Update Agent
Update an agent’s configuration.
Use this to change voice mode, system prompt, greeting, or voice. Switching voiceMode to “hosted” re-provisions the voice agent with built-in LLM.
Authentication
Bearer authentication of the form Bearer <token>, where token is your auth token.
Path parameters
Headers
Request
When true, hosted-mode agents can send and read SMS/iMessage during calls.
Live only: speaking identity, public call purpose (who you represent and why you call), style, and verbal rules. The voice model only knows what is written here; it delegates everything else. Required with modelTier live. Business instructions stay in systemPrompt. Omit on PATCH to preserve; null or empty clears when Live is inactive.
Hosted-mode only. User-defined function tools the agent can call during a call. Replaces the full set on each update; send [] to clear all tools, or omit to leave unchanged. Ignored in webhook mode. Maximum 20 tools.
Speech-to-text mode. ‘fast’ optimizes for latency, ‘accurate’ optimizes for transcription accuracy (~200ms additional latency).
Background ambience to mask synthetic silence between turns. ‘none’ disables; ‘office’, ‘coffee-shop’, ‘outdoor’ enable a quiet bed.
Audio denoising. ‘noise-cancellation’ (default) handles general noise. ‘noise-and-background-speech-cancellation’ is more aggressive for callers in cars, cafes, or near TVs.
Hang up the call after this many milliseconds of caller silence. Default 600000 (10 min). Raise for IVR/hold-music workflows; lower to fail fast on dead lines. Range 10000 (10s) to 3600000 (1 hour).
Voice speed multiplier. 1.0 is normal pace; lower slows speech (0.5 is half speed), higher speeds it up (2.0 is double). Range 0.5 to 2.0.
When true, the agent interjects short filler words like ‘uh-huh’ or ‘mhmm’ during longer caller utterances. Set false to keep the agent silent while the caller is speaking.
BCP-47 locale that drives the agent’s speech recognition and pronunciation. See the enum for the full set of supported codes.
Response
Speech-to-text mode. ‘fast’ optimizes for latency, ‘accurate’ optimizes for transcription accuracy (~200ms additional latency).
Background ambience to mask synthetic silence between turns. ‘none’ disables; ‘office’, ‘coffee-shop’, ‘outdoor’ enable a quiet bed.
Audio denoising. ‘noise-cancellation’ (default) handles general noise. ‘noise-and-background-speech-cancellation’ is more aggressive for callers in cars, cafes, or near TVs.
Hang up the call after this many milliseconds of caller silence. Range 10000 (10s) to 3600000 (1 hour). Default 600000 (10 min).
When true, the agent interjects short filler words like ‘uh-huh’ or ‘mhmm’ during longer caller utterances. Defaults to true.
BCP-47 locale that drives the agent’s speech recognition and pronunciation. Defaults to ‘en-US’.