Docs
Voice and speed
Models
Choose the model behind the replies. Match its reasoning and tool use to the calls you handle.
Calls run on OpenAI models tuned for live conversation.
| Model | Use it for |
|---|---|
gpt-5.6-luna | The default. Fast, low cost and good at ordinary front-desk calls. |
gpt-5.6-sol | The advanced model option on Pro. |
gpt-5.4-mini, gpt-4.1-mini, gpt-4.1 | Older models, kept for assistants that were tuned on them. |
#Settings
temperature0 to 1.5default 0.4- Higher gives more varied wording, lower gives more predictable wording. GPT-5 models ignore it.
max_tokens40 to 2000default 300- The longest single reply. Phone replies are short, so the default rarely needs changing.
prioritybooleandefault true- Uses OpenAI priority processing. The first word arrives about 80 ms sooner, for a slightly higher token price.
racebooleandefault true- Sends each turn twice and uses whichever reply starts first. It smooths out slow responses.
#Bring your own model
On Pro you can point an assistant at any OpenAI-compatible endpoint, such as a fine-tuned model or your own server. Set the provider to Custom, then enter the base URL and a key. The key is stored encrypted and never shown again.