FULL STORY

OpenAI Brings Voice and Multi-Agent to Desktop ChatGPT

Following rumors of testing voice and multi-agent modes in Codex, OpenAI officially rolled out ChatGPT Voice to its desktop apps, evolving it into a comprehensive voice assistant.

2026-07-23 ~ 2026-07-24 · 2 episodes · 34 posts

Episode 1 · OpenAI Codex Rumored to Test Realtime Voice and Multi-Agent Assistant Mode (2026-07-23, 8 posts)

Multiple sources reveal that OpenAI's Codex is testing a Realtime Voice Mode, aiming to transform it into a personal assistant that can be directly conversed with and directed via mobile phones. Leaked system prompts have unveiled its dual-layer architecture design and the potential for integrating external services. Currently, the feature remains in the testing and tracking phase and has not been officially rolled out.

Confirmed

According to widely circulated leaked system prompts, Codex's voice assistant mode will employ a dual-layer architecture: one acting as a "general-purpose agent assistant" for continuous conversation, and another serving as a background worker agent responsible for executing specific tasks. Regarding app integrations, this voice mode is expected to use voice commands to drive the Codex Agent to handle tasks across Slack, Spotify, documents, and calendars. Furthermore, developer @joshwhiton confirmed that prior to the official launch of the voice mode, offline CLI tools are already capable of supporting all Agents and achieving similar multi-agent collaborative workflows.

Unconfirmed

The Realtime Voice Mode and its related personal assistant features are currently only in the code tracking and internal testing phases. A specific release timeline and the final feature set have yet to be officially confirmed by OpenAI. Additionally, rumors suggest a recent Cerebras-based Sol update might be coming, but its connection to Codex and specific details remain to be seen.

Episode 2 · OpenAI Brings Voice Control and Multi-Agent Orchestration to Desktop (2026-07-24, 26 posts)

OpenAI has officially brought ChatGPT Voice to its macOS and Windows desktop apps, rolling it out to Plus, Pro, Business, Edu, and Enterprise users. The core feature allows users to control their computers via voice and coordinate multi-agent collaboration across Chat, Work, and Codex. This marks a shift for the desktop app from a simple chat window to a system-level control center.

Confirmed

ChatGPT Voice is now on desktop and available to various paid and enterprise users. Users can direct AI agents via voice to execute tasks, such as coding in Codex. The macOS system supports screen awareness. Additionally, the desktop app now supports local projects by mounting multiple folders. Developer @Dimillian noted that Codex Micro is fully compatible with voice, allowing task management without opening individual conversation threads; @kevinkern pointed out that users can ask real-time questions or change task directions via voice while Codex is working.

Why it matters

This update combines voice interaction with code execution, drastically improving workflow fluidity. @Dimillian called the experience of launching and managing Codex tasks directly via voice incredibly smooth, even dubbing it a true "AGI moment." @nickbaumann_ also believes that combining lifelike voice conversations with Codex's execution capabilities transforms traditional input or click operations into direct voice commands, significantly lowering the barrier to human-computer interaction.

6 more related posts →