Sun

Sun is a realtime multi-speaker voice model designed for AI agents that need to hold natural, back-and-forth conversations with humans. Unlike typical text-to-speech wrappers or simple voice chatbots, Sun is built as a Collaborative Voice Model (CVM) that understands overlapping speech, interruptions, and follow-ups without requiring a wake word. It connects via WebSocket API, allowing developers to integrate it into apps, meetings, or call systems where the AI can listen, respond, and even relay information from other tools in real time.

Sun stands out by solving common voice AI frustrations. It responds within a second, so there's no awkward pause. It detects when someone else is speaking and holds its response, preventing the AI from talking over people. If a user interrupts, Sun stops and addresses the interruption directly. It also stays ready for follow-up questions after speaking — no need to repeat a wake word. The model supports a 350K token context window, enough for hours of conversation, and includes features like speech injection (push announcements mid-call) and live info access for real-time lookups. Sun integrates with web search, dynamic context memory, and can relay structured outputs from other agents as natural speech.

This tool is ideal for developers building voice-enabled AI agents for meetings, customer service, or collaborative workspaces. Product teams can use it to add real-time voice to their apps without building complex speech pipelines. Enterprises benefit from its low latency, cost efficiency (50% cheaper than competitors for audio tokens), and 99.9% uptime guarantee. Whether you're creating a meeting assistant that checks metrics mid-conversation or a customer support bot that handles interruptions gracefully, Sun provides the infrastructure for natural, human-like voice interactions.