NTG Agent V-Next: Agent Delegation and Knowledge Isolation in a Multi-Agent Chatbot Platform

NTG Agent is a multi-agent AI chatbot platform built in C# on .NET 10 and .NET Aspire. It allows organizations to build AI assistants that answer from their own documents rather than from a language model's general training, using retrieval augmented generation.

Key capabilities:

Configurable agents: each agent has its own system prompt, language model, tools and document library, managed through a Blazor admin portal.

Agent delegation: a user facing agent can call specialist agents and combine their answers, keeping each focused on its own domain.

Grounded answers: files, web pages, and text are ingested into a knowledge graph for each agent via LightRAG.

Multiple model providers: Azure OpenAI, OpenAI, Anthropic, Google Gemini and GitHub Models, selectable for each agent.

Extensible tooling: agents connect to external tools through the Model Context Protocol.

Access control: permissions based on roles determine which users can use or manage each agent.

The solution runs as coordinated services orchestrated by .NET Aspire, using SQL Server for application data and PostgreSQL with vector and graph extensions for the knowledge store.