Open-source voice tool OpenLive now lets you talk to Claude Code and other coding agents
OpenLive is an open-source voice interface that gives AI models or agents ears, a mouth, and eyes. The entire pipeline — detecting when someone is speaking, converting speech to text, figuring out when a person has finished talking, and generating spoken responses — runs entirely on the user's own computer, so audio never leaves the machine and there are no per-minute fees.
The latest update connects OpenLive directly to including Claude Code, Codex, Cursor, OpenCode, and Hermes, all running locally under the user's own login. A person picks an agent, points it at a project folder, and gives instructions by voice; when the agent wants to run a command or edit a file, OpenLive reads that request aloud so the user can respond by voice, and it can also narrate what the agent is doing during a task.
Conversations are saved into each agent's own session history, so a task started by voice can later be resumed from that agent's , or vice versa. A voice-cloning feature is also included: recording just 5 to 30 seconds of audio lets the assistant speak back in that person's own voice, with the cloning process running entirely on-device and nothing uploaded.
Key points
- Voice detection, , end-of-speech detection, and all run locally on the user's own machine
- No audio leaves the device and there are no per-minute fees
- Connects directly to such as Claude Code, Codex, Cursor, OpenCode, and Hermes for voice-driven control
- Reads aloud when the agent wants to run a command or edit a file, so the user can respond by voice
- Cloning a voice from just 5-30 seconds of audio lets the assistant speak in the user's own voice, processed entirely on-device