How to Build an AI Voice Agent in 30 Minutes (No Code)
The full no-code guide to building an AI voice agent with ElevenAgents. Step-by-step walkthrough of every click, prompt, and workflow inside.
This article is sponsored by ElevenAgents. I built the agent, tested it, and documented it the way I do with every tool. All opinions are my own.
If you’ve spent any time around AI audio in the past two years, you’ve run into ElevenLabs. Their voice technology is everywhere. Creators use it to clone their voice for video content. Publishers use it for audiobooks. Game studios use it for in-game dialogue. And a lot of us, myself included, have been using their API inside n8n workflows to build voice agents for months.
That last part was tricky. Building a voice agent through n8n was a pretty complex workflow, and not something you could put together without a developer or technical experience.
Then ElevenLabs launched ElevenAgents, and going from zero to a working voice agent can now be done in as little as 20 minutes, without any technical background.
Which is a big deal, because a lot of businesses need one and don’t have one. You probably interact with a few every week. Maybe you run one yourself.
The dental clinic, the law office, the SaaS company with a support line, the e-commerce store handling returns. When the team clocks out at 6pm, the calls don't stop. And most of what comes in after hours means a frustrated customer who didn't get help when they needed it or a potential one who just moved on.
Once the barrier to building a voice agent drops this low, it's no longer just about missed calls. It's a 24/7 support line, a sales assistant, a booking agent. There for your customers whenever you're too busy or not available.
So in this article, I’ll walk you through what ElevenAgents is and how I put together a customer support voice AI agent for Amplifiers (the AI Blew My Mind MCP), so you can see how to build one just as fast for your own needs.
Whether you want a receptionist, an outbound agent, or something completely different, the process is pretty much the same.
Let’s get into it.
Here’s what we’ll cover
What is ElevenAgents? (and why it matters now)
ElevenLabs has been the go-to for AI voice for a while now. With ElevenAgents, they built a full agent platform on top of it.
Agents that talk, type, and do things.
Under the hood, it has three core parts working together:
The voice. Their text-to-speech engine, which has been the best on the market since day one. Sub-100ms latency, which means the conversation feels natural instead of feeling like a lagging video call.
The brain. A large language model that understands what the caller said and decides what to say back. You can pick the LLM you want, from Gemini Flash to GPT-4o to Claude.
The hands. Tools and integrations so the agent can do things during the call. Check a calendar. Send a text. Log a call in your CRM. Book an appointment. Pull data from a database.
And you can build everything in the dashboard with no code. It’s just buttons and clicks to get your agent up and running.
However, if you’re a developer or you want to build more complex post-call workflows, you can hook it into n8n or Make, or use the API.
What ElevenAgents gives you out of the box
Now, having a voice agent is one thing. Having one that people enjoy talking to is another.
If the agent sounds robotic, cuts people off mid-sentence, can't switch languages, or pulls generic answers that have nothing to do with your business, you're delivering a bad experience. And bad experiences don't bring people back.
Here’s what ElevenAgents gives you to make sure that doesn’t happen:
Turn-taking. The agent detects when someone is done talking. It picks up on cues like “um” and pauses so it doesn’t cut people off mid-sentence.
70+ languages with mid-conversation switching. A caller can start in English, switch to Spanish, and the agent follows. If your business serves clients from multiple countries, or you’re bilingual and want to build a personal voice AI agent, you have over 70 languages to choose from.
Knowledge base. Upload your FAQs, pricing, policies, SOPs, whatever documents describe how your business works. The agent pulls from them during calls, so it has all the information it needs to respond like your own employee would. This is retrieval-augmented generation (RAG), if you want the technical term.
ExpressiveMode. The agent adjusts its emotional tone based on the conversation. Someone calls upset? It softens and slows down. Someone calls excited? It matches their energy.
I linked a video below so you can hear it yourself, because describing it doesn’t do it justice. Listen to the conversation and tell me in the comments, would you say it’s an AI agent?
