Local Voice Agents
An experiment with voice agents that run entirely in the browser, on the user's device, based on the latest browser capabilities, including WebSpeech, and LanguageModel (aka prompt) APIs.
This is simpler and cheaper than traditional cloud solutions but also less powerful, so eventually a hybrid solution with some cloud components and some local components could be used in different scenarios.
Local First Development
Experiment with and develop voice agents entirely on your local machine.
Privacy Focused
Since everything runs locally, your data and interactions remain private.
Increased Reliability
Always available, even when the cloud is down.
Cost Effective
Completely free to use, unlike expensive cloud-based solutions.
Featured
Turn detection, interruptions, multiple voices, tools support, and more.
Note: You must use Chrome IN A MODERN COMPUTER and enable "Prompt API for Gemini Nano" in chrome://flags for this demo to work.
Voice Agent Demo
npm install @livetok/local-agents
const agent = new LocalRealTimeAgent({instructions: 'You are a helpful assistant that can answer questions and help with tasks.',
voiceName: 'Albert',
turnSilenceThreshold: 600,
});
agent.start();