Local Q&A chatbot

A local, grounded Markdown Q&A chatbot built with Go, Python, Weaviate, and Ollama Gemma 4 E4B.

Start

1. Copy .env.example to .env if you want to change model or retrieval settings.
2. Put Markdown files in knowledge/.
3. Run docker compose up --build.
4. Open http://localhost:8080.

The first launch downloads gemma4:e4b and embeddinggemma:latest, so it can take a while. The indexer watches knowledge/ and updates the vector index automatically; the UI also has a Reindex button.

Architecture

• Go serves the browser UI and performs retrieval-augmented generation.
• Python parses Markdown, chunks it, embeds it, and synchronizes Weaviate.
• Weaviate stores chunk metadata and externally generated vectors.
• Ollama generates answers with Gemma 4 E4B and vectors with EmbeddingGemma.

Only the web UI and Ollama are published to localhost. There is deliberately no authentication: this is a single-user local application.