Which tools ElevenAgents integrates with
ElevenAgents connects to the tools your business already runs on. When a call comes in, the agent pulls data, books calendars, logs records, and triggers workflows while the caller is still on the line.
Here’s what’s available, grouped by category:
CRMs: Salesforce, HubSpot, Zoho, Pipedrive, Monday.com, Jotform. Every call gets logged, every lead gets routed.
Customer support: Zendesk, ServiceNow. The agent can open tickets, pull account history, and escalate with full context.
Scheduling and communication: Google Calendar, Outlook, Cal.com, Calendly, Mailchimp, Slack. Booking appointments, sending confirmations, pinging your team.
Data platforms: Airtable, Asana, Supabase, Palantir Foundry. For when the agent needs to read from or write to a database during the call.
Payments and commerce: Stripe, Shopify. Check order status or process a payment in real time.
Telephony: Twilio, WebRTC, SIP. How the agent connects to a real phone line.
Inference providers: Swap the LLM powering your agent. OpenAI, Anthropic, Google, and others.
Automation: Zapier, n8n, Make. If one of the native integrations doesn’t cover your tool, you can always bridge it through a workflow step.
And if your tool isn't on the list, you can build a custom integration through their API. That's how the more advanced setups get done, connecting tools that aren't natively supported yet. The full directory is at elevenlabs.io/agents/integrations.
60+ ElevenAgents templates to start from
If you want to build a voice agent, you don’t even have to start from scratch.
There are more than 60 pre-built templates. They cover pretty much what you’d expect (and beyond):
Customer support and operations — technical support, IT help desk, order tracking, subscription support.
Receptionists by industry — healthcare, veterinary, legal intake, salons, restaurants, hotels, property management.
Sales and outreach — inbound lead qualifier, AI SDR, appointment setter, outbound sales rep, win-back caller.
Education and HR — language tutors, training roleplay coaches, new hire onboarding, benefits hotline.
Booking and after-hours — appointment scheduler, catering inquiry, after-hours answering, restaurant phone orders.
Finance and feedback — accounts receivable, FDCPA-compliant collections, survey intake, feedback collector.
Most businesses will find something close enough to their use case to start from.
To use a template, select the one you want, then click “Use template” → “Connect your tools” → “Create Agent”.
They already come with the system prompt written, guardrails in place, and a conversation flow mapped out, so the agent knows how to branch a call based on how it goes. You can adapt and change all of it however you want.
After that, you configure it further just like I’ll show you below with the custom agent I built from blank for Amplifiers.
How I built a 24/7 voice AI agent for Amplifiers, step by step
For this walkthrough, I'm building a voice agent for Amplifiers, the AI Blew My Mind MCP that you connect in two minutes to your Claude or ChatGPT and it supercharges it with tools it doesn't have out of the box and 90+ expert prompts across work domains.
Since we launched it a couple of weeks ago, I've received so many questions from you, especially around how to connect it and how to get started.
So I put together a document with the answers to the most frequent ones, plus info about everything inside of it.
Now let’s put it to use and build our voice AI agent.
Get started
First thing, sign up for ElevenAgents using my link. This gives you 11% extra credits when you upgrade.
Make sure you’re on the ElevenAgents platform. ElevenLabs has multiple products (ElevenCreative, ElevenAgents, ElevenAPI), so click the dropdown in the top left and select ElevenAgents.
Step 1. Create a new agent
I didn’t use a template for this one. I wanted to show you how easy it is to set everything up from scratch. So I went to “Agents” in the left side menu, clicked “+ New agent”, and started with a blank setup.
From there, you go through a few steps to start configuring your agent.
First, you pick what type of agent you want to create: Blank Agent, Personal Assistant, or Business Agent. I chose Business Agent.
Then it asks what industry your business is in. I chose Technology & Software.
And what use case you need it for. I chose Customer Support.
