Behavior tuning (creativity, precision, latency)
Three sliders that change how Anna talks — when to be creative, when to be precise, when to be fast.
Three behavior controls in Settings → Behavior shape Anna's voice without rewriting the prompt.
Creativity (temperature 0.0–1.0)
Default: 0.4
| Value | Behavior |
|---|---|
| 0.0–0.2 | Strict scripted answers. Use for compliance-heavy verticals (legal, healthcare reporting). |
| 0.3–0.5 | Balanced. Anna stays on-brief but allows minor variation. Recommended default. |
| 0.6–0.8 | Conversational. Better for hospitality, customer-care setups where rapport matters more than verbatim policy. |
| 0.9–1.0 | Creative. Use only for entertainment / pre-screening setups where exact wording doesn't matter. |
Precision mode
Default: on
When enabled, Anna refuses to invent values. If the answer requires a number, date, or status she doesn't know, she says so explicitly ("I don't have that information") instead of guessing. Turn off only for casual prospecting setups.
Latency target
Default: 500 ms (round-trip ASR → LLM → TTS)
| Setting | Effect |
|---|---|
| Ultra-low (300–400 ms) | Uses smaller LLM tier. Sometimes less nuanced answers. |
| Balanced (500–600 ms) | Recommended for most cases. |
| Quality (700–900 ms) | Uses larger LLM. Better at multi-turn reasoning, longer answers. |
When to tune
Defaults work for 90% of setups. Reach for these knobs only if:
- Your callers complain Anna sounds robotic → raise creativity to 0.6.
- Your compliance team flags an inappropriate Anna response → lower creativity to 0.2.
- Calls feel slow → lower latency target to 400 ms.
- Anna gets tripped up on multi-clause questions → raise latency target to 800 ms.
Tuning takes effect on the next call after you save.