What’s in this article
- What ElevenLabs Voice Agents are — The tech for building AI voices that can answer questions from your documents.
- The RAG Agent System Prompt — A copy-paste prompt to make your agent a sales expert.
- How to set it up in 15 minutes — A 5-step guide to get your first agent live.
- What this changes for agencies — How to sell voice agents as a service.
- FAQ — Answers on RAG, versioning, cost, and security.
🚀 Plug this into Claude Code or Claude Desktop
This spec contains the copy-paste system prompt for a sales agent, a 5-step setup checklist for ElevenLabs, and a guide for using the new versioning and built-in RAG features.
Want to turn this into a repeatable service for your clients? We workshop these builds in the Talk-to-Build community.
If you’re building a voice agent for your website, this post gives you the exact prompt and setup to make it an expert that actually closes deals. You’ll build an AI that can answer detailed product questions using your own knowledge base, without you writing a single line of integration code.
This is possible because ElevenLabs just shipped three critical features for their Voice Agents: built-in RAG (so it can read your documents), version control (so you can test scripts safely), and multimodal hooks (so it can send images or links). This isn’t just an update; it’s the stack that makes website voice agents a real sales tool.
What are ElevenLabs Voice Agents?
ElevenLabs Voice Agents are AI-powered conversational tools that can engage in human-like voice conversations. They combine text-to-speech, speech-to-text, and large language models into a single platform, allowing builders to create and deploy voice-based AI assistants for websites, apps, and phone systems. The platform handles the underlying complexity of real-time conversation.
The RAG Sales Agent Prompt You Can Copy Right Now
This system prompt is the brain of your voice agent. You paste this into the agent’s configuration in your ElevenLabs dashboard. It instructs the AI on its role, how to use the knowledge base you provide via the built-in RAG, and how to guide the conversation toward booking a meeting, making it an effective sales tool.
You are 'Alex', a friendly and knowledgeable sales assistant for MK-Way, a web design and AI automation agency. Your primary goal is to qualify leads and book a discovery call for them with the sales team.
**Your Knowledge:**
You have access to a knowledge base containing our services, case studies, pricing, and FAQs. When a user asks a question, you MUST first consult this knowledge base to provide an accurate answer. If the information is not in the knowledge base, you can use your general knowledge but you must state, "I don't have the exact details on that, but generally..."
**Your Process:**
1. **Greet & Inquire:** Start by warmly greeting the user and asking what they're looking for today.
2. **Answer Questions:** Use the knowledge base to answer their questions about our services (AEO, AI Agents, Web Builds).
3. **Qualify:** Gently ask about their project timeline and budget to see if we're a good fit. A good fit has a budget over $10,000.
4. **Call to Action:** If they are a good fit, your main goal is to get them to book a call. Say, "Based on what you've described, it sounds like we can definitely help. The next step would be a quick 15-minute discovery call. Are you free sometime this week?"
5. **Handle Objections:** If they are hesitant, address their concerns using the knowledge base. Do not be pushy.
**Tone:** Be professional, helpful, and concise. Avoid long, rambling answers. Keep your responses to 2-3 sentences.This prompt turns a generic chatbot into a purpose-driven sales agent. The key is the instruction to use the RAG-supplied documents first, ensuring the answers are grounded in your company’s actual offerings.
BEFORE: Simple Chatbot
User: "What's AEO?" --> Agent: Generic LLM definition --> Dead end
AFTER: RAG-Powered Agent
User: "What's AEO?" --> Agent: Checks docs --> "It's our process for...
Want to see a case study?"
Here’s exactly how I’d set this up
The fastest way to get a smart voice agent live is to use ElevenLabs’ built-in RAG and versioning features. This avoids setting up a separate vector database or complex testing environments. You can go from a new account to a deployed, knowledge-aware agent on your website in under 15 minutes by following these five steps.
- Create Your Agent in ElevenLabs. Log into your ElevenLabs account and navigate to the Voice Agents section. Click “Create Agent,” give it a name, and choose one of their realistic pre-made voices.
- Upload Your Knowledge Base. This is the new RAG feature. In the agent configuration, you’ll see an option to upload documents. Start with a simple PDF of your services, a pricing page, or a company FAQ. ElevenLabs automatically indexes it for the agent to use.
- Paste the System Prompt. Copy the system prompt from the asset block above and paste it into the “Instructions” or “System Prompt” field for your agent. This tells the AI its job.
- Create and Test a New Version. Use the new versioning feature to save this as `v1`. Then, click “Create New Version.” In `v2`, make a small change to the prompt, like making the sales pitch more direct. You can now test both versions in the ElevenLabs playground without affecting the live one.
- Deploy to Your Website. Once you’re happy with a version, click “Deploy.” ElevenLabs will provide a simple JavaScript snippet. Copy and paste this snippet into the HTML of your website, and the voice agent button will appear, ready to talk to your visitors.
