Sun

The second feature group addresses turn-taking and continuous interaction. Barge-in prevention detects active speakers and holds responses, allowing conversations to flow with proper turn-taking. This prevents the AI from talking over participants, a common frustration in group meetings. The follow-up ready feature eliminates the need for wake words after the AI has spoken—users can simply ask the next question naturally. This is a significant improvement over traditional voice assistants that require repeating a wake phrase for every query. By combining barge-in prevention with wake-word-free follow-ups, Sun enables long-running conversational state where the AI stays contextually aware across multiple exchanges. For example, after answering one query, it remains ready for the next without any explicit invocation, making interactions feel like a continuous dialogue with a human collaborator.

Sun offers additional capabilities that extend its utility. Speech injection allows push announcements or notifications to the AI to speak at any time during a call, enabling proactive information delivery. Live info access provides real-time lookups for prices, events, and specs, ensuring the AI always gives the latest information without leaving the conversation. Configurable context lets developers adapt the bot's behavior for specific meeting contexts, making it feel unique to each use case. These features transform the AI from a passive responder to an active participant that can inject alerts, fetch live data, and tailor its personality. For instance, a meeting agent can interrupt with a breaking news update or pull live metrics from a dashboard, all while maintaining the natural conversation flow.

Sun operates on a collaborative voice model architecture built on websockets for low-latency communication. The signal flow involves an input stream (audio or text) being processed through Sun's model, which then outputs voice and text streams. The model is designed with a large context window of 350K tokens, enabling hours of sustained real-time collaboration without losing conversational history. This is significantly larger than competitors like ChatGPT Realtime or Gemini Live, which support only a few minutes. Sun Zero, the complete real-time collaboration suite, integrates text and voice seamlessly, providing WebSocket API access, scalable minutes, and support for multiple concurrent connections. The architecture is built for tool and agent orchestration, allowing Sun to interpret JSON or structured outputs from other agents and relay them as natural speech. This makes it a bridge between humans and systems, handling long-running state and dynamic context updates.

Concrete use cases include real-time meeting assistants that can check metrics like MRR or churn without breaking flow. In a sample session, Maya asks 'Sun, can you check what our current MRR is?' and the agent responds with live data, even relaying updates from analytics agents. Jake then asks a follow-up about churn without any wake word, and the agent answers from the metrics dashboard. This enables teams to query databases, get reports, and receive proactive notifications mid-conversation. Other scenarios include customer support where the agent can handle interruptions like 'wait, what about the budget?' and immediately pivot, or product demos where live info access provides real-time pricing and specs. The outcome is increased meeting efficiency, reduced context switching, and a more natural collaborative experience that feels like having a knowledgeable team member in the room.

Sun is designed for developers, AI-native organizations, and product teams building voice agents for collaborative environments. The target platforms include any WebSocket-capable application, with SDK access through the Sun Zero API. Pricing ranges from a free tier (15 minutes/month, 2 connections) to Enterprise custom plans with unlimited minutes and dedicated infrastructure. The Pro plan at $29/month offers 200 minutes and priority support, making it accessible for production apps. The technology stack is model-agnostic but optimized for real-time voice. In summary, Sun is a revolutionary collaborative voice model that solves the fundamental pains of voice AI—latency, uninvited speech, wake word dependency, and lack of interruption awareness—all at a cost more than 50% lower than competitors. It enables a new class of voice applications where AI collaborates naturally with humans.