OpenAI Codex release notes: Voice steers threads and projects gain multiple folders

Two changes in OpenAI Codex release notes 26.715 show up immediately in daily work. ChatGPT Voice can coordinate Codex tasks, and one local project can span several related folders. That is useful when a change touches a web app, API, and infrastructure at the same time.

OpenAI Codex release notes 26.715: what changed?

A coding agent is an AI agent that can read a codebase, propose changes, and use tools such as Git or a terminal within its granted permissions.

Voice can start, check, and steer other threads

ChatGPT Voice in the desktop app is powered by GPT-Live. You can start a new chat or task in voice mode, then ask ChatGPT to start, check, or steer work in other threads across Chat, Work, and Codex. On macOS, Screen context can share an appshot, a snapshot of the frontmost app window, as extra context.

Voice is available on Plus, Pro, Business, Edu, and Enterprise plans in the desktop app and through Remote on iOS. OpenAI does not describe this as a new voice feature in Codex CLI, so use the desktop app for this workflow.

Source: OpenAI Codex changelog, ChatGPT Voice and multi-folder projects 26.715.

Local projects can contain multiple folders

A local project in the ChatGPT desktop app can now include several related folders. You choose one as the primary folder. It controls new chats, Git operations, and automatic discovery of AGENTS.md, skills, and config.toml. Secondary folders remain available for file search, reading, and editing.

This helps when one customer workflow spans several codebases. Codex can read the interface in one folder, the API contract in another, and deployment configuration in a third without forcing you to merge everything into an artificial project first.

Source: OpenAI Codex changelog, multi-folder local project support.

What this changes for Nordic teams

The practical gain is better context for work that crosses repository or folder boundaries. A developer can ask Codex to map dependencies between a frontend and an API. An agency can keep a client's theme, integration, and deployment code in the same local project. Voice adds a quick way to check ongoing work or steer the next step without leaving the current discussion.

Choose folders deliberately. Give the project access to the codebases it needs, keep tokens in environment variables or a secret manager, and add an approval gate before writes that affect several folders. Review the combined diff and run each codebase's tests before moving the change forward.

Human step

1. Open the project's menu in the ChatGPT desktop app and select Edit project.
2. Add the related folders and choose the primary folder.
3. To coordinate work by voice, start a new chat or task in voice mode. On macOS, turn on Screen context when the frontmost window is relevant.

Those are app actions. The prompt below starts after the folders are already part of the project.

Short example: use the new Codex feature

Replace the bracketed text and send this prompt to Codex:

Map how [the planned change] affects the primary and secondary folders. List affected files by folder, dependencies between folders, the applicable instructions from the primary folder's AGENTS.md, and the tests we need. Change nothing.

Good output should:

• cite real file paths in the affected folders,
• separate dependencies, proposed changes, and tests,
• state which primary-folder instructions govern the work,
• stop before editing files, so a person can approve the plan.

What to test now

Start with a change that genuinely crosses two folders, such as a new API property that also needs to appear in the web interface. Check whether Codex finds both sides of the contract and splits the test plan by codebase. Then try Voice to ask for the thread's status or steer the next bounded step.

The npm registry still lists 0.145.0 as stable latest, while 0.146.0-alpha.10.1 is alpha. Release note 26.715 covers the ChatGPT desktop experience, so do not treat it as a stable Codex CLI 0.146 launch.

Source: npm registry for @openai/codex.

Teams that want to turn this into a repeatable, reviewable operating pattern can handle it through Tool Forge: folders, permissions, tests, and human decisions need to work as one flow.