dsh-kb-rag

Your local documents, notes, and crawled articles keep growing, and you want semantic search plus knowledge linking without shipping a single byte to a cloud API. dsh-kb-rag wraps a complete RAG pipeline into a DSH plugin so you can query, ingest, crawl, and relate local knowledge straight from your conversation. The plugin is a thin forwarder; the heavy lifting lives in a Python side-project that uses Ollama bge-m3 for embeddings and ChromaDB for storage, with zero npm runtime dependencies and zero API cost. Your files never leave the machine.

Four tools cover the full workflow: kb_query returns top-k chunks with source and similarity scores; kb_ingest performs manifest-hash-idempotent incremental ingestion with optional front-matter stripping; kb_crawl batch-fetches URLs into clean Markdown via Scrapling and markdownify; kb_related recommends adjacent documents through a retrieval-style or graph-style knowledge walk. Everything is invoked as a simple tool call with no need to remember underlying script flags.

It is aimed at local knowledge workers, SiYuan note-takers, and anyone who needs fully offline, privacy-preserving search and curation without surrendering their documents to a third-party service.