OpenAI explained how to configure Codex: personalization, level of detail, and access permissions

OpenAI released a brief guide on Codex settings — a working AI assistant for computer tasks. The company shows which parameters to check in the first few minutes so the agent works predictably, doesn't stall midway through a task, and doesn't overwhelm the interface with unnecessary details.

Where to Start

OpenAI recommends not trying to learn the entire list of parameters at once. The settings menu is located in the bottom left corner of the Codex interface, and for the first runs, it's enough to go through several basic items. The approach is practical: first remove everything that could disrupt a long task, then choose a convenient display style, and only after that adjust the agent's behavior to your work rhythm. This order saves time and reduces the risk of accidental failures.

• Prevent sleep while running — prevents the computer from going to sleep while Codex executes a task.
• Detail level — determines how much service information the agent displays during its work.
• Personalization — allows you to choose a more friendly or more direct communication tone.
• Custom instructions — adds permanent instructions, just like in ChatGPT.
• Avatar — displays a floating avatar that helps track background tasks.

The point of this set is that it covers three main questions at once: how Codex behaves during long work, how many technical details the user sees on screen, and how well the interface fits into the usual work process. For a new user, this is more important than rare fine-tuning options that will be needed only after a few real runs. This is what OpenAI suggests focusing on first, not cosmetics.

Control Over Process

OpenAI places particular emphasis on configuring Detail level. If you need maximum control, you can choose Coding mode: then Codex shows the specific commands it executes during work. This is useful for developers and those who want to see the logic of actions almost step by step. If the task is routine and extra technical noise gets in the way, the company recommends switching to Default to keep the dialog cleaner and calmer.

Essentially, OpenAI here divides two usage scenarios. The first is observation mode, when the user wants to check each action and understand exactly what the agent launches. The second is delegation mode, when the result matters, not the full stream of service telemetry. For a product of the Codex class, such a split is critical: some need almost terminal-level transparency, others want the feeling of a simple assistant that doesn't distract with unnecessary details.

Personalization, Background, and Rights

The second important block is Personalization. It works by the same logic as personalization in ChatGPT: you can choose the tone of communication and add your own instructions. This seems minor at first glance. In practice, it's exactly these settings that determine whether Codex will respond briefly and to the point or, conversely, provide more explanations and act as a more conversational assistant throughout the session.

OpenAI also reminds about permissions separately. The company directly recommends starting with a conservative approach: don't grant the agent extra rights in advance, but expand access as it becomes clear what tasks you really trust it with. In the same block, there's also a more everyday but useful feature — Prevent sleep while running. If the computer falls asleep during a long operation, Codex might simply stop. And through the Appearance > Avatar section, you can enable a floating avatar, drag it around the screen, and return to the task while the agent works in the background.

What It Means

OpenAI's new material shows that Codex is being promoted not only as a powerful tool but also as an everyday work interface. The main idea is simple: for an agent to work properly, it needs not only models but also understandable user settings — from the level of detail to carefully granted permissions.