Quickstart
Create your first voice AI agent and make a live test call in under 5 minutes -- no credit card, no API keys, no infrastructure.
What you'll learn
By the end of this guide, you will have a working voice AI agent that you can talk to in your browser. You will also know how to review the call transcript, recording, and cost -- and optionally connect a phone number for real phone calls.
This guide walks you through the full process: sign up, create an agent, make a test call, review the results, and (optionally) connect a phone number. The entire flow takes under 5 minutes.
Prerequisites: A modern browser with microphone access (Chrome, Firefox, Edge, or Safari). That's it -- zoxaAI provides all the API keys and infrastructure.
Sign up for zoxaAI
Go to dashboard.zoxa.ai and create a free account using your email or Google sign-in.
You receive free credits when you sign up -- enough to run dozens of test calls before you need to add a payment method.
No credit card required
You can create agents, make web calls, and explore the full platform without entering any payment information. Credits are deducted automatically as you make calls.
Create your first agent
From the dashboard sidebar, click Agents, then click Create Agent.
The agent editor opens with several tabs. For your first agent, you only need to fill in a few fields on the Agent tab:
Name
Give your agent a recognizable name. This is only for your reference inside the dashboard -- callers never see it.
Acme Support BotSystem prompt
The system prompt is the core instruction set that tells the language model how to behave during the call. This is the single most important field -- it shapes every response the agent gives.
Write it in the second person, addressing the agent directly. Be specific about the agent's role, personality, knowledge boundaries, and response style.
Here is a good example for a support agent:
You are a friendly and professional support agent for Acme Inc, a SaaS
company that sells project management software.
Your job:
- Help callers troubleshoot login issues, billing questions, and basic
product questions.
- Gather the caller's email address early in the conversation so you can
look up their account.
- If you cannot resolve the issue, offer to transfer them to a human agent.
Rules:
- Keep responses concise -- no more than two sentences per turn.
- Never guess at account-specific information. If you don't know, say so.
- Be warm but efficient. Callers appreciate quick resolutions.
- If the caller asks about pricing, direct them to acme.com/pricing.Prompt quality matters
Vague prompts produce vague agents. Instead of "You are a helpful assistant," specify the role, the company, the topics the agent should handle, the response style, and what the agent should NOT do. The more specific your prompt, the better the agent performs.
First message
The opening line the agent speaks the moment the call connects. Make it natural, include a greeting, and set expectations for the caller.
Hi, thanks for calling Acme support! How can I help you today?If you leave this empty, the agent will generate its own greeting based on the system prompt. Setting an explicit first message gives you more control over the opening experience.
Voice
Click the Voice dropdown to browse available voices. You can preview any voice before saving by clicking the play button next to it.
Each voice comes from a TTS provider (ElevenLabs, Cartesia, Sarvam, etc.). The default is a good general-purpose voice, but you can switch to any provider and voice that fits your brand.
Language model
Select a language model from the LLM tab. The default model works well for most use cases. If your agent needs stronger reasoning (complex troubleshooting, multi-step logic), switch to a more capable model.
| Model tier | Best for | Example |
|---|---|---|
| Fast / small | Simple Q&A, greetings, routing | GPT-5.4-mini, Gemini 2.5 Flash-Lite |
| Balanced | General support, sales, scheduling | GPT-4.1, Claude Sonnet 4.6 |
| High reasoning | Complex troubleshooting, legal, medical intake | GPT-5.4, Gemini 3 Flash Preview |
Managed API keys
You do not need to bring your own OpenAI, Anthropic, or Google API keys. zoxaAI provides managed access to all providers out of the box. Just pick the model you want.
Click Save when you are done.
Make a test call
Open your agent and click the Web Call button in the top-right corner of the agent page.
Your browser will ask for microphone permission -- allow it. The call connects within a couple of seconds and you will hear your agent's first message.
Speak naturally. The agent transcribes your speech in real time, generates a response, and speaks it back -- typically in under one second.
Try a sample conversation:
You: "Hi, I'm having trouble logging into my account."
Agent: "I'm sorry to hear that! Can you give me the email address associated with your account so I can look into it?"
You: "Sure, it's [email protected]."
Agent: "Thanks, Sarah. Are you seeing an error message when you try to log in, or does the page just not load?"
When you are finished testing, click End Call.
Tips for a good test
Test edge cases, not just the happy path. Try interrupting the agent mid-sentence, asking off-topic questions, or saying something the agent should refuse. This reveals gaps in your system prompt that you can fix before going live.
Review the call
After the call ends, open the Runs tab on the agent page (or go to Calls in the sidebar for a global view). You will see a full record of the conversation.
Click on the call to open the detail view. Here is what you will find:
| Section | What it shows |
|---|---|
| Transcript | The full conversation, speaker-labeled (AI and Human), with timestamps for every turn. |
| Recording | An audio player with the complete call recording. You can download it as a file. |
| Duration | Total call length in seconds. |
| Cost breakdown | Total cost in USD, broken down by component: LLM tokens, TTS characters, STT minutes, and telephony (if applicable). |
| Status | How the call ended -- completed (normal), failed (error), or other statuses. |
| Metadata | Call ID, agent name, transport type (WebRTC), start and end timestamps. |
Use this data to evaluate your agent's performance. If the agent said something wrong, adjust the system prompt and make another test call. Iterate until the agent handles your key scenarios correctly.
Connect a phone number (optional)
If you want your agent to make or receive real phone calls, you need to connect a telephony provider.
zoxaAI supports eight telephony providers:
| Provider | Inbound | Outbound |
|---|---|---|
| Twilio | Yes | Yes |
| Vonage | Yes | Yes |
| Telnyx | Yes | Yes |
| Plivo | Yes | Yes |
| Cloudonix | Yes | Yes |
| VoBiz | Yes | Yes |
| Tata Smartflo | Yes | Yes |
| Asterisk ARI | Yes | Yes |
To connect Twilio as an example:
- Go to Telephony in the dashboard sidebar.
- Click Add Configuration and select Twilio.
- Enter your Twilio Account SID and Auth Token (found on your Twilio Console dashboard).
- Name the configuration (e.g., "Production Twilio").
- Click Save.
- Click into the configuration and add your Twilio phone numbers in E.164 format (e.g.,
+14155552671). - Go to your agent's Call tab and assign the phone number.
No webhook configuration needed
You do not need to configure any webhook URLs in Twilio. zoxaAI sets up the webhooks automatically when you connect the number through the dashboard.
Once connected, your agent answers inbound calls to that number automatically. You can also trigger outbound calls from the dashboard or the API:
curl -X POST https://dashboard.zoxa.ai/api/v1/call \
-H "X-API-Key: zsk_a1b2c3d4e5f6..." \
-H "Content-Type: application/json" \
-d '{
"type": "outbound",
"toNumber": "+14155551234",
"agentId": "your-agent-id"
}'With no telephony details in the body, zoxaAI dials from the configuration you just connected in the dashboard. To pass provider credentials inline instead, or to pick a specific configuration by id, see Place an outbound phone call.
What's next
What is zoxaAI?
zoxaAI is a managed voice AI platform for building and deploying conversational agents that make and receive phone calls, run browser-based voice sessions, and execute outbound campaigns at scale.
Key Concepts
The core building blocks of zoxaAI -- agents, calls, runs, campaigns, tools, knowledge bases, and providers -- explained with examples, tables, and practical context.