junkai-li/NetCoreKevin

What it solves

NetCoreKevin provides an enterprise-grade AI agent SaaS architecture that simplifies the deployment of AI middle-platform services. It solves the complexity of building scalable, multi-tenant AI applications by integrating knowledge bases, agent orchestration, and skill management into a unified .NET 9 framework.

How it works

Built on a microservices architecture using DDD (Domain-Driven Design), the system leverages several key components:

• AI Orchestration: Uses an AgentFramework for multi-agent collaboration and dynamic skill loading.
• RAG Pipeline: Implements Retrieval-Augmented Generation using the Qdrant vector database for knowledge base queries.
• Model Integration: Supports both cloud-based LLMs (like Zhipu AI) and local models via Ollama.
• Infrastructure: Utilizes Consul for service discovery, CAP and Hangfire for distributed tasks, and Redis for caching.
• Multi-tenancy: Employs a single-database multi-tenant architecture with strict data isolation at the interface level.

Who it’s for

It is designed for enterprise developers and architects who want to build a professional AI-powered SaaS platform using the .NET ecosystem, specifically those needing robust permission control, multi-tenancy, and integrated RAG capabilities.

Highlights

• Full-stack Architecture: Includes a .NET 9 backend and a Vue 3 + Ant Design frontend.
• Dynamic Skill Management: Allows for the online editing and controllable loading of agent skills.
• Local AI Support: Native integration with Ollama for offline model execution.
• Advanced Memory Management: Features chat history compression strategies to optimize context windows.
• Enterprise Tooling: Built-in RBAC permission management, distributed locking, and multi-cloud file storage.