A step-by-step guide to building an AI expense approval agent using the Microsoft Agent Framework (MAF) and Auth0 in a C# Blazor Web App on .NET 10. Part 1 of a four-part series covers scaffolding the project with the Auth0 .NET template, configuring GitHub Models (gpt-4o-mini) as the inference backend, implementing a tool-calling agent with session persistence via distributed cache, and wiring up a Blazor chat UI protected by Auth0 authentication. The agent can retrieve hardcoded expense reports and maintain multi-turn conversation state across renders. Future parts will add vector database integration, fine-grained authorization, Gmail tooling via Token Vault, and CIBA-based push approval flows.
Table of contents
What You'll NeedA Quick Look at the Microsoft Agent FrameworkCreating the Project with the Auth0 TemplateAdding the Agent PackagesBuilding the Agent ToolsBuilding the Agent ServiceBuilding the Chat ComponentRunning the ApplicationWhat the Agent Knows About the UserWhere This Leaves Us519 Impressions