by VibecodedThis

Codex CLI 0.130 Adds Remote Control for Headless Sessions, Points Toward ChatGPT Integration

OpenAI released Codex CLI 0.130.0 on May 8 with a new codex remote-control command that starts a headless, remotely controllable app-server — and a GitHub discussion thread confirms ChatGPT mobile is the intended controller.

Share

OpenAI released Codex CLI 0.130.0 on May 8 with a new top-level command: codex remote-control. It starts a headless, remotely controllable app-server — a stripped-down alternative to codex app-server designed for pipelines where Codex runs without any user interface. There’s no interactive TUI, just a running process that accepts instructions from an external controller.

The OpenAI changelog describes it as “a simpler entrypoint for starting a headless, remotely controllable app-server.” A GitHub discussion thread on the openai/codex repository confirms where this is heading: the intended controller is ChatGPT itself, specifically the mobile and web apps.

The Remote Control Architecture

The app-server mode in Codex CLI has been around since earlier in the year. It exposes an API surface that other tools can drive, but it’s been positioned mainly for developers building integrations. codex remote-control is a simplified entry point for the same concept: you start it, and something else drives it.

In the context of the ChatGPT discussion thread, the design makes sense. A user could start a Codex remote-control session on their local machine or a remote server, then use the ChatGPT mobile app to give the agent tasks. The local Codex instance handles execution, including shell access, file system operations, and the sandboxing Codex already provides on Linux and Windows. ChatGPT handles the conversation interface.

This pattern is already in use at the API level for developers building agent pipelines. The 0.130 release extends it toward consumer-facing scenarios.

Other Changes in 0.130

The release also adds:

Thread pagination: App-server clients can page through large threads with configurable view options: unloaded, summary, or full turn items. For sessions with long histories, this keeps memory use manageable.

Bedrock auth via AWS console-login: Codex now supports AWS console-login credentials from aws login profiles for Bedrock authentication. If you’re running Codex against Bedrock-hosted models in an AWS environment, you no longer need to configure separate credentials manually.

Multi-environment image resolution: The view_image function now resolves files through the selected environment for multi-environment sessions. Previously it could resolve against the wrong environment when multiple were configured.

Plugin improvements: Plugin details now show bundled hooks, and sharing a plugin exposes link metadata along with discoverability controls.

Live config refresh: Config changes now take effect in running app-server threads without requiring a restart.

Sandbox hardening: Bug fixes for Windows sandbox access permissions and desktop runtime binary cache access.

Context

Codex CLI has been shipping changes rapidly since March. Version 0.128.0 added the /goal command for persistent agent loops. Version 0.129.0, released a day before 0.130, focused on TUI improvements and plugin management.

The remote-control command is a different category of addition. It’s not a feature for users of the Codex CLI directly — it’s infrastructure for the systems that will use Codex as a backend. The ChatGPT integration direction, if it ships, would put a local AI coding agent into the hands of ChatGPT subscribers without requiring them to understand or interact with the CLI at all.

Alpha versions for 0.131 are already appearing in the releases list as of this writing, so the pace of development isn’t slowing down.

Sources:

Share