docs. No Codex product model recommendation change was needed, but the July 30 GPT-5.6 price-performance and ARC-AGI-3 posts update the pricing/API-operating context described above. The NTT DATA Group Codex story is treated as Enterprise adoption and incident-analysis automation context, not as a general feature or model-order change.
• Person-level practitioner signals were rechecked in Chrome against X, LinkedIn, and Threads. Section 20 adds VB's mid-task decision prompt tip and links it to the official config schema boundary: default_mode_request_user_input appears as a schema feature key, while behavior guidance stays practitioner context until the current app behavior is verified.

June 12, 2026 Refresh Notes

• Reflected the June 11 Codex app 26.609 release: rate-limit reset banking, Developer mode for Browser/CDP work, /init in the app composer, expanded Enterprise Computer Use, and Windows Computer Use per-app access controls.
• Rechecked pricing, quickstart, config/profile, app settings/features, remote connections, Sites, Bedrock, MCP, hooks, plugins, and skills through the fresh Codex manual. No model-order, remote-control, or Sites setup reversal was needed, but Quickstart now clearly foregrounds the standalone installers.
• Audited all 101 official Codex use cases after excluding collection pages. The catalog count, local/official IDs, promptEn, promptKo, and multiline Korean prompt shape are clean.
• Rechecked OpenAI launch/customer-story surfaces. The June 11 Ona acquisition announcement is captured only as pre-closing enterprise execution direction, not as an enabled feature, model, plan, platform, preview/GA, or feature-flag change.
• Person-linked practitioner tips stay separate from product documentation. The public guide points readers to original posts or official corroborating docs, and only verified, reader-useful guidance is promoted.

What's Covered

Section	Topic
Start Here	Quick paths for beginners and teams
01-03	Codex ecosystem, product surfaces, supported models
04-05	Requirements, pricing, installation, authentication
06-09	CLI, Desktop App, IDE Extension, Web/Mobile
10-14	Approvals, sandboxing, slash commands, AGENTS.md, config.toml, MCP
15-17	Session strategy, automation, prompt contracts
18-21	Advanced usage, FAQ, references, practitioner tips

Languages

The site currently supports Korean and English. Use the language toggle in the top-right corner of the live page, or update the translation blocks in i18n.js when contributing.

Local Preview

This is a static site. No build step is required for basic review.

open index.html
open use-cases.html

Project Structure

codex-101/
├── index.html          # Main Codex 101 guide
├── use-cases.html      # Use-case gallery
├── app.js              # Theme, language, navigation, interactions
├── i18n.js             # Korean/English translation content
├── use-cases-data.js   # Use-case gallery data
├── style.css           # Responsive styling
└── images/             # Screenshots and explainer visuals

Contributing

Contributions are welcome for corrections, translations, screenshots, examples, and workflow tips.

git clone https://github.com/<your-username>/codex-101.git
cd codex-101
open index.html

After reviewing your change locally, open a pull request with a short explanation of what changed and which page or section you checked.

License

This project is released under the MIT License.

---

Made by @swhan0329
Built with the help of AI. Pull requests are welcome.