Creates a new agent; optional fields override server defaults.
Send your organization API key as Authorization: Bearer <key>.
Fields accepted when creating an agent; server defaults apply to omitted values.
Display name (required).
Twilio number id belonging to this org, or null to leave unassigned.
Language model id in provider/model form. Use GET /v1/llms for values accepted on create and update.
Language model id in provider/model form. Use GET /v1/llms for values accepted on create and update.
Whether the agent accepts traffic.
active, inactive One of: multi, en, es, fr, de, it, pt, ru, zh, ja, ko, ar, hi (see Agent.default_language).
Inbound noise suppression profile.
Voice-activity detection timing for turn-taking. On PATCH, send only the fields to change; omitted properties keep their stored values.
Limits and escalation rules for user silence and overall call duration.
Optional list; empty omitted uses server default.
Voice and TTS provider for the agent. On create or update, include id and provider; optional tuning fields (when sent and non-zero) must be between 0 and 2.
Created agent
Voice agent configuration for your organization. llm is the primary model; fallback_llm is used when the primary cannot serve.
Stable agent id.
Assigned Twilio number id for this agent, if any. Empty when unassigned.
Display name.
Language model id in provider/model form. Use GET /v1/llms for values accepted on create and update.
Language model id in provider/model form. Use GET /v1/llms for values accepted on create and update.
Whether the agent accepts traffic.
active, inactive System or base instructions for the model.
Opening line the agent speaks when a call starts.
Window (seconds) where the first message can be interrupted.
Primary speech language: multi, en, es, fr, de, it, pt, ru, zh, ja, ko, ar, or hi (same set as the builder UI). multi is only appropriate for multilingual STT (e.g. Deepgram Nova).
Whether to generate summaries after calls.
Inbound noise suppression profile.
Voice-activity detection timing for turn-taking. On PATCH, send only the fields to change; omitted properties keep their stored values.
Play first_message on outbound calls when true.
Voice and TTS provider for the agent. On create or update, include id and provider; optional tuning fields (when sent and non-zero) must be between 0 and 2.
Limits and escalation rules for user silence and overall call duration.
Hints for transcription or keyword boosting, up to server limits.