Published: 2026-07-15

Codex Full Course: Build and Automate Anything in the Super App

Chapters / key moments (click to jump — plays here on the page)

This hour-long tutorial walks non-technical users through the Codex desktop "super app" — the same window that holds both the normal ChatGPT chat and the Codex build environment. The video covers downloading and logging in, using plan mode to architect a project before building, choosing an access level and effort/speed settings, running locally or in the cloud, forking conversations into git branches, and connecting skills (like Gmail and presentations) so Codex can build websites and run scheduled automations in the background.

Source video

"CODEX: Beginner to Expert Course! (1 HOUR!)" by Julian Goldie SEO — Watch on YouTube →

Note on the models named here

The video references a Codex "super app" and frontier models ("Soul," plus "Terra" and "Luna"). Treat product and model names as the creator's framing — always confirm current names, availability, and pricing on the official source before relying on them. See our ChatGPT / Codex hub for verified guidance.

Key Takeaways

One "super app," two modes. The same application holds the regular ChatGPT chat (pulls in your existing conversations) and the Codex build environment — chat for back-and-forth, Codex for building bigger projects locally and running automations 24/7.
Plan mode before build mode. Plan mode has Codex think through the project in depth (branding, pages, linking, scope) and ask clarifying questions before writing anything. You can switch to action/build mode afterward. The creator notes you don't always need plan mode for simple builds.
Access levels control autonomy. "Full access" lets it run autonomously and make decisions for you (fast, less control); "ask for approval" keeps you in the loop on every change (slower, safer). Even in full access, deleting or replacing existing files still prompts for permission first.
Effort and speed dials. Higher effort = better results but slower; lower effort = faster but less capable. A "speed mode" runs roughly 1.5× faster for demos; normal mode is about 50% slower.
Local or cloud, with git branching. Runs execute locally by default and can switch to cloud at any time. You can fork a conversation into branches (e.g., one for design, one for the plan), and manage runs as tasks — archive, pin, or open the created files in Finder.
Skills make it compound. Connecting plugins/skills (Gmail, presentations, social posts) lets Codex send emails, build slide decks, and run daily looped schedules — the demo builds a productized-agency website, a gamified habit-tracker prototype, and a daily Gmail-drafts automation.

What you can actually set up from this

Extracted from the video's own transcript — the specifics the original summary left out.

Reproducible steps

Start a project
In the Codex app (the desktop "super app" that holds both plain ChatGPT chat and Codex), create a new folder for the project, then choose Local or Cloud. Local keeps files on your machine; Cloud runs sandboxed, with no connection to your local files, which he recommends only when you need isolation. A usage-remaining indicator shows how much of your plan is left.
Plan before you build
Use plan mode first: give it the brief, attach files such as a slide deck of your offers, brand colours and links. Codex asks clarifying multiple-choice questions (for example whether to replace the existing site or host a separate subdomain, which framework, what clients can do after login), then produces a detailed plan. He cites Kilo Code research that planning first gives better outputs. When satisfied, say to implement the plan and it builds.
Choose the access level
Full access lets it run autonomously and fast with less control; ask-for-approval slows it down but keeps you in control. Even on full access he saw it ask permission before replacing existing files.
Tune speed, effort and model
Speed mode runs about 1.5 times faster (normal mode is about 50% slower, by his estimate). The effort dial trades speed for quality. The model picker offered Sol, Terra and Luna; he prefers Sol for Codex.
Add plugins and computer use
Plugins page: search, enable and authorise an app. Tag the Computer Use plugin and Codex drives your browser; in his demo it drafted a social post, accepted an edit (adding a call to action) and posted only after he said to.
Run several things at once
Tasks run in the background, so a website plan, an agent-OS project and a slide deck can progress in parallel. A built-in preview pane shows the locally hosted project beside the chat. Conversations can be forked into branches, archived, pinned, or opened in Finder.

Gotchas

For every new task, select a separate folder (or start from scratch) so a new task does not overwrite files from another conversation.
Frontier models are expensive to run: he says token use is the main constraint with Sol and promises optimisation tips later; the transcript is cut off before them.
He also describes an agent-OS demo: from a GitHub link he told Codex to install it, set up Hermes Agent and build a mission-control dashboard. Details of that install are not given.
Promotional course; the "hundreds of projects tested" claim is the speaker's.