Hybrid Knowledge Platform KI-4-KMU

This project is part of the KI-4-KMU initiative developed at FHNW, which provides SMEs with practical AI tools and methodologies.

It acts as a proxy server that exposes OpenAI-compatible endpoints, automatically injecting retrieved context from the KI4KMU-IngestionLayer into your prompts. This enables document-grounded question answering (RAG) using domain-specific knowledge. The system also launches an integrated Open WebUI container, providing a professional chat interface accessible at http://localhost:3000 immediately after startup.

Features

• OpenAI-compatible API: Implements /v1/chat/completions and /v1/models endpoints.
• Context Injection: Automatically queries the Ingestion Layer and injects results as a context role message.
• Integrated UI: Starts a pre-configured Open WebUI instance for a turnkey chat experience.
• Local Model Support: Fully compatible with LM Studio for secure, local inference.
• Streaming support: Pass-through streaming for real-time, low-latency responses.

Demo

knowledgeplatform-demo.mp4

---

Architecture

Environment Variables

Create a .env file based on .env.example:

BASE_URL="https://openrouter.ai/api/v1"
API_KEY="your-api-key-here"
RETRIEVAL_URL="http://localhost:8001/v1"

Variable	Description	Default
BASE_URL	Upstream OpenAI-compatible API endpoint	[https://openrouter.ai/api/v1](https://openrouter.ai/api/v1)
API_KEY	API key for authentication	(empty)
RETRIEVAL_URL	Retrieval engine API endpoint (Ingestion Layer)	http://localhost:8001/v1

Installation & Running

Due to specific container image requirements and the multi-service architecture (Proxy + Open WebUI), this project must be managed exclusively via Docker Compose.

Local Models (LM Studio)

To use local models, ensure LM Studio is running its Local Server. Update your .env to point to the Docker host gateway: BASE_URL="[http://host.docker.internal:1234/v1](http://host.docker.internal:1234/v1)"

Start the Stack

docker-compose up --build

Once the containers are running:

• API Proxy: http://localhost:8000
• Open WebUI: http://localhost:3000 (Access this in your browser to start chatting)

---

API Endpoints

POST /v1/chat/completions

Create a chat completion with retrieval-augmented context.

Request:

{
  "model": "openai/gpt-4o",
  "messages": [
    {"role": "user", "content": "What is the compression ratio of the engine?"}
  ]
}

GET /v1/models

List available models from the upstream API.

Project Structure

app/
├── main.py                 # FastAPI entry point
├── api/
│   ├── chat.py             # Chat completions & Context Injection
│   └── models.py           # Models listing router
├── core/
│   ├── config.py           # Configuration management
│   └── client.py           # API Client
├── models/
│   ├── schemas.py          # API models
│   └── retrieval.py        # Retrieval API models
└── utils/
    └── context_injector.py # Logic for merging context into messages

Testing

docker-compose run --rm app pytest tests/ -v