LiveMind
A RAG chatbot that goes live on any website in under a day — no database maintenance, no document uploads, just a URL.
The problem it solves
Most RAG chatbots require you to manually upload documents, keep them updated, and maintain an ever-growing knowledge base. LiveMind sidesteps that entirely. You give it a URL. It crawls the site, chunks the content, embeds it, and starts answering questions. When the site updates, you re-ingest. No stale documents, no manual curation.
How it works
Playwright crawls the target website (sitemap-first, with fallback path discovery). The content is chunked, embedded using nomic-embed-text via Ollama, and stored in PostgreSQL with pgvector. Each user session gets its own UUID for memory. Cosine similarity retrieval pulls the most relevant chunks, which are passed into a local LLM for generation. Re-ingestion clears and rebuilds automatically — one API call.
What makes it interesting
The decision to remove the database maintenance layer entirely came from thinking about what gets in the way when you try to give a non-technical team an AI assistant over their own content. The answer is usually "someone has to keep feeding it documents." LiveMind removes that person from the equation.