Quickstart Guide

Build and test your first conversational voice AI assistant in under 3 minutes.

In this tutorial, you will configure an assistant, give it a custom system prompt and voice persona, and place an in-browser test call via WebRTC.

Open Assistant Studio →

Step 1: Open Assistant Studio

Navigate to the Assistant Studio to view your existing voice assistants or create a new one:

1. In the left sidebar of the dashboard, expand Ecosystem and click Assistant.
2. Click the Create New Assistant button in the upper-right corner.
3. Select a starter template (e.g. Customer Support, Lead Qualification, Appointment Booking) or click Blank Assistant.

Assistant Studio Listing

Step 2: Configure System Prompt & Model

Once your assistant is created, open the Command tab:

1. Opening Message: Enter the initial greeting the assistant will speak as soon as the call connects:
   Plain textCopy

   Hello! Thanks for calling Acme Support. My name is Alex. How can I help you today?

2. System Prompt: Define the persona, conversational guardrails, and role:
   MarkdownCopy

   You are Alex, an expert support representative for Acme Corp.
   
   - Speak in a warm, concise, and professional tone.
   - Answer customer questions directly in 1-2 sentences.
   - If you do not know the answer, politely offer to transfer to a human specialist.
   - Never invent order statuses or billing details.

3. Provider & AI Model: Select OpenAI (gpt-4o or gpt-4o-mini) or Google Gemini (gemini-1.5-flash).
4. Max Tokens: Set to 250 for crisp, conversational responses without monologue delays.

Assistant Command Configuration

Step 3: Select Voice & Speech Settings

Switch to the Speech tab to choose a voice persona:

1. Synthesizer Provider: Choose ElevenLabs or Cartesia Sonic for ultra-low latency voice streaming.
2. Voice: Select from high-fidelity speaker presets (e.g. Rachel, Drew, Callum).
3. Transcriber (STT): Select Deepgram (nova-3) for accurate speech recognition and background noise suppression.
4. Click Save Config in the top navigation bar.

Assistant Speech Configuration

Step 4: Test Your Assistant Live

You don't need a phone number to test your assistant. You can test it immediately inside your browser:

1. Click the Test Web Call button in the top action bar of the assistant configuration panel.
2. Grant microphone access when prompted by your browser.
3. Speak naturally to your assistant and listen to the streaming voice response.
4. Watch the real-time transcript to verify latency and pronunciation.

Next Steps

Now that your assistant is running, explore these capabilities:

• Connect a Phone Number to make and receive PSTN calls via Twilio, Plivo, or Vonage.
• Add a Knowledge Base to ground your assistant in your company's PDFs and documentation.
• Add Custom REST Tools so your assistant can look up orders and trigger webhooks.
• Run an Outbound Campaign to dial a CSV list of contacts automatically.