ChatGPT Codex is OpenAI’s coding agent across the ChatGPT desktop app on macOS and Windows, CLI, IDE integrations, mobile Remote Control, and Cloud. Existing standalone Codex app projects and settings carry into ChatGPT desktop. Local OpenAI clients can share CODEX_HOME configuration, instructions, skills, plugins, and MCP setup; Codex Cloud keeps its own environment and workspace configuration.
This guide takes you from zero to productive across all four surfaces, then into the workflows that actually ship code. Somewhere between “it works” and “it works fast” sits a collection of small techniques experienced Codex users rely on daily — keyboard shortcuts that save five seconds per turn, config tweaks that eliminate approval fatigue, AGENTS.md patterns that keep the agent on track. Individually each is minor. Stacked across a full day, they change how you work.
Install every surface, authenticate, configure approval modes, write your first AGENTS.md, connect MCP servers, hook up GitHub, run your first real task, review the output, and recover when things break. Everything you need to be dangerous in an afternoon.
Codex mode in the ChatGPT desktop app gives you parallel threads, worktrees, built-in Git tools, automations, skills, and an integrated terminal — all in a focused window separate from your editor.
The Codex CLI is open-source, built in Rust, and runs on macOS, Linux, and Windows. Interactive TUI sessions, non-interactive scripting, approval modes, image input, web search, and local code review — all from your terminal.
The OpenAI extension supports VS Code, Cursor, and Windsurf. JetBrains AI and Xcode 26.3+ provide separate native Codex integrations rather than builds of that extension.
Most coding agents lock you into a single interface. ChatGPT Codex instead offers desktop, CLI, IDE, Cloud, and mobile Remote Control entry points. Local clients can share local state, while cloud tasks use separately configured environments. The GPT-5.6 family spans Sol (frontier), Terra (balanced), and Luna (efficient), but there is no universal Sol default: Free and Go use Terra, while Plus, Pro, Business, and Enterprise can choose all three, subject to rollout and managed-workspace policy. In the API, the alias gpt-5.6 resolves to gpt-5.6-sol.
ChatGPT desktop app: Codex projects, parallel threads with worktree isolation, inline editing, PR review, multi-repo projects, automations, and background notifications
CLI: Interactive TUI or non-interactive codex exec for scripting, piping output, CI/CD integration, and headless automation on remote machines
IDE Extension: Agent mode in the sidebar with file context, @file references, cloud delegation, and auto-context syncing with the App
Cloud: Browser-based at chatgpt.com/codex, GitHub PR integration with @codex mentions, environment configuration, and long-running background tasks
Each of these articles is a list of focused, self-contained tips. Scan the headers, find what applies to your current workflow, and apply it immediately. The progression runs from foundational (setup and config) to surface-specific (App, CLI, Cloud) to strategic (AGENTS.md, advanced techniques, team collaboration).
New to Codex? Start with setup and work forward through the surfaces you actually use. Been using Codex for weeks? Skip to the advanced and team articles. Either way, the rest of the guide follows the shape of the work: planning a change, building it, testing it, shipping it, operating it, automating what repeats, and the reference material behind all of it.