Choosing an AI model and a voice for your agent
Fast or most capable? Standard or custom voice? How to pick the right AI model and phone voice for the way your customers talk to you.
By the Sanma team ·
Sanma lets each business choose which AI models its agents may use, and each agent picks one, along with a voice for the phone. You don't need your own AI accounts; models and voices are included in your plan. Here is how to choose.
Models: fast versus most capable
Sanma works with models from OpenAI (GPT), Anthropic (Claude), and Google (Gemini). As a rule of thumb:
- Fast models reply quickest, which matters most on phone calls, and are ideal for everyday questions, bookings, and orders.
- More capable models handle long, nuanced conversations and complicated questions better, at a higher cost per reply.
Most businesses start with a fast model and only move up if conversations regularly need more judgment. You can switch any time in Settings, and if a model is ever unavailable, replies automatically use your default model instead.
Voices: natural, custom, or your own
On the phone, your agent speaks with a natural premium voice powered by ElevenLabs. You can also use a custom voice, including a clone of a real person's voice such as the owner's, created with ElevenLabs and assigned to your agent. Only clone a voice with the clear permission of the person it belongs to. If the premium voice service is ever unavailable, calls continue with a standard voice rather than going silent.
Language
Each agent can speak English, Spanish, French, German, or Hindi. Businesses serving several communities can run one agent per language.
Test before customers hear it
Use the test chat in your dashboard to try your agent with real questions, and call your number yourself before going live. Every change to an agent is saved as a version you can restore with one click. Curious how it sounds? Try a voice conversation with our own assistant.
