A DEMO of best-practices to implement a personal local RAG into a Frontier-LLM.

In real production, you should most likely go with three separate dbs for prod, dev, test instead of my TABLE_PREFIX approach.

FastAPI, PostgreSQL (with pgvector), Gemini (Gemini Embedding, Gemini 2.0 Flash-Lite for summaries in pgvector & Gemini 2.5 Flash-Lite for Q&A), pydantic AI, Crawl4AI, curl-cffi, python-markdown

• uv sync
• rename example.env to .env and add external DB credentials and API-Key (optional)

• add crawling fallbacks, in case sitemap is not existing
• add Ollama and/or vllm for more privacy-focused llm inference
• user-based accounts