OpenAI Prepares Real-Time Voice for Codex with Background Agent Support, July 24 Launch Speculated

According to Beating, OpenAI is preparing real-time voice mode for Codex that allows users to maintain continuous dialogue while background agents execute time-consuming tasks. Codex source code shows the feature has been in development for several months, with voice sessions able to bind to existing task threads via WebRTC or WebSocket connections. Users can adjust requirements or change direction while background tasks run.

The latest exposed Codex client resources indicate the real-time voice interface will display command execution, file modifications, MCP calls, and permission approvals. Supported task types mentioned include browsing, apps, documents, shopping, and food ordering. OpenAI product lead Tibo hinted today that "tomorrow has a Codex feel," leading community speculation that the feature could launch as early as July 24.