Build Your First Voice AI Agent on Acefone 

Every sales team hits the same wall. Lead volumes grow, but your team doesn’t. Calls wait in queues, follow-ups slip and hot leads go cold. 

That’s the gap Acefone’s Agent Studio (our DIY voice agent portal) is built to close. You can build an AI voice agent that calls your leads, qualifies them, and captures every answer. And you can do it all yourself, from one place. 

In this guide, we’ll walk you through the full journey. You’ll sign in, create an agent, set up its brain and voice, add tools and test it live. Then you’ll see how to launch it and learn from every call. We’ll use a real lead qualification agent as our example throughout since it is a common use case across industries. 

How to Set up your first DIY agent with Agent Studio? 

Agent studio lets you build a voice AI agent end to end, on your own. You sign in and create your agent. You set its AI model and voice, add tools and choose what data to capture. Then you test it, launch it and review every call. All from one place. 

Every agent follows the same eight steps. We cover each one in detail below. 

1. Sign in: Log in and land on your agent dashboard. 
2. Create: Pick “Auto Build” or “Start from Scratch”. 
3. Configure: Choose the AI models for your agent’s brain (LLM), ear (STT) and mouth (TTS). Set up the Denoiser to filter background noise. VAD and End of Turn run automatically. 
4. Instruct: Write the prompt, greeting line and guardrails. 
5. Equip: Add tools, data extraction fields and advanced settings. 
6. Test: Try your agent over a phone call, a web call or chat. 
7. Launch: Schedule and go live through Campaigns. 
8. Review: Check each call in Interaction Logs and track trends in Reports. 

It’s a loop, not a straight line. What you learn in step 8 feeds straight back into steps 3 to 5. 

What do you see when you sign in? 

After you sign in, you land on your agent dashboard. Here you can search every agent you’ve created and manage its versions. Pick any agent to open it and make changes. 

How are Agent Studio’s modules grouped? 

Agent Studio’s modules fall into three groups. Each group handles a different stage of your agent’s life. 

You’ll touch most of these as you build.  

So, let’s create your first agent. 

Auto Build or Start from Scratch? 

When you create an agent, you get two options: 

Auto Build puts you two to three steps ahead. It handles the early setup, so you start further along. 

Start from Scratch lets you set up everything yourself. That’s the route we take in this guide, so you can see every setting. 

Our tip: Use Auto Build when you want a quick start. Pick Start from scratch. 

How do you set your agent’s brain and voice? 

The first page of every new agent is the Config section. This is where you define your agent’s intelligence. You’ll set up three layers here: 

• LLM model: The AI model that does your agent’s thinking and writes its replies. 

• Speech-to-text (STT): How your agent recognizes what callers say. 

• Text-to-speech (TTS): How your agent’s voice sounds on the call. 

Fine-tune each layer to match your use case. A sales call and a support call may need different settings. 

Choose the voice 

In the TTS settings, you pick three things: 

• Language: The language your agent speaks in. 

• Gender: A male or female voice. 

• Speaking speed: Faster or slower, to suit your callers and your use case. 

Add background sounds 

A completely silent line can feel unnatural. You can add office sounds or traffic sounds in the background. Then adjust them to suit your use case. 

Fix words with the Pronunciation Dictionary 

Is your agent saying a brand name or a place wrong? Add the input word and its correct pronunciation to the dictionary. Your agent will then say that word correctly every time. 

Our tip: add your brand name, product names and the cities you serve on day one. 

How do you write your agent’s prompt? 

The Prompt section is where your agent’s whole brain sits. Here you add every instruction: its behavior, its tone and its guardrails. 

A strong voice prompt usually covers four things: 

1. Who the agent is. A name, a role and one clear goal. For example: “You’re Rahul from MG Motors. You qualify leads for a test drive.” 
2. How it talks. Short, spoken replies. One question at a time works best on a call. 
3. Its guardrails. What it must never say or promise. 
4. The call flow. Which details to collect, and in what order. 

Who speaks first? 

You also decide whether your agent speaks first or waits for the caller. Then you add a greeting line to match. 

Set this by use case. On an outbound call, your agent opens the conversation. On an inbound call, the caller may speak first. 

Once your prompt and greeting line are in, click Save changes. Your agent is ready to test. 

