// the find
jonfairbanks/local-rag
Ingest files for retrieval augmented generation (RAG) with open-source Large Language Models (LLMs), all without 3rd parties or sensitive data leaving your network.
A Streamlit app for local RAG over files, GitHub repos, and websites, built on LlamaIndex with Ollama for both chat and embeddings. Aimed at developers who want a fully offline RAG playground without wiring API keys into a cloud LLM.
Multiple ingestion sources (local files, GitHub repos, websites) covered by dedicated components rather than one monolithic uploader. Decent test coverage for a Streamlit project — security controls, browser storage, wheel metadata, and import boundaries all have their own test files, which is more discipline than most repos this size bother with. Docker Compose variants split out for CPU/ROCm/GPU instead of forcing one config on everyone.
Streamlit as the UI layer caps this at demo/prototype use — no path to a real multi-user deployment without a rewrite. Chat and embedding quality are entirely dependent on whatever Ollama model the user pulls, so the RAG pipeline itself isn't the differentiator; you're really evaluating Ollama plus LlamaIndex glue code. No mention of chunking strategy or retrieval tuning in the README, which is the part that actually determines whether RAG output is useful or garbage. Website/repo ingestion at arbitrary URLs is a real SSRF surface for anyone running this on a shared network, even with the 'ingestion guardrails' mentioned.