Configure Your Assistant
After creating an assistant, you’ll land on the configuration page. Here’s how to set up each section.
Model (LLM)
The model is the AI brain powering your assistant’s responses.
| Setting | Description | Default |
|---|
| Provider | AI provider (OpenAI, Groq, Anthropic) | OpenAI |
| Model | Specific model (GPT-4o, Llama 3, Claude) | GPT-4o |
| Temperature | Creativity level (0 = precise, 1 = creative) | 0.7 |
| Max Tokens | Maximum response length | 250 |
For voice assistants, keep Max Tokens between 150–300. Longer responses feel unnatural in conversation.
Voice (TTS)
Controls how your assistant sounds.
| Setting | Description |
|---|
| Provider | Voice engine (ElevenLabs, Spitch, PlayHT) |
| Voice | Specific voice selection |
| Language | Voice language (auto-detected from transcriber) |
| Speed | Speaking rate |
| Stability | Voice consistency |
When using African language transcribers (Spitch), the voice provider is automatically set to match.
Transcriber (STT)
Controls how your assistant hears and understands callers.
| Setting | Description | Default |
|---|
| Provider | STT engine (Deepgram, Spitch) | Deepgram |
| Model | Recognition model | Nova-2 |
| Language | Expected caller language | English |
System Prompt
The system prompt defines your assistant’s personality, knowledge, and behavior. Write clear instructions about:
- Who the assistant is
- What it should do
- How it should respond
- Any constraints or rules
You are a friendly customer support agent for Acme Corp.
Help customers with order tracking, returns, and general inquiries.
Always be polite and concise. If you can't help, offer to transfer to a human agent.
First Message
What your assistant says when it picks up a call:
| Setting | Description |
|---|
| First Message | Opening greeting text |
| Mode | assistant-speaks-first or user-speaks-first |
- Tools — API integrations (check order status, book appointments, etc.)
- Files — Knowledge documents (PDFs, docs) for context
Saving
Click Save to persist your changes. If your assistant is already published, changes sync to the voice engine automatically.