Last step: complete your agent. You give it a name, optionally add your website so it can pull public info to personalize the agent, and describe the main goal. There’s also a “Chat only” toggle here if you want a text-based agent without voice. I left it off since I’m building a voice agent.
Then you hit “Create Agent”, wait a few moments for the agent to be finalized, and you’re in the dashboard
Step 2. Configure the agent (system prompt, voice, language, LLM)
When you land in the dashboard, the agent already comes with a system prompt, a voice, a language, and a first message, generated from what you filled in during setup. All of it configurable with clicks, no coding.
Here's how I configured it to make it final and ready to go.
System prompt
The system prompt is the core of your agent. It defines who the agent is, how it behaves, what it should and shouldn’t do.
The auto-generated one was a solid start, but I wanted something more tailored. So I used the ElevenAgents Voice Prompt Builder I created inside Amplifiers.
It interviews you about the agent, then writes the full prompt following a six-section structure that ElevenAgents is tuned to work with: Personality, Environment, Tone, Goal, Guardrails, and Tools.
A few questions, a ready-to-paste prompt, dropped it into the dashboard, done.
If you don't use Amplifiers, you can either tweak the system prompt that's already generated or use the standalone prompt builder I created. Either way, keep the six-section structure. That's what ElevenAgents is tuned to work with.
Set the first message
The first message is what the agent says when the conversation starts. Mine is:
“Hi, thanks for reaching out about Amplifiers. I can help you with setup, troubleshooting, or any questions about the tools inside. What can I help you with?”
Pick the voice
The voice library has a lot of options. For a product support agent, I wanted something warm and encouraging, so after testing a few, I picked Lauren.
You can also turn ExpressiveMode on or off from here. I’d keep it on. You saw how natural it sounds. And there are suggested audio tags you can add if you want the agent to laugh, sigh, or speak with specific emotion. I didn’t select anything here for now.
Select the languages
This is also where I selected multiple languages. I turned on English and a bunch of other languages based on where most of my audience is.
The agent detects which language the caller is using and switches automatically. For a global product with users from different countries, this was a no-brainer.
Choose the LLM
You can pick which language model powers the agent. I kept Gemini 2.5 Flash, which is fast and cost-effective for an FAQ agent. If your agent needs to reason through more complex situations, you can switch to other more powerful ones.
And this is what the full configuration looks like when everything is set:
On the right side, you can already test it from here. But I didn’t want to do that yet. First, I needed to add the knowledge base.
Step 3. Upload your knowledge base
This is where the agent gets the expertise it needs to help people. The system prompt tells it how to behave. The knowledge base tells it what to say.
Navigate to the Knowledge Base tab and click “Add document”. From there you get three options:
Add URL: paste a link to your website, public documentation, or any page you want the agent to reference.
Add files: upload documents directly. Supported formats include PDF, DOCX, EPUB, HTML, and Markdown. This is what I used. I uploaded the FAQ document I had already prepared before starting the build.
Create text: write or paste content directly into the platform.
There’s also more advanced configuration like RAG settings, but we won’t get into that here.
At this point, the agent is configured and could go live as is, after testing it. For a setup like mine where I just need it to answer questions on the Amplifiers website, this would be enough. You don’t even have to connect it to any external tools.
But I wanted to show you how to take it a step further. So before publishing, I connected a tool and built a small workflow inside the agent.
Step 4. Add tools to your voice agent (WhatsApp example)
This is where the agent goes from “answers questions” to “answers questions and does stuff”.
My setup isn’t that complex. I don’t have a CRM or a ticketing system to connect to like you might. But I wanted the agent to send users a WhatsApp summary of the conversation if they need the steps in writing, like for troubleshooting or setup.
Here’s how I set it up:
First, I navigated to the “Integrations” tab in the left side menu. From there, I clicked “+ Add integration”.
I selected WhatsApp from the Telephony category.
Then it asked me to configure the WhatsApp connection and I clicked “Import an account here”.
Which took me to the WhatsApp accounts page where I clicked “+ Import account”.
