# Build Custom Websites with AI Agents Source: https://www.youtube.com/watch?v=5FAYO-A0jZA Recap page: https://rapidrecap.app/video/5FAYO-A0jZA Generated: 2025-12-22T11:07:27.286+00:00 --- ## Quick Overview The presenter demonstrates building personalized, one-shot websites using Relevance AI agents orchestrated with Vercel's V0 platform, showing how agents can independently research a subject like Neil Gaiman, generate content, and deploy a functional website via custom API tools. **Key Points:** - The agent successfully created a website for Neil Gaiman, which included sections on his books, awards, and spontaneously generated features like a 'random fact tool' and a summoned bee. - The process involves a multi-step agent workflow: Google search for research, scraping website content, and then using a custom tool to interface with the V0 API for generation and deployment. - The custom V0 tool requires an API key, the generation prompt, and the project name, utilizing a specific API call structure involving system messages like 'next.js and tailwind developer'. - The presenter highlights the self-healing ability of the agent, noting that when a non-compliant country was initially passed to the Google Search tool, the agent retried the search using 'United States' instead of erroring out. - The agent can be layered with other tools, such as integrating with a CRM (like HubSpot) to pull details for generating personalized landing pages, or adding an email tool to notify the subject. - The custom tool setup recommends using an 'API key input' within Relevance for security, requiring the user to match the variable name (e.g., 'Vzero key') in the integration settings. - The process incurs Relevance credits, with the example of the Lovable useless fact website costing 465 credits per run, though costs can be reduced by adjusting the models used. **Context:** The video presents a tutorial on leveraging Relevance AI agents to automate the creation of dynamic, personalized web experiences, specifically using Vercel's V0 platform for the actual site generation. The presenter, who emphasizes not being a professional coder, walks through configuring an agent with research tools (like Google Search and LinkedIn extractors) and a custom tool designed to interact with the V0 API to handle content generation and deployment steps. ## Detailed Analysis The core demonstration involves instructing an AI agent to research an author, Neil Gaiman, and build a personal website using V0. The agent autonomously performs research, scrapes content, and uses a custom-built tool to interface with the V0 API. This custom tool handles the necessary API calls, including setting system context (e.g., 'next.js and tailwind developer'), passing the prompt, initiating generation, waiting for the project ID, creating the project name, and finally publishing the site. The presenter stresses that complex integrations, like figuring out the required API headers and variable passing for V0, can be achieved by asking the Relevance chat feature for assistance. Beyond simple generation, the presenter demonstrates layering capabilities: first, integrating a CRM tool (like HubSpot) to pull specific user data for hyper-personalized landing pages (as shown with the 'Tyler' example), and second, adding an email tool to automatically notify the subject about their newly created website, incorporating value propositions stored in Relevance variables. The presentation concludes by showcasing another use case—a Slack-triggered agent that generates a useless fact, creates a corresponding image via another API (like Banana), and publishes both to a Lovable-hosted website, illustrating the broad application of these web experience agents for sales proposals, marketing assets, or automated content publishing. ### Agent Workflow Steps - Google search for information - Extract website content from sources - Generate website via custom V0 tool - Publish and deploy the live site ### Custom V0 Tool Configuration - Inputs include API key, generation prompt, and project name - Tool uses an API call structure requiring specific headers and authorization - It uses the project ID and chat ID from previous responses to complete deployment ### Security and Integration Practices - Recommends using an 'API key input' in Relevance for security rather than direct copy-pasting - API key variable name must match the name in the integration settings for custom keys ### Agent Self-Healing Demonstration - The agent exhibited self-healing when a Google search input was non-compliant, automatically retrying with 'United States' instead of failing. ### Layered Business Use Cases - Agents can capture details from a CRM (HubSpot) to create personalized landing pages - An additional layer involves creating and emailing the subject about the new website using stored value propositions in variables. ### Alternative Triggers and Platforms - Demonstrated a Slack-triggered agent for creating useless fact websites - This agent also generated custom header images on the fly using an image generation API and published content to a Lovable-hosted page. ### Credit Consumption - Agent runs consume Relevance credits; the complex Lovable example used 465 credits per run, while the simpler Tyler example used 122 credits.