BookWorm：基于 Aspire 的云原生 AI 多智能体示例项目

Warning

This project is for demo purposes only and is not production-ready.

Introduction

⭐ BookWorm showcases Aspire in a cloud-native application with AI integration. Built with DDD and VSA, it features multi-agent orchestration and standardized AI tooling through MCP with A2A & AG-UI Protocol support.

View Screenshots

Project Goals

• Developed a cloud-native application using Aspire
• Implemented Vertical Slice Architecture with Domain-Driven Design & CQRS
• Enabled service-to-service communication with gRPC
• Incorporated various microservices patterns
  • Utilized outbox and inbox patterns to manage commands and events
  • Implemented saga patterns for orchestration and choreography
  • Integrated event sourcing for storing domain events
  • Implemented a microservices chassis for cross-cutting concerns and service infrastructure
• Implemented API versioning and feature flags for flexible application management
• Set up AuthN/AuthZ with Keycloak
  • Used Authorization Code Flow with PKCE for user authentication
  • Enabled Token Exchange for service-to-service authentication
• Implemented caching with FusionCache
• Incorporated AI components:
  • Use Microsoft Foundry for LLM and embeddings
  • Orchestrated multi-agent workflows using Agent Framework
  • Standardized AI tooling with Model Context Protocol (MCP)
  • Enabled agent-to-agent communication via A2A Protocol
  • Supported Agent interactions via AG-UI Protocol
  • Agent governance with policy-based controls and monitoring
  • Enabled AI agents to generate rich, interactive UIs using A2UI
• Configured CI/CD with GitHub Actions
• Created comprehensive documentation:
  • Used OpenAPI for REST API & AsyncAPI for event-driven endpoints
  • Utilized EventCatalog for centralized architecture documentation
• Built modern client applications:
  • Monorepo architecture powered by Turborepo
  • Customer-facing storefront and admin backoffice dashboard with Next.js
  • Targeted WCAG 2.1 AA accessibility standards (accessibility statement)
• Established a testing strategy:
  • Conducted service unit tests
  • Implemented Pact-based contract testing
  • Established architecture testing strategy
  • Implemented frontend unit tests and component tests
  • Conducted end-to-end testing with BDD

Project Architecture

Getting Started

Prerequisites

Required

• mise — tool version manager
• Docker — container runtime (must be running before starting the app)

Optional

• Buf CLI — gRPC schema & codegen
• Spec-Kit — spec-driven development
• GitHub Copilot CLI — AI-assisted development

Note

• Use mise tasks ls to view available tasks.
• Azure Subscription is required for deploying to Azure Container Apps and using Microsoft Foundry.
• Email uses SMTP through MailKit, with Mailpit locally; configure the endpoint and credentials for providers like SendGrid or production SMTP servers.

Run locally

# 1. Clone the repository
git clone git@github.com:foxminchan/BookWorm.git

# 2. Navigate to the project directory
cd BookWorm

# 3. Install tools (.NET SDK, Bun, JDK & tools — skip if already installed globally)
mise install

# 4. Setup Azure subscription and location for deployment
aspire secret set "Azure:SubscriptionId" "your-subscription-id"
aspire secret set "Azure:Location" "your-location"

# 5. First-time setup
mise run prepare

# 6. Start the application
mise run run

Note

On first run, you'll be prompted to enter the required environment variables.

Self-Deploy to Azure

1. Authenticate with Azure

az login

1. Deploy

aspire deploy

1. Get the app URL

az containerapp show --name <app-name> --resource-group <resource-group> \
  --query properties.configuration.ingress.fqdn --output tsv

1. Clean up resources

az group delete --name <resource-group> --yes --no-wait

1. Clean up resources:

To remove all deployed resources and avoid charges:

az group delete --name <resource-group> --yes --no-wait

Documentation

For full documentation, visit the GitHub Wiki.

Contributing

Contributions are welcome — see CONTRIBUTING and CODE OF CONDUCT for details. Thanks to all contributors!

For bugs or feature requests, open an issue.

License

MIT — see LICENSE for details.