Wohnpro Guide

Der zentrale Wissens- und Orientierungsort für gemeinschaftliches Wohnen. Ein KI-gestützter Guide für Hausordnungen, Verträge und Beschlüsse.

🚀 Local Setup

Prerequisites

• Node.js (v18 or higher)
• npm or yarn

Installation

1. Clone the repository.
2. Install dependencies:

   npm install

Configuration Options

You can configure the application to use either Google AI Studio (Recommended for quick start) or Google Cloud Vertex AI (Recommended for enterprise).

Option A: Google AI Studio (API Key)

1. Create a .env file in the root directory.
2. Add your API Key.

   # .env
   API_KEY=AIzaSyYourKeyHere...

3. To obtain an API key, visit Google AI Studio.

Option B: Google Cloud Vertex AI

1. Ensure you have a Google Cloud Project with the Vertex AI API enabled.
2. Configure your .env file with the following variables instead of (or in addition to) the API Key:

   # .env
   GCP_PROJECT=your-gcp-project-id
   GCP_LOCATION=us-central1

   Note: If running in a browser environment without a proxy, you may still need to handle authentication (e.g., OAuth tokens) depending on your network setup.

Option C: Nextcloud Backend (Enterprise)

For production environments where document management and user administration should be handled externally via Nextcloud: 👉 Read the Nextcloud Integration Guide

This setup allows:

• Document management via Nextcloud folders (drag & drop).
• Automatic PDF/DOCX to Text conversion via Webhooks.
• User management via a users.csv file.

⚙️ Configuration Reference

The application can be configured using environment variables (e.g., in a .env file).

Variable	Description	Required For
API_KEY	Google Gemini API Key.	Standard Mode
GCP_PROJECT	Google Cloud Project ID.	Vertex AI Mode
GCP_LOCATION	Google Cloud Region (e.g. us-central1).	Vertex AI Mode
NEXTCLOUD_URL	Base URL of your Nextcloud instance (e.g. https://cloud.example.com).	Nextcloud Mode
NEXTCLOUD_USER	Username or Bot Account name for WebDAV access.	Nextcloud Mode
NEXTCLOUD_PASSWORD	App Password for the user (Generated in Nextcloud Security Settings).	Nextcloud Mode
NEXTCLOUD_ROOT_FOLDER	The root folder name where 02-Bot-Memory resides. Default: FAQ-Bot-Project.	Nextcloud Mode
WEBHOOK_SECRET	Secret token to verify webhook requests in the backend.	Backend Server

Running the App

npm start
# or depending on your build tool (e.g., Vite)
npm run dev

The application will launch at http://localhost:3000.

🐳 Running with Docker

You can run the entire stack (Frontend + Optional Middleware) using Docker Compose.

1. Ensure your .env file is configured (see above).
2. Run the containers:

   docker-compose up --build

3. Access the services:
   • Frontend: http://localhost:3000
   • Backend Middleware: http://localhost:3001

Running the Backend Middleware (Manual Node.js)

To enable automatic document conversion via Nextcloud Webhooks without Docker, run the Node.js middleware:

1. Ensure you have installed backend dependencies: npm install express webdav mammoth pdf-parse.
2. Run the server:

   ts-node backend/server.ts

   (Or compile TS to JS and run with node).
3. Configure your Nextcloud Webhook to point to http://YOUR_SERVER_IP:3001/nextcloud-webhook?secret=YOUR_SECRET.

---

🤖 Models & Capabilities

This project uses the @google/genai SDK to access specific Gemini models.

1. Text & RAG

• Model: gemini-3-flash-preview
• Purpose: Used for the main chat interface, generating FAQs from documents, and verifying user invitations.

2. Speech-to-Text (STT) & Text-to-Speech (TTS)

• Model: gemini-2.5-flash-native-audio-preview-12-2025
• Purpose: The "Voice Mode" uses the Multimodal Live API.
• How it works: We do not use separate STT (Transcription) or TTS (Synthesis) models. Instead, we open a bidirectional WebSocket connection to the model. We stream raw audio bytes in, and the model streams raw audio bytes out. This provides much lower latency than traditional STT -> LLM -> TTS pipelines.

---

💾 Database Structure

Currently, this application uses the browser's localStorage for persistence. For a production deployment, use a backend database (e.g., PostgreSQL).

Table	Description
users	Stores authentication and authorization details (role: USER/ADMIN).
documents	Stores extracted text content for RAG. Critical: The content column holds the full text used for AI context.
personas	Defines target audiences for dynamic FAQ generation. Contains the "Meta-Prompt".
faqs	Stores pre-generated questions and answers linked to documents and personas.
chat_sessions	Stores conversation history.