企业AI集成架构

poc-mcp-llm-stack

Proof-of-concept: an AI architecture combining a Large Language Model (LLM), a Vector Database, and an MCP-style proxy to integrate enterprise systems.
This project demonstrates how easily AI can be integrated into enterprise workflows as part of the Infinito.Nexus vision.

🚀 Installation

Using Kevin's Package Manager:

pkgmgr install pmls

Or clone the repository and start with Docker Compose.

🧩 Components

• Ollama + Open WebUI → run and interact with LLMs
• LiteLLM → OpenAI-compatible proxy (MCP-style)
• Qdrant → vector database for embeddings & retrieval
• MinIO → object storage for documents and knowledge
• Flowise → orchestration and experimentation tool
• Mock APIs → simulated ERP, CRM, and ticketsystem services

⚡ Quickstart

Start the stack

docker compose up -d

Load model (one-time)

docker exec -it ollama ollama pull llama3
# optional: embeddings
docker exec -it ollama ollama pull nomic-embed-text

🔗 Access Points

• Open WebUI → http://localhost:3001 (direct Ollama connection)

• Flowise → http://localhost:3000 (login: admin / admin)

• Qdrant Console/API → http://localhost:6333

• LiteLLM (OpenAI API) → http://localhost:4000/v1 (API key arbitrary, e.g. dummy-key)

• MinIO Console → http://localhost:9001 (admin / adminadmin)

• Mock APIs →

  • Ticketsystem: http://localhost:7010/__admin
  • ERP: http://localhost:7020
  • CRM: http://localhost:7030

🛠️ Building Flows in Flowise

• OpenAI Node

  • Base URL: http://litellm:4000/v1
  • API Key: dummy-key
  • Model: ollama/llama3

• Embeddings Node

  • Base URL: http://litellm:4000/v1
  • Model: ollama/nomic-embed-text

• Vector Store Node

  • URL: http://qdrant:6333

• Tools/HTTP

  • Call ticketsystem:8080, erp:8080, crm:8080 (via container DNS in the same network)

📚 Knowledge Ingest

1. Upload documents to MinIO (e.g., bucket knowledge/).
2. Build a Flowise pipeline: Loader → Embedder → Qdrant.

This enables retrieval-augmented generation (RAG) using your documents. To integrate real systems (e.g., OTRS, ERPNext, Odoo), replace the WireMock containers 1:1 with their official images and update Flowise endpoints.

👤 Author

Created by Kevin Veen-Birkenbach

🙏 Acknowledgments

• Special thanks to Bernd Erk, CEO of NETWAYS, for his inspiring talk at the Bitkom Forum Open Source 2025.
• Thanks also to Bitkom for organizing the conference and providing valuable inspiration.

📖 Context

This project is a PoC within the Infinito.Nexus initiative to evaluate how seamlessly AI can be integrated into modular IT infrastructure.

See the full conversation that led to this project here: ChatGPT Log