How to Build an AI Agent That Searches the Web

The biggest limitation of any Large Language Model is its static knowledge cutoff. If you want to deploy a reliable voice or chat agent, you need an online search agent in ai that can retrieve real-time data. Voice AI agents powered by accurate, up-to-date information transform customer interactions by eliminating hallucinations.

So, how do you build an AI agent that searches the web? In the past, this required complex RAG (Retrieval-Augmented Generation) pipelines. Today, we are going to show you exactly how to add web search to an AI agent natively using the Syllable Platform. We will break down the exact steps to perform a web search with tools, demonstrating how to use the web_search tool to turn your company’s live website into the agent's primary source of truth. Finally, we will pair this web search agent capability with a static data store so your agent can handle complex online information requests, referencing warranty data, policies, and real-time pricing simultaneously.

0:00 Intro
0:20 What are Web Search and Data Source Search Tools?
0:40 Starting off with AI Agent prompt creation
1:35 Create your Agent's greeting
1:56 Adding data tools to your AI Agent
2:16 Build your Agent with ease
2:54 Chat with your Agent
3:40 Interact with your Agent over voice

I am a huge music fan, and own many instruments, so I'll be using one of my favorite brands as my example, but you can use any website.

Step 1: Creating Your Agent’s Prompt and Tools

Start by creating a prompt tailored to your agent’s purpose. Our Music agent helps customers find product information and policies, so we’ll use two powerful tools:

• Web Search Tool: Enables your agent to search your website, turning it into a single source of truth for dynamic content like product descriptions or availability.

• Data Source Search Tool: Allows your agent to pull information from a static data source, perfect for policies, warranties, or other structured content.

• Click Prompt → New Prompt.

• Fill out the Provider, Model, and Version fields.

• Under Tools, select:

  • Web Search

  • Data Source Search

  • Hang Up

• Clearly instruct the agent to utilize these tools in your Prompt Text.

• An easy example prompt from video:

  • You are a support agent for a music company.

    Your primary function is to assist users with general product information.

    You can use the web_search tool to answer questions related to products.
    You can use the data_source_search tool to look for information related to support policies like refunds, shipping, warranties and department information like address, phone number, etc.

• Click Save.

Step 2: Crafting Your Agent’s Greeting (Messages)

Next, define the greeting your agent will use when answering calls. This is your customer's first impression, so keep it concise and welcoming.

• Click Message → New Message.

• Enter your greeting. For my example, I used:

  “Thank you for contacting Music Company Support! How can I help you today?”

• Click Save.

(You can set multiple greetings based on days or times, but a single greeting works for now.)

Step 3: Adding a Data Source

Now, let’s create your static data source. This is useful for unstructured information that rarely changes, such as refund policies or warranty details.

• Click Data Source → New Data Source.

• Name your data source clearly (e.g., “Company Policies”).

• Type out or paste relevant policy content or structured data into the provided text box. (I copied the e-commerce agent's prompt details)

  • Company policies: Orders are processed within 1-2 business days after confirmation, with standard shipping taking typically 5-7 business days, expedited shipping 2-3 business days, and overnight shipping available for next business day delivery if orders are placed by 2pm PST. Shipping costs are calculated at checkout, offering free standard shipping on orders over $100, and various shipping methods are available, including standard, expedited, and overnight. Once shipped, customers receive a confirmation email with tracking information active within 24 hours, and are responsible for any customs, duties, and taxes on international orders. Returns are accepted within 30 days of delivery for unused items in original packaging. Return shipping costs are free to the customer. Approved refunds are processed within 7-10 business days to the original payment method.

• Click Save.

With your prompt, message, and data source ready, you’re all set to build your agent.

Step 4: Building Your Agent

Navigate to the Agents section to bring everything together:

• Click Agents → New Agent.

• Name your agent clearly (e.g., “Music Company Support Agent”).

• Select the Prompt and Greeting Message you previously created.

• Under Tools Configuration:

  • Click Web Search and input your website URL.

  • Click Data Source Search and select your recently created data source.

• Choose additional preferences (e.g., voice options, busy sounds, etc).

• Click Save.

Your agent is now ready for testing!

Step 5: Testing Your Agent

Test your agent via text to ensure both web and data source searches function correctly,

💡 Tip: You'll see a tag appear on the reply that indicates the tool was called

• Click Test Chat.

  • Enter a query about a specific product detail to test web search.

  • Enter a policy-related question to test data source search.

• Confirm your agent retrieves accurate answers for both queries.

Step 6: Going Live with Voice

Give your agent a real-world test by placing a call. First we'll add a voice channel:

• Click Channels → New Channel.

• In Mode select "Voice."

• In Channel name select "syllable-voice"

• Channel target select your agent's phone number

• Enter your fallback number with country code. In the US that's +1. I used my cell phone # here.

• Under Agent, connect it to the agent you just created.

• Click Save.

• Call the provided phone number and interact with your agent:

Simple, smooth, and effective!

And that’s all there is to it. With web and data source search tools, your AI Agents can deliver reliable and up-to-date information seamlessly.

Ready to build smarter agents for every idea?

Start Building Your Agent Today →