Voice agents
Voice agents
Everything a voice agent is made of, and where each setting lives in the builder.
Anatomy of a voice agent
A voice agent is one configuration that the voice engine runs on every call. It has five tabs in the builder:
| Tab | What you set |
|---|---|
| Model | Provider, model, system prompt, first message, temperature, max tokens, knowledge base |
| Voice | Voice provider and voice, speed, stability and style, emotion and background sound |
| Transcriber | Speech-to-text provider, model and language, background denoising |
| Functions | Your custom functions: name, description, JSON schema, server URL, sync or async |
| Advanced | Silence timeout, maximum duration, interruption handling, server URL, fallback server URL, server timeout |
Behavior settings such as idle messages, end-call phrases, voicemail detection, recording, compliance and call analysis are on the agent's settings panels.
Lifecycle
- Draft — the agent exists but can't take live phone traffic.
- Live — the agent is synced to the voice engine and assigned to a number.
- Paused — taken offline by you.
- Suspended — taken offline automatically because the workspace's credit balance went below zero. Top up to restore service. See Auto top-up and suspension.
Every save re-syncs the agent to the voice engine. The agent is tagged with its workspace id so events from it can never be attributed to another workspace.
In this section
Model and promptChoose the model and write the system prompt.VoicesPick and tune how the agent sounds.TranscriberChoose how the agent hears.Conversation behaviorTurn-taking, interruptions, silence and ending calls.Prompting guideWrite prompts that work on the phone.Variables and memoryPersonalize calls with caller history.Call analysisSummaries, structured fields and success scores.Voicemail and keypadDetect machines and accept DTMF input.Recording and complianceRecording, consent, HIPAA and PCI.
Last updated November 2026 · Found a problem? Email support@katexs.com
