wly8691-jpg/knowlp-rag

KnowLP-RAG: dual knowledge-graph RAG for Markdown notes — dsh plugin add @eqman00003/knowlp-rag · MCP + native Cordis plugin for DeepSeek Harness (dsh) & Claude Code

Bundle verified MIT Python Unknown
Bundle verified

Listed

3

Memory

Bundle verified

★ 3 View on GitHub
VersionUnknown
LanguagePython
LicenseMIT
View on GitHub

Preview

Preview 1 of 2: wly8691-jpg/knowlp-rag
Preview 2 of 2: wly8691-jpg/knowlp-rag

What it does

Dual knowledge-graph RAG for Markdown notes: prerequisite + similarity graphs (P/S-Agent traversal), paragraph chunking, n-gram/embedding hybrid search, and an explicit weight feedback loop — via MCP.

Best for

  • Markdown note collections where prerequisite relationships and similar alternatives should shape retrieval order.
  • DSH or Claude Code workflows that need paragraph-level hybrid retrieval across n-grams, embeddings, and knowledge graphs.
  • Knowledge bases that can benefit from explicit feedback to evolve graph weights over time.

Not ideal for

  • Collections that are not stored as Markdown notes or do not need relationship-aware retrieval.
  • Setups that cannot provide both a Markdown vault path and a writable graph-index directory; without them, only full-text search remains.
  • Users seeking a simple one-off keyword search without indexing, traversal, or feedback.

README

KnowLP-RAG

双图检索 + 衰减遗忘的记忆插件 —— 把你的 Markdown 笔记建成会”用进废退”的知识图谱,给 DSH / Claude Code 提供带阅读路径的检索:该读哪几篇、按什么顺序读、哪篇是相似替代。

Python License: MIT

快速开始(3 步)

# ① 安装(npm 官方源)
dsh plugin add "@eqman00003/knowlp-rag"

# ② 必配两个环境变量(不配则双图引擎空转,只剩全文搜索)
export KNOWLP_VAULT="$HOME/Notes"              # 你的 Markdown 笔记目录
export KNOWLP_GRAPH_DIR="$HOME/.knowlp-dsh"    # 可写索引目录

# ③ 重启 dsh web,首次搜索触发 Python 环境自举(约 30s,勿中断)

五个工具

工具 作用
knowlp_search 四引擎扇出检索(双图 P/S-Agent + 向量 + 全文)
knowlp_get_note 读笔记内容(只读,防路径穿越)
knowlp_stats 引擎/图健康度自检(排障第一入口)
knowlp_record_feedback 显式反馈(权重闭环唯一入口)
skill_search 技能索引检索

文档

为什么是 KnowLP?

Grep 搜 “RAG architecture” 给你 105 个文件。KnowLP 给你 3 条带依赖上下文的排序命中。

  grep 朴素向量库 KnowLP
结果排序 ❌ ✅ ✅
依赖链(P-Agent) ❌ ❌ ✅
相似替代(S-Agent) ❌ ❌ ✅
无 GPU 可用 ✅ ❌ ✅(n-gram 模式)
越用越好(反馈) ❌ ❌ ✅(权重闭环)
段落级匹配 ❌ ❌ ✅
衰减遗忘(用进废退) ❌ ❌ ✅(半衰期三档)

区别:向量搜索找”包含关键词”的文档;KnowLP 找”因你的查询而该读”的文档,附阅读路径。笔记间的权重随使用演化——被消费的边加强,长期不用的边按半衰期衰减(过程性 1 天 / 一般 30 天 / 陈述性永不)。

Demo

$ knowlp_search "RAG architecture"
  1. [HIT]  RAG检索架构.md (score 0.77)
  2. [LINK] 向量数据库选型.md (score 0.61)     ← 前置依赖链
  3. [LINK] 检索评估踩坑记录.md (score 0.42)
  4. [LINK] _索引-阅读顺序 (depth 1)            ← 告诉你从哪里读起
  5. [LINK] _索引-相关概念.md (depth 2)

本地开发

git clone https://github.com/wly8691-jpg/knowlp-rag.git
cd knowlp-rag
pip install -e .            # 生成 knowlp-mcp / knowlp-build / knowlp-search

# config.yaml 里配 vault → 建图 → 检索
python build_graph.py
python knowlp_search.py "RAG architecture"

View Architecture Diagram

Frequently Asked QuestionsFAQ

Use the verified command dsh plugin --profile default add github:wly8691-jpg/knowlp-rag in a DSH-enabled shell. The command resolves the public package metadata and keeps the plugin attached to the catalog identity shown on this page.

volcengine/OpenViking#examples/dsh-memory-plugin★ 28,960

OpenViking memory and context bundle for DeepSeek Harness: pre-step auto-recall and profile injection, session capture, `viking://` URI guarding, and recall/write memory tools backed by an OpenViking server.

AGPL-3.0
vectorize-io/hindsight#coding-agents★ 20,126

Hindsight, agent memory that learns: long-term project memory with auto recall and retain, knowledge pages, deep reflection, and per-repo memory banks.

MIT
Ikalus1988/MisakaNet★ 404

Failure-recovery memory: search and record failure-recovery lessons from real engineering sessions, with BM25 + semantic RAG retrieval and a lessons knowledge base.

Apache-2.0
text2future/flowix#dsh-flowix-memory★ 316

Registers the local flowix-cli MCP server so the agent can search, read, create, and edit Flowix memos and mind-map artifacts.

MIT