When Notion pages, GitHub issues, and wiki pages each live in different places, building RAG requires creating connectors for every source from scratch. OpenDocuments is a self-hosted RAG tool that comes with that wiring pre-built. One-Line Installation Start with a single npm install. Auto-loading of API keys minimizes manual setup, and the opendocuments doctor command catches configuration errors on the spot. Built on TypeScript and Hono, it is lightweight. Using SQLite and LanceDB means no separate database server is required. Connect Ollama and everything processes locally without external API calls. This architecture works in environments with strict network isolation like financial institutions. Korean-English Cross-Language Queries Work It parses various formats including PDFs and Jupyter notebooks, splits them into semantic units, and stores them as vectors. Search uses a hybrid approach combining vector retrieval and keyword matching. Ask in Korean and it finds answers from English documents, with source documents cited so you can filter out hallucinations. MCP server support means coding agents can query the internal knowledge base directly. The monorepo structure makes it straightforward to add custom parsers and connectors as plugins. Hundreds of test cases are in place, so stability is not a concern when layering custom logic on top. Key Takeaways Ollama integration processes data locally only, making it suitable for air-gapped environments Hybrid search with multilingual support finds English documents from Korean queries MCP server support connects coding agents directly to internal knowledge bases Related Posts Why DataNexus — The problem of structuring scattered domain knowledge How We Chose These Four Open-Source Tools — Evaluation criteria for RAG engine selection Sources https://github.com/joungminsung/OpenDocuments https://news.hada.io/topic?id=27910