OpenAI updated its ChatGPT desktop application on Thursday to include support for ChatGPT Voice, enabling users to speak directly to the software to control AI agents and execute tasks on their machine. This feature relies on the ChatGPT-Live family of voice models launched earlier this month and functions with ChatGPT Work and Codex. The system can also access computer use skills to locate websites and applications. On macOS, integration with Appshots allows the application to view screen content, including alternative text descriptions. Unlike the smartphone version which prioritised smoother conversation flow without action capabilities, this desktop update permits users to dictate complex multi-step commands and respond when the AI requires input. A demonstration video showed a developer instructing ChatGPT to create a new thread, submit a pull request, and identify the root cause of a bug in a single command. Anthropic has similarly updated its voice mode for Claude to interact with apps such as Gmail, Calendar, Slack, Notion, and Canva using Opus, Sonnet, and Haiku models.
The shift from passive conversation to active task execution represents a practical change in how desktop users interact with generative models. By allowing direct control over local workflows, the feature reduces the friction between planning a task and performing it. This capability moves the interface closer to a traditional command-line utility while retaining natural language processing.
- Supports complex multi-step command dictation
- Integrated with Appshots for macOS screen awareness
- Compatible with Codex via remote access from iOS
