Vui

Vui is an open source, real time voice assistant from Fluxions that runs on your own machine. Out of the box it provides a streaming voice loop: speech recognition, a language model and text to speech, delivered through a WebRTC and WebSocket server with a browser interface. Its central idea is a tool calling layer that hands spoken instructions to whatever agent stack the developer connects, so a voice request can drive Claude, MCP servers, a shell or custom code. The release includes the vui Nano text to speech model, a 300 million parameter decoder trained on two way conversation rather than scripted reads, with inline tags for breaths, sighs, hesitations and similar non verbal sounds. It also ships an endpoint compatible with OpenAI's Realtime API, a one shot voice note endpoint that returns rendered audio, an optional Claude sidecar for tasks such as email, calendar and web research, and a standalone text to speech demo. It is meant to run via Docker compose on a single GPU. It suits developers who want to build voice interfaces, voice note bots or hands free control of their tools without relying on a hosted assistant. The model and server are released as open source. Fluxions also sells a hosted voice platform, priced at $0.10 per minute all in for agent conversations, billed by the second, with transcription at $0.20 per hour of audio and speech synthesis at $9 per million characters. Telephony is bring your own or at cost.