OpenAI expands Codex with computer use, memory, and plugins

OpenAI has updated Codex so it can use a computer’s apps, handle more developer workflows, and remember context across tasks. The company said the new features are rolling out now to desktop app users signed in with ChatGPT, with some personalization features and macOS computer use expanding later.

OpenAI expands Codex with computer use, memory, and plugins

OpenAI has released a major update to Codex, adding computer use, memory, more workflow tools, and broader app integrations for the developer assistant used by more than 3 million people each week, according to the company. The update is meant to move Codex beyond code generation and into more of the software development process.

The biggest change is background computer use. OpenAI said Codex can now see, click, and type with its own cursor across apps on a user’s computer. The company said multiple agents can work in parallel on a Mac without interfering with other work in other apps.

⚡ New to this?

This matters because Codex is moving from a code-writing helper to a broader work assistant. Computer use means the system can interact with apps the way a person does, while memory means it can keep useful context from earlier sessions.

For non-experts, an API is a formal way for software to talk to other software. When a tool can work without an API, it can still operate on apps by using the screen and mouse/keyboard actions instead.

🦞 OpenClaw angle

If you build self-hosted AI workflows, treat this as a sign to separate tasks that need browser or GUI control from tasks that can stay API-based. Add logging and approval steps around any agent that clicks through apps, because those actions are harder to audit than direct API calls.

If you are designing long-running automations, store task context outside the model as well, then compare it against the agent’s memory before resuming work. For teams using plugins or MCP servers, keep a tight allowlist and test each connector in a sandbox before letting it touch Jira, Slack, email, or Git repositories.

OpenAI said this is useful for iterating on frontend changes, testing apps, and working in software that does not expose an API. The feature is starting on macOS, with rollout to EU and UK users planned later.

Codex is also getting a stronger web interface. The app now includes an in-app browser where users can comment directly on pages and give precise instructions to the agent. OpenAI said that is useful for frontend and game development, and that the company plans to expand browser control further over time.

Image generation is now part of the workflow as well. Codex can use gpt-image-1.5 to generate and refine images, which OpenAI said can help with product concepts, frontend designs, mockups, and games when combined with screenshots and code.

To extend what Codex can do across other services, OpenAI said it is adding more than 90 plugins. These combine skills, app integrations, and MCP servers, a reference to the Model Context Protocol, which lets AI tools connect to external systems. Among the plugins OpenAI highlighted are Atlassian Rovo for Jira, CircleCI, CodeRabbit, GitLab Issues, Microsoft Suite, Neon by Databricks, Remotion, Render, and Superpowers.

The Codex app is also getting deeper support for day-to-day developer workflows. OpenAI said users can now address GitHub review comments, run multiple terminal tabs, connect to remote devboxes over SSH in alpha, and open files in a sidebar with rich previews for PDFs, spreadsheets, slides, and docs.

A new summary pane shows agent plans, sources, and artifacts. OpenAI said that should make it faster to move between writing code, checking outputs, reviewing changes, and collaborating with the agent in one workspace.

OpenAI is also expanding automations so Codex can reuse existing conversation threads and keep context that has already been built up. The company said Codex can now schedule future work for itself and wake up automatically to continue long-running tasks over days or weeks.

According to OpenAI, teams already use automations for landing open pull requests and following up on tasks across tools such as Slack, Gmail, and Notion. The company is also previewing memory, which lets Codex remember preferences, corrections, and other useful context from past work.

OpenAI said memory and context-aware suggestions can help Codex propose useful work to continue, such as identifying open comments in Google Docs, pulling relevant context from Slack, Notion, and a codebase, and returning a prioritized list of actions.

The update is rolling out starting today to Codex desktop app users signed in with ChatGPT. OpenAI said personalization features, including context-aware suggestions and memory, will reach Enterprise, Edu, and EU and UK users soon.

Source: OpenAI News ↗

More from AI News