OpenAI Codex Ships Broad Release with First-Class Plugins and Multi-Agent Overhaul
OpenAI's Codex coding agent received its biggest update yet on March 27, 2026, bringing first-class plugin support, a redesigned multi-agent architecture, and a fully capable app-server TUI. The release signals OpenAI's push to make Codex the default interface for professional software development.
OpenAI shipped version 0.117.0 of Codex today — a broad release that touches nearly every layer of the tool. Plugins are now first-class, multi-agent sessions got a structural overhaul, and the app-server TUI is enabled by default for all users.
Plugins stop being an afterthought. Previously, plugin integration in Codex required manual wiring and wasn’t well-documented. Starting with this release, Codex syncs product-scoped plugins at startup, exposes them in a browsable /plugins interface, and handles authentication and setup inline. Install or remove plugins without leaving the session. This is the change that unblocks the integrations ecosystem that’s been building around Codex.
Multi-agent gets readable. Sub-agents now use path-based addresses like /root/agent_a instead of opaque identifiers. That sounds minor, but it changes how you debug and monitor parallel sessions — you can see at a glance which agent is doing what. Inter-agent messaging is now structured, and agents can be listed programmatically, which matters for automated orchestration pipelines.
The /title terminal-title picker now works in both the classic TUI and the app-server TUI, solving a real annoyance for anyone running parallel Codex sessions in split panes. Prompt history recall also works across app-server TUI sessions for the first time.
Image workflows are cleaner. In code mode, Codex now returns URL references for generated images instead of inline blobs. Generated images can be reopened, and they persist across session resumptions — so picking up a multi-hour task no longer means losing visual context from earlier in the session.
Under the hood, this release fixes stability issues across sandboxing, multi-agent sessions, and plugin integration — the three areas that had the most friction in production use. The Codex app and VS Code extension now share key settings, closing a gap where preferences set in one place didn’t carry over to the other.
Context: OpenAI released the Codex desktop app for macOS in February, followed by a Windows version on March 4. The app hit over 1 million downloads in its first week. The underlying model, GPT-5.3-Codex, scores 57% on SWE-Bench Pro — above every competing coding agent at time of release. GPT-5.4, released March 17, added native computer-use mode, and its Codex integration allows agents to operate desktop applications as part of multi-step workflows.
What’s actually happening here is OpenAI is making Codex the default professional interface for software development. The plugin ecosystem, multi-agent orchestration, and TUI improvements all point to the same direction: Codex as a platform, not just a chatbot you ask for code snippets. The developers who build on top of this release will be shipping production integrations before Anthropic or Google catch up with comparable tooling breadth.
Sources: OpenAI Codex Changelog, OpenAI Developers