A voice agent is an AI you build once and it then holds a live, spoken conversation with each person who reaches it, instead of typing replies one at a time in a chat window. It answers from your knowledge base, works through the questions you’ve configured for it, and can take a call over the phone or run right on your website.
What it can be
Every agent runs in one of two modes. Voice only gives callers audio with a dynamic voice visualization; Video avatar puts a lifelike talking face on the call instead. Either way, the agent speaks and listens in real time and can hold up its end of a natural back-and-forth, not a scripted menu.
The pieces you configure
Every agent is built from the same four pieces, no matter which mode it runs in:
- Voice & avatar: which mode it runs in, the voice it speaks with, and, for a video agent, the face it shows.
- Instructions & behavior: the personality and instructions that shape how it responds, and how quickly it replies.
- Questions: what it should ask during the conversation, and which field on your account each answer gets saved to.
- Knowledge base: the collections it can draw on so it answers from your documents instead of guessing.
Build one, then put it to work
Once an agent exists, the rest of this section covers everything you do around it: create it from a blank agent or a template, run it through testing with scripted scenarios before real people talk to it, share or embed it so people can actually reach it, then read its conversations and analytics to see how it’s doing and where it needs work.
Related: Create an agent · Voice & avatar · Instructions & behavior · Questions · Knowledge base · Testing · Share & embed · Conversations · Analytics · Voice agent API reference