Knit: Rank Workspace Documents by Conversation Relevance

This is an introduction to Knit, a memory plugin for DeepSeek Harness. The plugin ranks and categorizes Markdown documents alongside conversations by relevance (BM25 + IDF), supports dual-mode switching, keyboard navigation, and live preview, and is designed to help agents quickly locate relevant context documents. It runs entirely locally, requires no model calls, and can scan the session workspace with automatic refresh.