How do you test your agent? 

Never send an untested agent to real customers. Our portal gives you three ways to test. 

How do tools give your agent new abilities? 

A talking agent is useful. An agent that can act in real time is far more useful. The Tools section adds those abilities. 

• API Request Connect your agent to external systems. Fetch customer details mid-call, or pass data to your systems live during the call. 

• Transfer: Add a destination number. Your agent can then connect the caller to a human during a live call. 

• Hangup: Tell your agent when and where to end the call, based on your use case. 

• Agent: Build multi-agent workflows. For example, an intro agent can pass a qualified lead straight to a sales agent. 

• DateTime: Help your agent work with business hours, working days and real-time scheduling. Set your showroom’s hours, and it won’t book a test drive when you’re closed. 

How do you capture data from every call? 

The Data Extraction section tells your agent what to capture from each call. It turns the conversation into structured output your team can use. 

For each field, you add a name, mark whether it’s required or not, and write a short description. Here’s the field we set up for our MG Motors agent: 

After the call, the captured value shows up in Interaction Logs. If the caller picks the Hector, you’ll see “Hector” as the interested car model. 

You can add more fields the same way. For lead qualification, think purchase timeline, trade-in status, and city. 

Which advanced settings keep calls on track? 

Real calls get messy. Callers go quiet, talk over the agent or don’t pick up at all. These settings help your agent handle all of it. 

• Turn timeout: How long your agent waits before a turn ends with no response. If the caller goes silent, you set when the agent nudges them. 

• Maximum attempts: How many nudges your agent gives before it disconnects the call. 

• Maximum duration: Leave calls open-ended, or set an exact limit. Say you cap calls at five minutes. Add an optional timeout message like “Thank you for connecting, have a nice day.” 

• Interruption settings: Choose whether callers can interrupt your agent mid-reply. 

• Voicemail detection: Real calls often land in voicemail. Set a custom message for those moments, like “Hello, I’m calling from MG Motors about your interest. Please call back later.” 

Why are backchannel settings important? 

When people listen, they say things like “acha”, “okay” or “hanji”. It’s a listening acknowledgement, not an interruption. 

Most voice agents miss this. The caller says “acha”, and the agent suddenly stops mid-reply. It doesn’t sound human at all. 

Backchannel settings fix this. Add words like “acha”, “okay” and “hanji”. Your agent then treats them as listening cues and keeps talking. 

• Without backchannel settings: Caller says “acha” → agent treats it as an interruption → agent stops mid-reply. 

• With backchannel settings: Caller says “acha” → agent recognises a listening cue → agent keeps talking. 

How do you take your agent live? 

Once your tests look good, head to Campaigns. This is where you schedule your agent and launch it when it’s ready to go live. 

As calls start flowing, Reports gives you combined analytics and insights across all your campaigns. It’s your big-picture view of how every agent is performing. 

How do you read your Interaction Logs? 

Interaction Logs show you what actually happened on every call. Come here after each call to check the details and spot what to improve. You get four views. 

Transcript and recording 

Read the call turn by turn: what your agent said and what the caller said. You can also play the full recording. 

Latency 

See the overall latency of the call, and the latency of each turn. Spot exactly which turn took longer than usual. 

Each turn is split three ways: how long STT, the LLM and TTS took. If one is slow, switch that layer to an option that suits your use case better. 

Logs 

See everything under the hood: what was sent to your agent, what STT picked up and what TTS generated. Handy when something doesn’t behave as expected. 

Analysis 

This is your call summary. It shows: 

• End reason: Whether your agent or the caller ended the call. 

• Barge count: How many times the caller interrupted your agent. 

• Tool failures: Whether any tool failed during the call. 

• Post-call analytics: The data your extraction fields captured, like the interested car model. 

Our tip: a high barge count often means your agent’s replies run too long. Check your backchannel settings too. 

Ready to build your first voice agent? 

You’ve now seen the full journey on the Agent Studio portal. Sign in, create your agent and set its brain and voice. Write its prompt, add tools and choose what data to capture. Then test it, launch it and learn from every call. 

Start small with one lead qualification agent. Get it right, then roll the same pattern out across your other campaigns. 

Want a hand with your first build? Book a demo, and our team will walk you through it.