Skip to main content
Configurations determine how an agent listens, reasons, and responds. In Feather, most of these settings live on the agent version so you can tune behavior without creating a new agent identity.

Main Configuration Areas

  • LLM for reasoning and response generation
  • STT for speech recognition on call agents
  • TTS and voice for spoken output
  • Language for locale-aware behavior
  • Turn-taking settings such as interruption and silence controls
  • Message settings such as SMS or email token limits on text-capable versions

Start with Defaults, Then Tune

For most teams, the best rollout pattern is:
  1. Start with known-good default configs
  2. Validate the prompt and tools first
  3. Tune voice or model settings only after the baseline works

Inspect Available Configurations

curl "https://prod.featherhq.com/api/v1/available-voices" \
  -H "X-API-Key: $FEATHER_API_KEY"

What to Tune First

Voice Agents

Prioritize:
  • Voice selection
  • Silence timeouts
  • Interruption sensitivity
  • Maximum call duration

Text Agents

Prioritize:
  • Response length
  • Temperature
  • Inactivity timeout
  • Sender identity for email

Example Version Settings

{
  "llmConfigId": "YOUR_LLM_CONFIG_ID",
  "sttConfigId": "YOUR_STT_CONFIG_ID",
  "ttsConfigId": "YOUR_TTS_CONFIG_ID",
  "voiceId": "YOUR_VOICE_ID",
  "languageId": "en-US",
  "voiceSpeed": 1.0,
  "voiceVolume": 100,
  "initialSilenceTimeout": 30000,
  "finalSilenceTimeout": 10000,
  "minInterruptionWords": 1,
  "minInterruptionDuration": 1.0
}

Configuration Advice

  • Change one variable at a time when testing quality.
  • Keep a stable production baseline for each major use case.
  • Version changes to voice, prompt, and tools together when the customer experience depends on all three.
  • Match tone and pacing to the job, not just brand preference.