Woofun AI reports that OpenAI is integrating a real-time voice mode into Codex, enabling continuous user dialogue while background agents handle time-consuming tasks. Publicly accessible source code confirms this feature has been under development for several months, utilizing WebRTC or WebSocket connections to link voice conversations with existing task threads. Users can modify requests or adjust directions during background execution.

Exposed client resource strings indicate the interface will display states such as command execution, file modification, MCP calls, and permission approvals. Cue words reference tasks including browsing, document handling, shopping, and food ordering. Although OpenAI has not officially confirmed these features, product lead Tibo’s recent comment has led to speculation of a potential release on July 24.