openlive
Opensource, on-device voice + vision layer for AI agents. Bring any model or coding agent; the whole speech loop (VAD, STT, TTS, barge-in) runs locally. An open alternative to ElevenLabs Agents, Gemini Live, and OpenAI Realtime.
OpenLive gives any AI agent local voice and vision, so you can talk to it and let it see your screen without cloud audio fees.
OpenLive is an open source tool that gives an AI agent the ability to hear, speak, and see, all running on your own computer instead of a cloud service. It aims to be a free alternative to paid platforms like ElevenLabs Agents, Gemini Live, and OpenAI Realtime, without charging by the minute for audio.
The tool handles the full voice conversation loop: detecting when you start and stop speaking, converting your speech to text, sending that text to an AI model, converting the model's reply back into speech, and letting you interrupt mid sentence if you want to jump in. All of this listening and speaking work happens locally on your device using WebGPU, so your voice never leaves your machine. Only the final text of what you said, plus camera or screen images if you turn that feature on, gets sent to whichever AI model you choose.
You can connect almost any AI model you already have an account with, including Anthropic, OpenAI, Google, xAI, DeepSeek, Groq, or fully local models through Ollama. You are only charged whatever that model provider normally charges, with no extra fee for the voice pipeline itself. OpenLive can also connect directly to coding agents such as Claude Code, Codex, Cursor, OpenCode, or Hermes, letting you talk to your coding assistant out loud, hear its plan as it works, and answer its permission requests by voice instead of typing.
Other features include cloning your own voice from a short recording so replies sound like you, a camera and screen sharing mode so the AI can see what you are looking at, and a floating mini window that keeps listening while you continue working elsewhere. Conversations can be exported as text transcripts, and API keys are stored encrypted on your device.
The project describes its approach honestly: it is a chain of separate speech to text, AI, and text to speech steps rather than a single model that handles voice directly, which is the tradeoff it makes in exchange for working with any AI model of your choice, entirely on your own machine. It is aimed at developers and vibe coders who want a private, no subscription way to talk to their AI tools.
Where it fits
- Talk out loud to a coding agent like Claude Code or Cursor and hear its plan as it works.
- Add a private, local voice interface to any AI model you already have API access to.
- Clone your own voice so an AI assistant replies to you in a voice that sounds like yours.
- Show an AI agent your camera or screen so it can respond to what it sees.