> ## Documentation Index
> Fetch the complete documentation index at: https://docs.unpod.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Building Your First Voice AI Agent
> From blank canvas to a live, phone-ready AI agent - identity, persona, voice, knowledge base, and telephony
**Last updated:** July 22, 2026 · **Product version:** v2 · **Tested against:** `openapi.yaml` 2.0.0
***
Voice AI agents are replacing traditional IVR systems and transforming how businesses handle calls. Unpod makes it possible to build a production-ready voice agent in under 30 minutes - no ML background required. This guide walks through every step.
## What We're Building
A customer support agent that:
* Answers inbound calls for a SaaS product
* Has a defined persona and voice
* Can look up information from uploaded docs
* Hands off to a human when needed
***
## Step 1: Create Your Agent in AI Studio
Log in to [unpod.ai](https://unpod.ai) and open **AI Studio**. Click **New Agent**.
### Identity Tab
Set the core details:
| Field | Example |
| -------- | ------------------------------------------- |
| Name | Aria |
| Handle | `aria-support` (URL-safe slug, used in API) |
| Language | English (US) |
| Timezone | America/New\_York |
The **handle** is how you reference this agent in the API and telephony config.
***
## Step 2: Define the Persona
The persona determines how your agent thinks, speaks, and behaves. Open the **Persona** tab.
### System Prompt
Write a clear system prompt that defines:
* Role and purpose
* Tone and communication style
* What the agent should and should not do
Example:
```
You are Aria, a friendly and professional customer support agent for Acme SaaS.
Your goal is to help users resolve issues with their account, billing, and product features.
Rules:
- Always greet the caller by name if available
- Be concise - callers are on the phone, not reading
- If you cannot resolve an issue, offer to connect them with a human agent
- Never make promises about refunds or SLA violations without checking policy
```
### Greeting Message
The first thing your agent says when the call connects:
```
Hi, you've reached Acme support. I'm Aria, your AI assistant.
How can I help you today?
```
***
## Step 3: Configure the Voice Profile
Open the **Voice Profile** tab. Choose:
* **TTS Provider**: ElevenLabs, OpenAI, Azure, or Google
* **Voice**: Pick from available voices for your chosen provider
* **Speed**: 0.9-1.0 works well for support calls
* **Stability / Similarity**: Higher stability = more consistent, lower = more expressive
Test voices using the built-in preview before finalizing.
***
## Step 4: Choose Your AI Model
Under the **Advanced** tab, select the LLM powering your agent:
| Provider | Model | Best For |
| ------------ | ---------------- | ------------------------------ |
| OpenAI | gpt-4o | General purpose, high accuracy |
| Groq | llama-3.1-70b | Low latency, cost-efficient |
| Google | gemini-1.5-flash | Multimodal, long context |
| Azure OpenAI | gpt-4o | Enterprise compliance |
For support bots, `gpt-4o` or `llama-3.1-70b` via Groq are solid starting points.
***
## Step 5: Add a Knowledge Base
Upload product documentation, FAQs, or policy docs so your agent can answer accurately.
Open **Knowledge Base** → **Upload Document**. Supported formats:
* PDF, DOCX, TXT, Markdown
* Web URLs (Unpod crawls and indexes the content)
Unpod uses RAG (Retrieval-Augmented Generation) - the agent searches relevant chunks before generating a response.
**Tips for better retrieval:**
* Break large PDFs into topic-focused sections
* Include an FAQ document with exact question phrasing customers use
* Add a "what not to say" policy document to constrain agent behavior
***
## Step 6: Set Up Telephony
Your agent needs a phone number to receive calls.
### 6a: Create a Bridge
Go to **Telephony → Bridges → Create Bridge**. A bridge connects your agent to a voice infrastructure provider.
* **Name**: `support-bridge`
* **Provider**: LiveKit or Vapi
* **Agent**: Select `aria-support`
### 6b: Add a Phone Number
Go to **Numbers → Provision Number**. Select a country and area code. Unpod provisions a real phone number via your connected SIP provider.
### 6c: Link Number to Bridge
Assign the provisioned number to your bridge. Incoming calls to this number will now route to Aria.
***
## Step 7: Test the Agent
1. Click **Preview** in AI Studio to test via the browser interface
2. Call the provisioned phone number from your mobile
3. Check **Call Logs** to review the transcript and see how the agent performed
***
## Monitoring & Iteration
After going live:
* **Call Logs**: Review transcripts to find gaps in knowledge base coverage
* **Analytics**: Track call duration, resolution rate, and sentiment over time
* **Analysis Tab**: Configure post-call extraction - pull structured data (intent, outcome, customer tier) from each conversation
Iterate on the system prompt and knowledge base based on real call patterns.
***
## What's Next
Full platform guide for business users.
Trigger outbound calls to your agent via API.
Detailed knowledge base configuration guide.
Advanced telephony and provider setup.