Computer Use is delivered as an official OpenAI plugin, not a built-in Codex tool. See the OpenAI Codex Computer Use docs for the full reference and supported tasks.
What It Does
When Computer Use is enabled, Codex can:- Open and navigate web apps in a managed browser
- Click, type, scroll, and drag inside the virtual machine
- Take screenshots and reason over the visible UI
- Run command-line tools inside the sandbox
- Complete multi-step tasks that mix browsing, file manipulation, and shell commands
Installing the Plugin
1
Open the Plugins page
Click Plugins in the sidebar. The page opens on the Plugins tab (Codex marketplace).
2
Find Computer Use
Search for
computer-use in the marketplace, or browse the curated list and pick it from the OpenAI plugins.3
Add the plugin
Click Add on the plugin detail page. Mains forwards the install request to your Codex account, which provisions the plugin and reports back when it’s ready.
4
Open a Codex workspace
Start a new Codex run — the agent now has the Computer Use tool available and will use it when the task calls for a UI.
Using It in a Run
Computer Use is invoked like any other Codex tool — the agent decides when it’s needed. You can also nudge it explicitly:click, type, screenshot) and a thumbnail of the virtual screen at each step.
Permissions & Safety
When Not to Use It
Computer Use is the right tool when there is no API or CLI for what you need. If a job can be done with HTTP, gh, kubectl, or another command-line tool, prefer that path — it’s faster, cheaper, and more reliable. Reach for Computer Use only when the only way to complete the task is through a UI.Related
- Plugins marketplace — browse, install, and manage all Codex plugins (Computer Use is one of many)
- Codex Configuration — sandbox modes, approval policy, model selection
- Subagents — delegate the UI work to a focused sub-agent while the parent keeps coding