A new window opened where I had to set up my WhatsApp Business profile. I already had a Facebook page for AI Blew My Mind, so this was just 5 quick steps to connect it.
Once I went through all the steps, I clicked “Connect”. Done.
Next, I went to the “Tools” section in the left side menu.
I clicked “Add Integration tool”, selected my WhatsApp integration, and checked the “Send Message” tool.
Finally, I navigated back to my agent from “Agents” in the left side menu, went to the “Tools” tab, and clicked “Add tool”. WhatsApp was showing up there, so I selected it.
Once it’s added, you can also adjust the System tools on the right side. These let the agent perform built-in actions like ending a conversation, detecting language, transferring to another agent or phone number, and more.
Now that we have the WhatsApp tool set up, we can create a small workflow to integrate it into the conversation flow.
Step 5. Build the conversation workflow
If your agent just answers questions and that’s it, you can skip this. But if you want the agent to do something specific at certain points in the conversation, like send a message, transfer to a human, or change behavior based on what’s happening, this is where you build that.
You connect nodes, set conditions, and the agent follows the flow during the conversation.
My workflow is simple. All I want is for the agent to offer to send the user a WhatsApp message with a summary of the conversation once it’s wrapping up.
If you don’t want to build from scratch, you can start from a template like I showed you earlier. And if you want to dive deeper into how workflows work check the workflows documentation.
My workflow overview
The user and the AI agent have a conversation. When the conversation reaches an end, the agent asks whether the user wants a WhatsApp summary. If they do, it asks for their phone number and sends the message, then ends the conversation. If the user doesn’t want a summary, it just ends the conversation.
How I configured each node
I built this one from scratch (not from a template).
Start node. This is the entry point of the workflow. Nothing to configure here, it just kicks things off.
From the Start node, I clicked the + icon and added a Subagent node.
Subagent 1: Main conversation. This is where the agent has the full conversation with the user. I set the conversation goal to answer all questions about Amplifiers using the knowledge base, help with setup and troubleshooting, and stay in the conversation until the user says they’re done.
Now that the main conversation subagent is in place, I needed a second subagent for when the conversation reaches an end. This one handles the wrap-up: it asks if the user wants a WhatsApp summary of what was discussed.
Subagent 2: WhatsApp summary offer. This is where the agent asks if the user wants a conversation summary sent to their WhatsApp, collects their phone number if they say yes, and sends it. You can see the full conversation goal in the screenshot.
Now we need a transition rule between the main conversation and the WhatsApp summary offer, so the agent knows when to stop answering questions and move on to asking about the summary. I clicked on the connecting line between the two subagent nodes and set up the condition.
Edge: “Conversation wrapping up”. I set the transition type to LLM Condition. This means the agent evaluates in real time whether the conversation has reached an end before moving to the next node and asking about the WhatsApp summary.
From the WhatsApp summary offer subagent node I created earlier, I added two more nodes: a Dispatch tool node (for sending the WhatsApp message) and an End node (for when the user doesn’t want a summary). You need to create the nodes first so you have the connecting lines to configure.
Dispatch tool node. I selected the whatsapp_send_message tool. This is the tool I set up earlier. When the conversation reaches this node, it sends the summary to the user’s WhatsApp.
After the Dispatch tool, I added another End node so the conversation closes once the message is sent. Nothing to configure here. It just ends the conversation.
Now with all nodes in place, I configured the two edges coming out of the subagent node.
Edge: “User wants WhatsApp summary”. LLM Condition: “The user has agreed to receive a summary on WhatsApp and has provided their phone number including country code.” This leads to the Dispatch tool node.
Edge: “User declines summary.” LLM Condition: “The user has declined to receive a WhatsApp summary or said they don’t need one.” This leads directly to the End node.
And that’s the full workflow. Six nodes, three conditions, and my agent now offers a WhatsApp summary at the end of every conversation.
