# Build a VOICE Agent That Sells Your Product 24/7 Source: https://www.youtube.com/watch?v=ZPD-Ky0_UhY Recap page: https://rapidrecap.app/video/ZPD-Ky0_UhY Generated: 2025-12-16T15:34:49.745+00:00 --- ## Quick Overview The video demonstrates how to build a custom voice agent using Vapi that automatically sells products 24/7 by leveraging custom tools connected to an Airtable database for product information and conversational memory for context retention across calls. **Key Points:** - The voice agent is configured using Vapi to handle two main tasks: answering product questions via the 'Get Product Details' tool and booking sales calls via the 'Book Call' tool. - The agent's system prompt is customized to define its role as a professional voice assistant for SGP Labs, outlining its tasks, greeting, and interaction guidelines. - Two custom tools were created in the Tools library: 'get_product_details' (to fetch product info from Airtable) and 'book_call' (to schedule appointments). - The 'get_product_details' tool's prompt was enhanced to specifically request brief answers for general questions and allow the user to ask for more detailed information on specific products. - The agent uses a Simple Memory node configured with the call ID as the key to retain conversation context across subsequent calls. - The workflow connects the initial webhook trigger to the AI Agent, which then calls the necessary tools based on user intent, and finally responds back to the webhook. - The entire system is designed to be self-sufficient, sourcing product information from an external Airtable base, allowing the agent to sell products like the $5,000 'Kontent Engine DB™ Extreme' product automatically. ![Screenshot at 00:00: A high-level overview of the Vapi workflow diagram showing the AI Agent node connected to necessary tools like OpenAI Chat Model, Simple Memory, get\_product\_details, and get\_products, which forms the core logic of the sales automation agent.](https://ss.rapidrecap.app/screens/ZPD-Ky0_UhY/00-00-00.png) **Context:** The video presents a tutorial on creating a sophisticated, automated voice agent using the Vapi platform, specifically tailored for sales tasks for a fictional company called SGP Labs. The agent is designed to answer customer inquiries about products and book sales calls, utilizing custom tools that interact with an external Airtable database acting as the product knowledge base (RAG). The presenter walks through setting up the agent's core logic, configuring external tools, and ensuring conversational memory persists across interactions. ## Detailed Analysis The tutorial details the process of building a 24/7 sales voice agent on Vapi. The setup begins with establishing a workflow triggered by a webhook (00:07). The core logic is encapsulated in an AI Agent node (00:01) which utilizes an OpenAI model (01:46) and Simple Memory for context retention (09:49). Crucially, the agent is equipped with two custom tools defined in the Tools library: 'get_product_details' and 'book_call' (00:44). The system prompt is meticulously constructed to instruct the agent on when and how to use these tools, providing guidelines for conversational flow and specifying that the 'get_product_details' tool should be used to answer product feature, pricing, and availability questions, while also providing brief answers for general questions (03:09, 07:17). The tool definitions require specifying the expected input parameters, such as 'product_question' for the product details tool (06:15). The agent is connected to an external Airtable database (08:54) to source up-to-date product information, which is indexed via embeddings (00:00). The webhook response is configured to correctly format the tool calls output back to the platform (10:50). The workflow structure demonstrates how the initial webhook entry point feeds into the AI Agent, which orchestrates calls to the RAG tool ('get_product_details') and the booking tool ('book_call') before sending the final response back via the 'Respond to Webhook' node (10:59). The presenter emphasizes that by using the call ID in the Simple Memory key, the agent retains context across subsequent calls (10:03). ### Platform Setup & Initialization - The process starts by creating a new AI Assistant using a Blank Template in Vapi (01:35); Integrations for OpenAI and necessary voice providers (like Vapi's built-in voices) must be configured in Settings (00:57). ### Tool Definition - Two custom tools, 'get_product_details' and 'book_call', are defined in the Tools library (05:35) to enable the agent to query the Airtable knowledge base and schedule appointments, respectively. ### System Prompt Engineering - The AI Agent's system prompt is defined to mandate using the defined tools for specific tasks (product questions use 'get_product_details', sales calls use 'book_call') and includes explicit guidelines for conversational style (04:14). ### RAG Integration - The 'get_product_details' tool is configured to query product information stored in an external Airtable database, which is indexed using vector embeddings (08:41). ### Context Management - A Simple Memory node is connected to the AI Agent, using the call ID mapped from the webhook input as the unique key to maintain conversation history across calls (09:52). ### Workflow Assembly & Testing - The final workflow connects the initial Webhook trigger to the AI Agent, which then routes calls to the tools and memory nodes (07:26); testing confirms the tool calls are correctly structured in the webhook response (11:50). ![Screenshot at 00:00: A high-level overview of the Vapi workflow diagram showing the AI Agent node connected to necessary tools like OpenAI Chat Model, Simple Memory, get\_product\_details, and get\_products, which forms the core logic of the sales automation agent.](https://ss.rapidrecap.app/screens/ZPD-Ky0_UhY/00-00-00.png) ![Screenshot at 00:08: The payment screen showing a product costing $5000, illustrating the high-value items this automated agent is designed to sell.](https://ss.rapidrecap.app/screens/ZPD-Ky0_UhY/00-00-08.png) ![Screenshot at 00:44: The Tools section within the Vapi interface, showing the custom functions created, including 'book\_a\_call' and 'get\_product\_details', which are essential for the agent's sales capabilities.](https://ss.rapidrecap.app/screens/ZPD-Ky0_UhY/00-00-44.png) ![Screenshot at 01:00: The Integrations settings screen within Vapi, displaying connected services like ElevenLabs, OpenAI, and Anthropic, which power the agent's voice and intelligence capabilities.](https://ss.rapidrecap.app/screens/ZPD-Ky0_UhY/00-01-00.png) ![Screenshot at 04:04: The initial system prompt generated by Claude \(or similar AI\) based on the user's request, which outlines the agent's role, main tasks, and guidelines.](https://ss.rapidrecap.app/screens/ZPD-Ky0_UhY/00-04-04.png)