OpenAI Codex adds computer use, memory, images, and 90+ plugins for desktop agent workflows
OpenAI’s Codex desktop update expands the coding agent into broader software-workflow automation with computer use, an in-app browser, image generation, memory, scheduled automations, and more than 90 plugins.
What changed in OpenAI Codex
OpenAI’s latest Codex desktop update turns the product from a coding-only assistant into a broader desktop agent for software work. The release adds background computer use, an in-app browser, image generation through gpt-image-1.5, memory, scheduled automations, and more than 90 plugins. Independent coverage from Web Developer corroborates the same core feature set and frames the change as part of the wider shift from IDE helpers to ambient software agents.
Key takeaways
- Codex can now use macOS apps in the background by seeing, clicking, and typing with its own cursor while the user keeps working elsewhere.
- The desktop app adds an in-app browser for commenting directly on pages, with OpenAI positioning it first around frontend, app, and game iteration.
- OpenAI says Codex can generate and revise images through gpt-image-1.5 inside the same workflow as screenshots and code.
- More than 90 new plugins extend Codex into tools such as Jira, CircleCI, CodeRabbit, GitLab Issues, Microsoft Suite, Neon by Databricks, Remotion, Render, and Superpowers.
- Automations can reuse existing conversation threads, while memory and proactive suggestions are designed to carry project context across longer-running work.
Practical LinkLoot angle
The useful decision is whether Codex should stay inside your coding lane or become a workstation-level helper. For teams already using ChatGPT and Codex, this update makes Codex more interesting for multi-step workflows such as reviewing a pull request, testing a frontend screen, revising a mockup, checking a document, and scheduling follow-up work from the same project context. The limitation is equally important: computer use is initially macOS-focused, personalization is not rolling out everywhere at once, and any agent that can click/type across apps needs tighter permissions, review checkpoints, and artifact logging than a pure code-completion tool.
| Capability | Best use | Main limitation | Source |
|---|---|---|---|
| Background computer use | Testing local apps or tools without APIs | Initially macOS-focused and permission-sensitive | OpenAI |
| In-app browser | Frontend, app, and game iteration with page comments | Full browser command is described as a future direction | OpenAI |
| gpt-image-1.5 in Codex | UI mockups, product visuals, game assets | Quality and brand-review still need human checks | OpenAI |
| 90+ plugins | Pulling context from team tools and CI/dev platforms | Plugin permissions and data exposure must be reviewed | OpenAI / Web Developer |
| Memory and automations | Recurring project follow-ups and long-running tasks | Rollout varies by region and account type | OpenAI |
For a broader tool stack view, pair this with LinkLoot’s guide to AI agent tools and AI workflow automation.
What to verify before you act
Check which Codex desktop features are actually available for your account, region, and operating system before redesigning workflows around them. Review plugin scopes before connecting Jira, GitLab, Microsoft, CI, or database-related tools, because the productivity gain depends on how much context Codex can safely access. For production code or client work, require a human review step for generated images, browser actions, PR changes, and scheduled automations so the agent does not silently carry an old assumption into a later task.
Source check
OpenAI confirms the feature launch, named capabilities, plugin examples, availability notes, and the claim that more than 3 million developers use Codex weekly. Web Developer independently confirms the same desktop-agent direction, including background computer use, the in-app browser, image generation, plugins, developer workflow additions, automations, memory, and rollout boundaries.
Codex adds background computer use, an in-app browser, gpt-image-1.5 image generation, memory, scheduled automations, and more than 90 plugins.