# Codex Goes Beyond Code: PC Control, Memory, and 90+ Plugins
OpenAI has unveiled a major update to Codex — a tool used by over 3 million people weekly. The update transforms the assistant from a narrowly specialized coding helper into a universal digital agent capable of interacting with the operating system, remembering context, and handling long-term tasks without constant user involvement.
PC Control at the GUI Level
The key feature of the new release is background computer control (background computer use). Codex can now see the screen contents, move its own cursor, click on interface elements, and type text into any apps — including those without an API. This is especially useful for GUI tools that lack programmatic interfaces for automation.
On macOS, multiple agents can work alongside the user without stealing input focus. The feature is currently limited to this platform, and users in the EU and UK will get access later due to regulatory requirements. For developers, this opens up new scenarios: rapid frontend component iteration, UI/UX testing, and automating routine tasks in IDEs or configurators.
Browser Integration and Image Generation
The desktop app now includes a built-in browser that lets you comment on page elements and give the agent precise instructions. At this stage, the functionality focuses on frontend development and web games, but OpenAI plans to expand it to full browser control.
Codex also gained access to the gpt-image-1.5 model, allowing it to generate layouts, icons, concept art, and other visuals right in the middle of a task. For example, when building a landing page, the agent can suggest design options, implement them in HTML/CSS, and test them in the browser immediately.
Expanded Plugin Ecosystem
The integrations catalog has added over 90 new plugins. Highlights include:
- Atlassian Rovo (for working with JIRA)
- CircleCI and GitLab Issues (CI/CD and bug tracking)
- Microsoft Suite (Office tools)
- Neon from Databricks (cloud databases)
- CodeRabbit and Superpowers (code review and advanced dev tools)
- Remotion and Render (video and hosting)
These integrations let Codex handle the full development lifecycle: from task assignment in trackers to deployment and monitoring.
Long-Term Tasks and Persistent Memory
The agent can now retain context across sessions. The new memory system remembers user preferences, style tweaks, commonly used templates, and interaction history. Previously, this info had to be duplicated via system prompts; now it's stored automatically.
On top of that, Codex can set its own "alarms" — postponing tasks and returning to them after hours, days, or weeks. Example scenarios:
- Automatically completing unattended pull requests
- Monitoring notifications in Slack, Gmail, and Notion, then taking action
- Building a unified context from scattered discussions across tools
Proactive Recommendations and Multitasking
The interface now offers a personalized action list at the start of the day. The agent analyzes open docs in Google Docs, unread Slack messages, Notion tasks, and code repo status to create a prioritized work plan.
The desktop client also adds:
- Parallel terminal tabs
- SSH access to remote devboxes (in alpha)
- Sidebar preview for PDFs, spreadsheets, presentations, and text docs
- GitHub review integration right in the interface
Key takeaways:
- Codex is no longer limited to code generation — it's a full OS agent.
- Background GUI control works even in API-less apps.
- Memory and long-term tasks enable complex workflow automation.
- Over 90 new plugins cover the entire dev cycle.
- Personalization features are temporarily unavailable in the EU and UK.
The update is now available to ChatGPT desktop app users logged in with personal accounts. Enterprise and Edu versions will get memory and personalization features later.
— Editorial Team
No comments yet.