Build a conference concierge with Agent Context and Anthropic

At a fictive conference, a Telegram bot was developed to assist attendees in navigating the event schedule and content more efficiently, overcoming the limitations of traditional search methods. This bot, which operates within the Telegram platform where attendees are already active, provides real-time answers to questions about sessions, speakers, and workshops using natural language processing. The bot utilizes Sanity's Agent Context to access the conference content model, ensuring read-only permissions and allowing for model flexibility and controlled conversation flow. This setup contrasts with a separate organizer bot that uses the Content Agent API for read and write access, highlighting the different integration patterns based on user needs. The choice of Agent Context over Content Agent API for the attendee bot was driven by the need for read-only access, model choice, and explicit control over interactions, making it suitable for public-facing interfaces. The implementation involves creating an Agent Context document in Sanity Studio, configuring environment variables, and using the @ai-sdk/mcp to connect and query content, while maintaining structural boundaries to prevent unauthorized content modifications.