Sun is a real-time multi-speaker voice AI model (Collaborative Voice Model) with a long context window, barge-in prevention, interruption handling, and WebSocket API access for building voice-native applications.
At a Glance
Engagement
Available On
Alternatives
Listed Jun 2026
About Sun
Sun is a real-time voice AI platform built by AGI Deeptech Pvt Ltd, accessible at getsun.io. It introduces what the company calls a new model class — the Collaborative Voice Model (CVM) — designed specifically for multi-speaker, real-time conversations rather than simple turn-based voice interactions. The platform is available via a WebSocket API and a web playground for immediate experimentation.
What It Is
Sun is a real-time voice AI API that handles the complexities of natural human conversation: overlapping speech, barge-ins, follow-up questions without wake words, and agent-to-human relay of structured data. Unlike text-to-speech layers or voice wrappers, Sun is positioned as a purpose-built model for collaborative voice sessions — meetings, podcasts, one-on-one conversations, and multi-agent orchestration scenarios. The core interface is a WebSocket connection to wss://api.getsun.io/ws, through which developers send transcription events and receive streamed audio responses.
How the Collaborative Voice Model Works
Sun's architecture centers on a set of event-driven behaviors that make voice feel natural:
- Instant responses: The site claims users hear the AI start speaking within a second of being addressed.
- Barge-in prevention: The agent waits for silence before speaking if someone is already talking, preventing it from talking over participants.
- Interruption handling: If a user interrupts mid-response, Sun stops immediately, waits for the speaker to finish, then decides whether to respond to the new input or go quiet.
- Follow-up window: After finishing a response, Sun listens for 20 seconds for follow-up questions without requiring a wake word, allowing up to 2 follow-ups per window.
- Speech injection: External systems or background agents can push content into the live session via
inject.verbatimorinject.naturalevents, which Sun speaks at a natural moment without disrupting conversation flow. - Trigger word detection: The agent distinguishes between being directly addressed by name and casual name mentions, avoiding uninvited responses.
Context Window and Architecture
The homepage claims Sun Zero has a context window of 350,000 tokens — described as "enough for hours of realtime collaboration" — and positions this as 10x larger than ChatGPT Realtime and Google Gemini Live, which the comparison table characterizes as supporting only "few minutes" of sustained conversation versus Sun's "few hours." These are vendor-published claims. The underlying audio format is 24kHz, 16-bit mono raw PCM, streamed as base64-encoded chunks over WebSocket. Sessions are ephemeral — all conversation state is lost on disconnect, with no session resumption.
Integration and Agent Orchestration
Sun is designed to act as a bridge between human participants and external systems during live sessions. Key integration patterns documented in the API include:
- Agent-to-Human Relay: Structured JSON outputs from other agents can be converted into natural speech via
inject.natural. - Built-in web search: The agent automatically performs real-time lookups for facts, prices, and current events without additional configuration.
- Dynamic context updates: The
context.updateevent allows background information to be changed mid-session, taking effect on the agent's next response. - Multi-pattern support: The docs describe six named patterns — one-on-one conversation, meeting with agent, agents/people/systems sharing updates, proactive contributions, podcast-style sessions, and announcements.
Developer Setup Path
Connecting to Sun requires a WebSocket client and an API key (prefixed sk-sun-) generated from the dashboard at app.getsun.io. The connection flow is: open WebSocket → receive connection.established with session_id → send config.update and context.update → begin streaming transcription events. The API supports JavaScript and Python examples in the documentation. Rate limits are 200 messages per 60 seconds per connection, with a maximum message size of 64 KB and context limited to 4,000 characters.
Current Status
Sun appears to be in active commercial availability with a tiered subscription model ranging from a free tier to enterprise. The site references a playground at app.getsun.io and a live demo at demo.getsun.io. The company is incorporated as AGI Deeptech Pvt Ltd and operates the getsun.io domain. No version numbers or changelog dates are published in the available sources.
Community Discussions
Be the first to start a conversation about Sun
Share your experience with Sun, ask questions, or help others learn from your insights.
Pricing
Free
Get started with voice AI.
- 15 minutes / month
- 2 concurrent connections
- WebSocket API access
- 1 API key
Starter
For developers building prototypes.
- 100 minutes / month
- 10 concurrent connections
- WebSocket API access
- Unlimited API keys
Pro
Best value for production apps.
- 200 minutes / month
- 20 concurrent connections
- WebSocket API access
- Unlimited API keys
- Priority support
Premium
For high-volume production use.
- 1,000 minutes / month
- 50 concurrent connections
- WebSocket API access
- Unlimited API keys
- Priority support
Enterprise
Custom plans for large-scale deployments.
- Unlimited minutes
- Unlimited connections
- Dedicated infrastructure
- Custom SLA
- Dedicated support
Capabilities
Key Features
- Real-time multi-speaker voice AI (Collaborative Voice Model)
- Barge-in prevention
- Interruption handling and detection
- Follow-up questions without wake word (20-second window)
- Speech injection (verbatim and natural)
- Built-in real-time web search
- Dynamic context updates mid-session
- Agent-to-human relay for structured data
- Trigger word and alias detection
- WebSocket API with low-latency audio streaming
- 24kHz 16-bit mono PCM audio output
- Configurable bot name and trigger words
- Proactive contribution mode
- Injection queue with priority support
- Multi-connection support per tier