OpenAI Adds Hands-Free Voice Control to Codex and ChatGPT

AI News

July 26, 2026

OpenAI Adds Hands-Free Voice Control to Codex and ChatGPT

OpenAI has added GPT-Live voice control to the ChatGPT desktop application, allowing users to direct Codex and ChatGPT Work agents through natural spoken conversations.

The new GPT-Live Codex voice control is available in the ChatGPT desktop app for macOS and Windows. Users can describe tasks, interrupt the assistant, answer questions and redirect ongoing work without relying entirely on a keyboard.

Full-duplex conversations

GPT-Live uses a full-duplex architecture, meaning it can listen and speak at the same time.

Traditional voice assistants usually wait for a clear pause before responding. GPT-Live can continue listening while it speaks, making interruptions and quick corrections feel more natural. It can also recognise pauses and decide whether to respond or remain silent.

This design is particularly useful for coding and agent workflows, where users may need to provide additional instructions while a task is already running.

Voice control for Codex

Codex is OpenAI’s agentic software-development system. It can inspect code, edit files, run tests and manage longer technical tasks.

With the new desktop voice interface, developers can ask Codex to investigate a bug, explain a code change, run tests or coordinate several coding agents using spoken instructions.

Codex tasks continue to operate through the same tools and permissions available in the desktop application. Users can review changes, inspect diffs and approve sensitive actions before they are completed.

Managing several agents

The feature also works with ChatGPT Work, OpenAI’s agent for longer, multi-step projects.

Users can speak to agents that are researching information, creating documents or working across connected applications. Voice can be used to change priorities, provide missing details or move between tasks while the agents continue operating.

This could make complex workflows easier to manage when several agents are running simultaneously. Instead of switching between multiple text threads, users can coordinate work through a live conversation.

Desktop availability

OpenAI is rolling out the feature globally to Plus, Pro, Business, Edu and Enterprise users on macOS and Windows.

GPT-Live initially launched as a more natural voice model before being connected to Work and Codex. The broader integration turns voice from a conversational feature into an interface for controlling AI agents and completing practical tasks.

A new interface for agentic coding

Voice control does not remove the need to review code or monitor autonomous agents.

Developers still need to inspect proposed changes, confirm permissions and verify that generated code behaves as expected. Background noise, overlapping speech and network quality can also affect how accurately spoken instructions are understood.

However, the update shows how agentic coding tools are moving beyond text-based prompts.

By combining Codex with full-duplex voice, OpenAI is creating a workflow in which developers can supervise software agents through ongoing conversation while the agents write, test and revise code in the background.

AI News