About Sun
Overview
Sun is a realtime multi-speaker voice model from AGI Deeptech that lets an AI agent take part in live conversations with more than one person. The company calls it a Collaborative Voice Model rather than a text-to-speech layer, a voice wrapper, or a speech SDK. It targets agent-to-human collaboration, overlapping speech, barge-ins, and long-running conversational state.
Developers reach Sun through a WebSocket API, which the site says they can wire into an app in minutes. The model carries a 350K context window that the page presents as enough for hours of realtime collaboration. Sun also advertises a 99.9% uptime guarantee during peak usage hours.
Key Benefits
- Listeners hear the agent begin speaking within one second of a question.
- Sun tells a real command apart from a passing mention of its name and stays quiet when it is not addressed.
- When a participant barges in, the model stops speaking and answers the interruption.
- It detects active speakers and holds its response so turns pass back and forth properly.
- Follow-up questions work without repeating a wake word, so users ask naturally.
- A 350K context window keeps long sessions coherent instead of dropping after a few minutes.
How It Works
A developer creates API keys in the dashboard and opens a session over the WebSocket API. They set the context for a specific meeting or call so the agent responds in the right frame, then stream events in and receive voice output back. During the conversation Sun tracks who is speaking, takes interruptions, and converts structured output from connected agents into speech for the room.
Use Cases
- Meeting agents answer questions about MRR, churn, and other metrics mid-call so nobody leaves the conversation to open a dashboard.
- Analytics and metrics agents relay JSON or structured output from other agents as natural speech to the team.
- Developers test latency, multimodal interactions, and agent behavior in the playground before connecting the API in production.
- Teams use speech injection to push announcements or notifications into a live call at any moment.
- Organizations running high-volume voice workloads take Enterprise dedicated infrastructure with custom SLA and support.
Why Choose This Product
The site compares Sun Zero against the ChatGPT Realtime API and Google Gemini Live API on context window, price, and collaboration quality. It marks Sun as the only one of the three with collaborative proactive follow-up and interruption handling, and it claims audio token processing at roughly half the competitor cost. Free and paid plans let a team start on the API before committing to volume.
Sun Pros & Cons
- 350K context window supports hours of realtime conversation
- 99.9% uptime guarantee advertised for peak usage hours
- Free tier includes WebSocket API access and 15 minutes per month
- Claims audio token processing at 50% lower cost than competitors
- Handles barge-ins, overlapping speech and wake-word-free follow-ups
- Free plan caps usage at 15 minutes per month
- Free plan allows only 2 concurrent connections
- Unlimited API keys start at the $19 Starter plan
Key Features
Instant Responses
Users hear the agent start speaking within a second of asking a question.
Intelligent Intent
Sun distinguishes an actual command from an incidental mention of the bot's name and does not jump in uninvited.
Natural Interruptions
When a user interrupts, the model stops speaking and addresses the interruption instead of talking over them.
Barge-in Prevention
The model detects active speakers and holds its response so turns flow with proper turn-taking.
Follow-up Ready
Sun stays available for further questions after it speaks, so users never repeat a wake word.
Speech Injection
Developers push announcements or notifications to the agent to speak at any point during a call.
Live Info Access
Integrated web search pulls real-time prices, events, specs, facts and stats without anyone leaving the conversation.
Configurable Context
Dynamic context editing adapts the bot's behavior to a specific meeting or situation while the session runs.
Agent-to-Human Relay
Sun reads JSON or structured output from other agents and converts it into natural speech for the team.
WebSocket API
Free+Low-latency WebSocket access carries text and voice, with scaled connection and minute limits per plan.
Sun Pricing
- 15 minutes of usage per month
- 2 concurrent connections
- WebSocket API access
- 1 API key
- 100 minutes per month
- 10 concurrent connections
- WebSocket API access
- Unlimited API keys
- 200 minutes per month
- 20 concurrent connections
- WebSocket API access
- Unlimited API keys
- Priority support
- 1,000 minutes per month
- 50 concurrent connections
- WebSocket API access
- Unlimited API keys
- Priority support
- Unlimited minutes
- Unlimited connections
- Dedicated infrastructure
- Custom SLA
- Dedicated support
| Feature | Free Free/mo | Starter $19/mo | Pro $29/mo | Premium $99/mo | Enterprise Custom |
|---|---|---|---|---|---|
| 15 minutes of usage per month | |||||
| 2 concurrent connections | |||||
| WebSocket API access | |||||
| 1 API key | |||||
| 100 minutes per month | |||||
| 10 concurrent connections | |||||
| Unlimited API keys | |||||
| 200 minutes per month | |||||
| 20 concurrent connections | |||||
| Priority support | |||||
| 1,000 minutes per month | |||||
| 50 concurrent connections | |||||
| Unlimited minutes | |||||
| Unlimited connections | |||||
| Dedicated infrastructure | |||||
| Custom SLA | |||||
| Dedicated support |
Pricing extracted from the product website and may change. Check the source for current details.
Frequently asked questions about Sun
What is Sun?
Sun is a realtime multi-speaker voice model from AGI Deeptech, described on the site as a Collaborative Voice Model. It is built for agent-to-human collaboration, overlapping speech, barge-ins, tool and agent orchestration, and long-running conversational state rather than serving as a TTS layer, voice wrapper, or speech SDK.
How does Sun differ from a text-to-speech layer or speech SDK?
The page states that Sun is not a TTS layer, a voice wrapper, or a speech SDK. It classifies Sun as a Collaborative Voice Model built for agent-to-human collaboration, overlapping speech, barge-ins, and orchestration-ready tool use.
Does Sun require a wake word for follow-up questions?
No. Sun stays ready for questions after it speaks, so users ask naturally without repeating a wake word. The model also distinguishes an actual command from an incidental mention of its name, so it does not jump in uninvited when someone simply says the bot's name.
What happens when people talk over each other?
Sun detects active speakers and holds its responses so turns flow with proper turn-taking. When a user interrupts, the model stops speaking and addresses the interruption, which the page describes as behaving like a real person in the conversation.
How long can a Sun session run?
Sun carries a 350K context window, which the site says is enough for hours of realtime collaboration. Its comparison table lists a few hours of sustained conversation for Sun Zero against a few minutes for the ChatGPT Realtime API and Google Gemini Live API.
How do developers integrate Sun?
Developers create API keys in the dashboard and connect through the WebSocket API, which the site says takes minutes to integrate. Each session carries text and voice, and plans vary by monthly minutes, concurrent connections, and number of API keys.