KnowLP-RAG

Dual knowledge graph retrieval for your Markdown notes.

Install as a DeepSeek Harness (dsh) bundle: dsh plugin add "github:wly8691-jpg/knowlp-rag#main" — see dsh/README.md. Also runs in Claude Code.

Works with Obsidian, Logseq, Joplin, or any plain Markdown folder. 306 notes → 555 prerequisite edges + 624 similarity edges → P/S-Agent graph traversal + paragraph chunking + embedding + visual PixelRAG.

View Architecture Diagram

---

Why KnowLP?

Grep gives you 105 files for "因子回测". KnowLP gives you 3 ranked hits with dependency context.

	grep	Naive vector DB	KnowLP
Ranks results	❌	✅	✅
Dependency chain (P-Agent)	❌	❌	✅
Finds similar alternatives (S-Agent)	❌	❌	✅
Works without GPU	✅	❌	✅ (n-gram mode)
Gets better with use	❌	❌	✅ (feedback loop)
Paragraph-level matching	❌	❌	✅ (542 chunks)
Visual search	❌	❌	✅ (PixelRAG)

The difference: vector search finds documents that contain your keywords. KnowLP finds documents you should read because of your query — complete with a reading path.

---

Demo

$ knowlp_search "因子回测"
  1. [LINK] 因子回测：从理论到实践.md (score 0.77)
  2. [LINK] 因子库设计.md (score 0.61)     ← Prerequisite chain
  3. [LINK] 回测踩坑记录.md (score 0.42)
  4. [LINK] _索引-阅读顺序 (depth 1)          ← tells you where to start
  5. [LINK] _索引-相关概念.md (depth 2)
  Done.  Top-5 matches returned.