What a voice agent is
A voice agent is a regular Orbit agent whose channel of work is the phone. On create you attach a system prompt, optional knowledge bases, and the tools it may call; the runtime takes it from there. There is no separate “voice SDK” to wire — the phone call lands on the Orbit voice pillar and the agent answers it. What changes for voice is timing. The caller hears delays, so every turn has to complete in well under a second of speaking time and the agent must handle being interrupted. The Orbit voice runtime manages barge-in and turn-taking for you; your prompt and tools stay the same as for a text agent.How a call moves through the stack
Each spoken turn passes through a managed media pipeline:Inbound: the agent answers your number
For inbound, you route a phone number you own to the agent. Every caller to that number is greeted and served by the agent. The shortest end-to-end walkthrough is the Voice quickstart — claim a trial number, attach your agent, and call it from any phone.Outbound: the agent dials a list
For outbound, you request a call overPOST /voice/calls with an
agent_id. The dial is placed only after the recipient’s compliance
gates pass — quiet hours, do-not-call, and AI-voice consent run before
anything rings, and a rejected call is never billed. For more than a
handful of recipients you run a dialer campaign rather than per-call API
dials. See Outbound calls with an AI voice
agent for the request schema and error
matrix.
Human handoff when the conversation needs one
A voice agent is not a dead end — you can let a caller reach a person, or have the agent escalate on its own when its confidence or policy says so. The full transfer flow, including warm handoff to a specific queue or colleague, is covered in AI agent handback. Set the targets the agent is allowed to reach in handoff targets.Where to go next
- Voice quickstart — claim a number, attach an agent, place your first call in minutes.
- Creating agents — prompt, tools, and knowledge bases.
- Model selection and cost controls — pick latency-appropriate models and cap spend per run.
- Guardrail effectiveness and voice eval runs — test the agent against recorded and simulated speech before pointing a real number at it.
- Live monitor and human-in-the-loop oversight — watch production calls and intervene.