The Ultimate Codex Guide: From Installation to Real-World Use — AI That Operates Your Computer for You

Title

Codex: The Ultimate Guide — From Installation to Hands-On Practice, AI That Directly Operates Your Computer

Meta Description

A Bilibili creator demonstrates Codex in action: from installing via the app store and configuring work mode and project folders, to organizing messy files into Excel and building a visual dashboard with Plan Mode. Learn how this AI tool that directly operates your computer can get work done for you.

Article Content

AI assistants are no longer just for chatting. Bilibili creator "Computer Story" (计算机物语) released what they call "the most detailed Codex tutorial on all of Bilibili," demonstrating how this tool can directly operate a computer, organize files, generate reports, and even build a visual dashboard. This article is based on that video and summarizes Codex's core usage and key design.

What Is Codex: Crossing from "Chatting" to "Getting Work Done"

Codex offers a chat interface just like Doubao and DeepSeek^Ad, but the essential difference is this: Doubao can only chat, whereas Codex can directly operate your computer and get work done for you. According to the creator's demo, it can organize files, create reports, read work content, write weekly reports, and even directly operate a browser to build complex software. You can also set tasks to run automatically while you sleep, so they're done when you wake up.

"It can do almost anything you can think of, and you don't need any technical knowledge."

The installation barrier is extremely low: Codex is available in the Microsoft and Apple app stores—just search and install. Once opened, the interface resembles the web version of Doubao, with an input box in the center and a left sidebar offering extensions like plugins and automated workflows.

Codex is an AI coding agent launched by OpenAI. It was initially released as a cloud-based software engineering agent and later gradually expanded to desktop. Unlike ordinary conversational AI, Codex has "agentic execution" capabilities: it can read and write local files, run command-line instructions, operate a browser, and—once authorized—autonomously complete multi-step tasks. This capability relies on the tool use mechanism of large language models—the model selects and calls external tools such as the file system and terminal based on task requirements, rather than merely generating text. This is precisely why Codex can organize folders, extract data from PDFs, and generate Excel reports. Users don't need to know programming syntax; they simply describe the goal in natural language, and the model plans and executes the corresponding operations.

Key Configuration Before You Start: Work Mode and Project Folders

The creator suggests new users first adjust one setting: work mode. It controls how much technical detail Codex shows you. After switching to "Everyday Work," Codex's capabilities remain unchanged, but it no longer displays a large amount of technical detail, making for a cleaner experience.

Another core concept is "projects." Since Codex works directly on your computer, it needs a working folder, and these folders appear as projects in the sidebar. The creator categorizes different tasks into different folders so the corresponding content can be found at a glance. After selecting the project for the current conversation below the input box, Codex can read all the files in it to complete the task.

Practice Case 1: Organizing Messy Files and Extracting Data into Excel

The creator prepared a real folder called "Computer Things & Cafe" containing a mix of sales records, invoices, and inventory lists, in formats spanning spreadsheets, PDFs, and images. The task: tidy up the files and extract the information and data into a single Excel file.

A few minutes later, the task was complete. Codex organized the files by purpose, generated a detailed data table, categorized sales data and product rankings, and even extracted line items from PDF invoices. More importantly, the creator deliberately planted some "traps" in the files, and Codex successfully discovered missing transaction records as well as anomalies in two invoices.

This type of task—reading computer files, organizing and summarizing them, and generating reports—is exactly the everyday scenario Codex excels at.

Practice Case 2: Building a Visual Dashboard with Plan Mode

For a more complex task, the creator demonstrated using the command line add to specify files, having Codex read the previously generated summary table, and requesting a visual dashboard showing business conditions such as sales, expenses, costs, and inventory, along with an analysis of items that need attention.

For complex tasks, the creator recommends enabling "Plan Mode." In this mode, Codex doesn't execute directly; instead, it first plans a detailed execution checklist, confirms the task details, and then executes.

During execution, Codex proactively asks questions to confirm details, and users can adjust the plan or submit it for execution directly. The final dashboard contained all the requested content, along with suggestions and key items. The creator then requested a "more modern, intuitive, Apple-style" theme, and Codex completed that smoothly as well.

Plan Mode is a common interaction design in agentic tools. When enabled, the AI doesn't start acting immediately; instead, it first outputs a structured execution plan listing the steps to be completed, the files involved, and the expected results. Users can review, modify, or supplement requirements item by item, and then let the AI officially execute once everything is confirmed. This "plan first, execute later" workflow significantly reduces the risk of complex tasks going off track, and is especially suited to scenarios involving multiple files and multiple processing stages. The counterpart to Plan Mode is direct execution mode, which is faster but carries higher rework costs if the AI's understanding is off.

An Interesting Detail: Desktop Pet

Typing a slash in the input box brings up the "pet" feature. A small pet appears on the desktop; beyond being cute, it can also display all the tasks currently being executed, and clicking it jumps to view progress. Its appearance can be changed in settings.

Usage Tips and Boundaries

The creator repeatedly emphasizes that Codex doesn't require dedicated technical study—a few sentences can produce very cool things. But judging from the demo, two habits can significantly improve the experience: first, create project folders categorized by task; second, enable Plan Mode for complex tasks to confirm before executing.

As for the actual business value of the dashboard, the creator frankly admits not being a sales professional and thus unable to judge whether it's truly useful, suggesting that relevant practitioners evaluate it themselves. This also reminds us: Codex can efficiently complete work at the execution level, but business judgment still needs to be overseen by humans.

One-line Summary

A hands-on Codex tutorial: from chatting to directly operating your computer to get work done—organizing files, generating reports, and building dashboards.

Paragraph Summary

Based on Bilibili creator "Computer Story"'s hands-on Codex tutorial, this article summarizes the core usage of this AI tool. The essential difference between Codex and chat AIs like Doubao and DeepSeek is that it can directly operate a computer to complete real tasks, such as organizing files, extracting data into Excel, and building visual dashboards. The article covers key configuration before starting (work mode, project folders) and, through two practice cases, demonstrates the file organization and dashboard-building workflow, emphasizing that complex tasks should enable Plan Mode to confirm before executing. It also mentions the desktop pet feature and usage boundaries: Codex can execute efficiently, but business judgment still needs to be overseen by humans.