cekura-personality-design

First list existing personalities with personalities_list filtered by language, then fork a close match or create from scratch with name, prompt, voice_id, voice_model, background_noise, and language. Always send provider alongside voice_id and verify the result with personalities_retrieve.

Accent is a read-only label derived from the voice, so you change it by selecting a different voice_id. Query personalities_elevenlabs_voices or personalities_cartesia_voices for the target language, pick a voice with the matching accent, and patch the personality with that voice_id and provider.

Evaluator instructions cannot reach the voice layer, so accent, interruption, volume, silence, and noise instructions change nothing about the audio while the run still passes. Call-wide voice behaviors must be set on the personality instead.

A personality defines who is calling and how they sound across the whole call, including voice, language, speed, and background noise. An evaluator defines what the caller tries to accomplish in one test, including its turns, conditions, and expected outcome.

The idle timeout defaults to 10 seconds and fires regardless of instructions. Raise message_plan.idle_timeout_seconds on a personality your organization owns, or use a hold step in the evaluator for a bounded scripted pause.

No, globally available predefined personalities are shared and read-only. Fork one into your project with personalities_fork_create, then apply changes to the fork with personalities_partial_update so other organizations are unaffected.