Hi. I wanted to learn more about how Vapi build and deploys voice ai agent. Does it build and deploy voice agents separately or does it run different voice agents in a same runtime engine? You guys are really great at managing the latency.

Hey there! 👋 I wasn't able to find specific information about whether Vapi builds and deploys voice agents separately or runs them in the same runtime engine. However, I can definitely help with how Vapi builds and deploys voice AI agents in general! According to the [documentation](https://docs.vapi.ai/introduction), Vapi uses a modular architecture to build voice AI agents. At its core, Vapi functions as an orchestration layer over three primary modules: 1. The transcriber (Speech-to-Text) 2. The model (LLM) 3. The voice (Text-to-Speech) These three modules can be swapped out with any provider of your choosing, including OpenAI, Groq, Deepgram, ElevenLabs, PlayHT, etc. You can even [plug in your own server to act as the LLM](https://docs.vapi.ai/quickstart). For deployment, Vapi offers several options including: - Calls - Vapi SDKs - Server URLs - SIP Telephony Regarding latency management (which you mentioned Vapi is great at), the [documentation](https://docs.vapi.ai/quickstart) explains that Vapi "optimizes the latency, manages the scaling & streaming, and orchestrates the conversation flow to make it sound human." The system performs each phase in realtime (sensitive down to 50-100ms level), streaming between every layer, with the goal of keeping the whole voice-to-voice flow under 500-700ms. Vapi also runs a suite of specialized models for features like endpointing, interruption detection, noise filtering, and emotion detection to create more natural conversations, as mentioned in the [How Vapi Works documentation](https://docs.vapi.ai/how-vapi-works). Is there anything specific about Vapi's voice agent architecture you'd like to know more about? Sources: - [Vapi Documentation Introduction](https://docs.vapi.ai/introduction) - [Vapi Documentation Quickstart](https://docs.vapi.ai/quickstart) - [Vapi Documentation How Vapi Works](https://docs.vapi.ai/how-vapi-works)

Thank you for reaching out to us. We've received your message and will get back to you as soon as possible.