Step 6. Test your voice agent with simulations
This is one of the best parts of ElevenAgents. Instead of calling your agent yourself fifty times to check if it handles every situation correctly, you can run automated simulations.
You set up different scenarios, different types of users, different questions, and the agent runs through all of them on its own. You get to see the full transcript and whether it passed or failed. If something’s off, you go back and adjust the prompt or the knowledge base.
There are three types of tests under the "Tests" tab:
Simulation testing is what I used. It runs a full end-to-end conversation with a simulated user, so you can see the entire workflow playing out without being on the call yourself. You click “Add test”, pick a simulation, and the agent runs through the conversation on its own.
Scenario testing evaluates specific conversational moments. You set up a situation and check how the agent responds to it.
Tool call testing verifies that the agent uses tools correctly and passes the right parameters.
One thing to watch out for when testing: before running a simulation, increase the maximum conversation turns. It defaults to 5, which cut off my first tests before the workflow reached the WhatsApp summary step.
Here’s how one of my simulations went from start to finish. Everything worked smoothly:
You can also hit “Preview” in the top right corner to start a live conversation with the agent yourself. You talk to it, it talks back, in real time.
Step 7. What else lives inside the dashboard
There’s more in the ElevenAgents dashboard that I didn’t use for this build but worth knowing about.
Analysis. Set up post-conversation evaluation. You define what counts as a successful conversation and what data to extract (contact details, issue types, feedback). The agent evaluates this automatically after each conversation. More in the docs.
Widget. If you’re embedding the agent on a website, this is where you configure how it looks.
Security. Set rate limits, data retention rules, conversation recording preferences, and PII handling. Worth spending a few minutes here before you go live.
Advanced. Custom LLM configurations, post-call webhooks, event handling, and more. If you’re building something more complex or integrating the agent into a larger product, this is where you’d go.
Step 8. Go live and iterate
Deploy the agent. Watch the conversation logs for the first week. Every time the agent gets something wrong, the transcript usually shows you exactly what to fix.
Missing info in the knowledge base? Add it. Agent giving too long an answer? Tighten the system prompt. Someone asking a question you didn’t anticipate? Update the knowledge base.
The first version is never perfect. That’s fine. The iteration loop is fast because everything is in one dashboard and only takes a couple of clicks to adjust.
I’m not deploying mine just yet. The Amplifiers landing page and onboarding are still a work in progress. But the agent is built, tested, and ready to go live the moment the site is.
What you can build with ElevenAgents (beyond customer support)
I know this might look like a lot of steps, but that’s because I documented every single one of them. In practice, the whole process took me about 30-40 minutes.
As you saw, you don’t need to be technical to do this. You don’t need to know how to code. You just need to know what your agent should do, write a good system prompt (or use the Amplifier to write it for you), and create a workflow if your agent needs to do more than just talk. That’s it.
You can build a personal voice AI agent for yourself, like a personal assistant. Or one for your business: a customer support agent that handles questions 24/7, a receptionist that takes calls and books appointments, a lead qualification agent that pre-screens prospects and routes them to your sales team, or an outbound agent for follow-ups and campaigns. For your product, your app, your website, wherever you need it.
Either way, the result is the same. You convert more, deliver a better experience, show up when your customers need you and you’re not available, and stop losing people to a voicemail nobody checks.
Sign up using my link to get 11% extra credits at checkout. Set aside 30 minutes, and get started. This is the easiest way to build an AI voice agent right now, and you’ve seen every step in this article.
What will you build? Leave a comment and tell me.
And if you found this useful, share it with someone who could use a voice agent for their business and didn’t know it’s this easy.
This post is free. Paid subscribers get access to all premium prompts and tools inside Amplifiers (the AI blew my mind MCP), weekly premium articles, all premium resources inside the AI blew my mind Lab, and exclusive partner discounts. Upgrade here.
Voice agents are getting easier to build than most people realize.
Fantastic breakdown thanks for the share. Voice AI ftw.