This path takes you from nothing to a production AI voice agent. The agent answers or places calls. It listens and talks in real time. It calls your tools. It hands off to a human when necessary. The sidebar now shows only these steps. Work through them top to bottom with Next.

​
What you’ll build

You will build an agent that:
  • Connects to a call over QUIC.
  • Runs a speech pipeline or a single speech-to-speech model.
  • Detects turns and barge-in.
  • Calls tools mid-call.
  • Reaches callers over the phone network or the browser.
1

See the shape of it

Voice overview — the two primitives (calls + audio bridge) and the wire model.
2

Ship a hello-world agent

Managed-cloud quickstart — keys, install, and your first call with an agent attached.
3

Choose a runtime

Agent runtime, then wire providers with BYO ASR / LLM / TTS (or a speech-to-speech model).
4

Make it feel human

Turn detection & barge-in and tool calling.
5

Reach real callers

Outbound and inbound calls.
6

Put it together & watch it

Building voice agents end-to-end, then observability.
To build a full contact center with human agents and queues, switch to the Contact Center path.