Codex: Codex Computer Use
Last updated: 2026-08-31
Codex's Computer Use feature is a key differentiator from traditional coding tools — it doesn't just write code, it can operate your computer.
📋 Prerequisites: Understanding Codex basic operations and permission settings
1. What You Will Learn
- Computer Use concepts and capabilities
- Browser control
- Desktop application control
- Security considerations
2. What is Computer Use
Computer Use lets Codex operate a computer like a human:
| Capability | Description |
|---|---|
| Browser Control | Open web pages, click, type, take screenshots |
| Desktop App Control | Operate local applications |
| File Management | Browse, create, modify file systems |
| Command Execution | Run commands in the terminal |
graph LR
A[Codex Agent] --> B[Browser]
A --> C[Terminal]
A --> D[File System]
A --> E[Desktop Apps]
B --> B1[Navigation]
B --> B2[Form Filling]
B --> B3[Screenshot Analysis]
C --> C1[Run Commands]
C --> C2[View Output]
D --> D1[Read/Write Files]
D --> D2[Directory Operations]
3. Browser Control
(1) Basic Operations
TEXT
📖 Display only
> Open https://example.com and take a screenshot
> Type "Codex tutorial" in the search box and search
> Fill in the registration form with alice@example.com as email
> Click the "Submit" button
(2) Web Testing
TEXT
📖 Display only
> Open localhost:3000, verify login functionality works
> Screenshot the homepage, check responsive layout
> Test shopping cart flow: add item → checkout → confirm
▶ Example 1: Alice Tests a Web Page
TEXT
📖 Display only
# Alice has Codex do end-to-end testing
> Open http://localhost:8080
> Register a test account test@example.com / password123
> After login, create a new project
> Screenshot to confirm the project list displays correctly
> Delete the test account
# Codex output:
# → Opening browser to http://localhost:8080
# → Filling registration form
# → Clicking "Register" button
# → Logging in with test credentials
# → Creating project "Test Project"
# → Taking screenshot of project list
# → Account deleted successfully
# → Test completed ✓
4. Desktop Application Control
(1) Operation Examples
TEXT
📖 Display only
> Open VS Code and open the current project
> Run npm start in the terminal
> Open the browser to access localhost:3000
> Screenshot to confirm the page displays correctly
(2) Multi-step Operations
TEXT
📖 Display only
> Execute the following:
> 1. Start the database in the terminal
> 2. Run database migrations
> 3. Start the development server
> 4. Open the browser to verify
> 5. Screenshot to confirm all functions work
▶ Example 2: Bob's Deployment Verification
TEXT
📖 Display only
> Deploy to staging environment:
> 1. git push origin staging
> 2. Wait for CI to complete
> 3. Open staging.example.com
> 4. Test key functions: login, search, checkout
> 5. Screenshot each step
> 6. Generate test report
5. Security & Privacy
(1) Computer Use Permission Requirements
Computer Use requires Full-access sandbox mode, meaning:
| Consideration | Description |
|---|---|
| High Privileges | Codex can operate any application |
| Visual Access | Codex can see screen content |
| Input Simulation | Codex can simulate keyboard and mouse operations |
| Network Access | Codex can access any website |
(2) Security Recommendations
TEXT
📖 Display only
⚠️ Computer Use Security Checklist:
1. Do not use on screens containing sensitive information
2. Do not let Codex access banking/payment websites
3. Always monitor Codex's operations
4. Check for abnormal operations after use
5. Extra caution for production environment operations
(3) Operation Confirmation
In Computer Use mode, critical operations still require confirmation:
TEXT
📖 Display only
⚠️ Codex requests operation confirmation:
- Open https://production.example.com
- Operation type: Browser navigation
- Risk level: High (production environment)
[Approve] [Reject]
6. Computer Use Configuration
TOML
# ~/.codex/config.toml
[computer_use]
enabled = true
browser = "chromium" # Browser engine
headless = false # Headless mode
screenshot_interval = 5 # Screenshot interval (seconds)
max_operations = 100 # Max operations per session
require_confirmation = true # Critical operations need confirmation
7. Computer Use vs Traditional Automation
| Dimension | Computer Use | Traditional Scripts |
|---|---|---|
| Flexibility | High (AI understands page changes) | Low (depends on fixed selectors) |
| Maintenance Cost | Low (auto-adapts to UI changes) | High (scripts need updating when UI changes) |
| Learning Curve | Low (natural language descriptions) | High (need to learn Playwright/Selenium) |
| Reliability | Medium (AI may misjudge) | High (exact matching) |
| Best For | Exploratory testing, quick verification | CI/CD automation, regression testing |
❓ FAQ
Q What permissions does Computer Use require?
A Full-access sandbox mode is needed because Codex needs to operate browsers and desktop applications. Use only in trusted environments.
Q Can Codex operate any browser?
A Codex uses the Chromium engine to control browsers. Support for other browsers is limited.
Q Does Computer Use record video?
A Codex takes periodic screenshots for analysis and decision-making, but does not continuously record video. Screenshots are saved in a temporary directory.
Q Can Computer Use work in headless mode?
A Yes. Set
headless = true, and Codex operates in a headless browser, suitable for server environments.Q Can Computer Use replace Playwright?
A Not entirely. Computer Use is more flexible but less reliable, great for exploratory testing; Playwright is more precise and reliable, better for CI/CD automation. They complement each other.
📖 Summary
- Computer Use lets Codex operate browsers and desktop applications
- Browser control: navigation, forms, screenshots, testing
- Requires Full-access mode, higher security risk
- Always monitor operations, protect sensitive information
- Complements traditional automation tools, not a full replacement
📝 Exercises
- Basic (⭐): Have Codex open a web page and take a screenshot.
- Intermediate (⭐⭐): Have Codex automatically complete a web form filling flow.
- Advanced (⭐⭐⭐): Use Computer Use to implement end-to-end testing — from registration to login to operations, generating a test report.