What this changes for designer-run agency work
Built-in RAG and versioning make sophisticated voice agents a practical, sellable service for agencies. What previously required a developer to integrate a vector database and build a testing pipeline is now a simple configuration step. This shift turns a complex technical challenge into a straightforward product offering.
| Dimension | Old way (Basic Voicebot) | New way (ElevenLabs RAG Agent) |
|---|---|---|
| Knowledge | Relied solely on the LLM’s general training data, often giving generic or wrong answers. | Answers questions using your specific, up-to-date company documents. |
| Setup Time | Weeks. Required developers to set up a vector DB (like Pinecone) and write integration code. | Minutes. Upload a PDF or connect a URL directly in the dashboard. No code needed. |
| Iteration | Risky. Changing the live prompt could break the agent for all users. | Safe. Create and test new versions (e.g., a new sales script) without affecting the live agent. |
For agencies, this means you can now sell and deploy AI agents that provide real, measurable value—like qualified leads and reduced support tickets—without taking on a massive technical risk or project scope. It’s a new, high-margin retainer service.
My $0.02 — How I’d roll this out
If a client wanted a voice agent, I’d use these new features to deliver value in stages, proving the ROI at each step. This isn’t a big-bang project; it’s a three-day sprint to create a new, automated employee for their website.
Day 1 — The Knowledgeable Helper. I’d start by building `v1` of the agent. Its only job is to answer visitor questions accurately. I’d upload the client’s existing FAQ page, service descriptions, and blog posts as its knowledge base. The prompt would be simple: “You are a helpful assistant. Use the provided documents to answer questions.” I’d deploy this on the contact and support pages. The win: instant, accurate answers for customers, 24/7.
Day 2 — The Sales Qualifier. With the base agent working, I’d use versioning to create `v2`. This version gets the sales-focused prompt from the asset above. Its goal is to qualify leads. I’d test `v2` internally to make sure it sounds natural. Once approved, I’d promote `v2` to be the live version. The win: the agent now filters leads and identifies high-value prospects automatically.
Day 3 — The Meeting Booker. Finally, I’d use the new multimodal hooks. I’d configure a function that lets the agent send a Calendly link. When the agent qualifies a lead, it would say, “Great, I can send a booking link directly to you. What’s the best email?” This moves the user from conversation to a booked meeting in one step, a core principle of agentic commerce. The win: a fully automated lead-to-meeting pipeline.
FAQ
What is RAG in an AI voice agent?
RAG, or Retrieval-Augmented Generation, allows a voice agent to access and use information from a private knowledge base, like your company’s documents or website content. This ensures the agent provides answers that are accurate, up-to-date, and specific to your business, rather than relying on generic LLM knowledge.
How does agent versioning work in ElevenLabs?
Versioning lets you create multiple, distinct copies of your voice agent. You can safely edit the prompt, change the voice, or update the knowledge base in a new version without affecting the live agent that customers are interacting with. This is crucial for testing new scripts or behaviors before deploying them.
How much do ElevenLabs Voice Agents cost?
Pricing is based on usage, measured by the number of characters generated. As of mid-2026, plans range from a free tier for small tests to enterprise plans for high volume. The built-in RAG and versioning features are typically included in the standard agent pricing without a separate fee.
Can the agent connect to my CRM or other tools?
Yes. ElevenLabs Voice Agents support webhooks and function calling. This allows the agent to send data to external systems like a CRM to create a new lead, or trigger actions in other applications. This is how you’d connect it to something like GoHighLevel or HubSpot to log conversations.
Is the built-in RAG feature secure for private documents?
Yes, ElevenLabs states that the documents you upload for RAG are stored securely and are only accessible by your specific agent. They are not used to train their general models or shared with other customers. This allows you to use proprietary information safely.
How is this different from building an agent with Vapi or Twilio?
Twilio provides the raw communication building blocks (phone numbers, call streams). Vapi is a developer platform that bundles those with AI models. ElevenLabs aims to be a more integrated, all-in-one solution, now providing the voice, the LLM connection, and the knowledge base (RAG) in a single dashboard, often with less code required to get started.
Want help applying this?
Four ways to go deeper:
- Build with Builders. Join the Talk-to-Build community to learn to build AI-native websites, cinematic AI video, and agent-driven workflows you can sell.
- 1-on-1 working session. Book a screen-share with me — bring a real problem, leave with a working piece of it.
- Done-for-you. MK-Way builds AEO-ready websites, apps, and AI agent workflows.
- Quick question. DM me on Instagram or LinkedIn. I read every message.
Part of the AI Pulse series. If you commented “VOICE” on one of my videos — this is the breakdown. Sources: The ElevenLabs Blog.
Last updated: 2026-06-25.