nashsu/llm_wiki
LLM Wiki is a cross-platform desktop application that turns your documents into an organized, interlinked knowledge base — automatically. Instead of traditional RAG (retrieve-and-answer from scratch every time), the LLM incrementally builds and maintains a persistent wiki from your sources。
TypeScript · ★ 18,340 · Fork 2,111 · 在 GitHub 查看
在手机上浏览源码、订阅每日 AI 趋势日报与项目架构深度解读:获取 CopoHub 客户端(iOS & HarmonyOS)→
GitHub Trending 上榜记录
nashsu/llm_wiki 已在 GitHub Trending 每日榜单上榜 2 天,最高排名第 14,首次上榜 2026-09-10,最近一次上榜 2026-09-11。
| 日期 | 排名 | Star | 当日增长 |
|---|---|---|---|
| 2026-09-11 | #14 | 18,340 | +142 |
| 2026-09-10 | #14 | 18,050 | +94 |
同期上榜的项目
- ayghri/i-have-adhd – A skill to stop your coding agent from burying the answer. ADHD-friendly output.
- bilawalsidhu/gods-eye-view – God's Eye View: a spy satellite simulator in your browser, except the data is real. Live open source spatial intelligence on a photorealistic 3D globe. Official site: godseyeview.ai
- obra/superpowers – An agentic skills framework & software development methodology that works.
- alsk1992/CloddsBot – Open Source AI trading agent that operates autonomously across 1000+ markets - Polymarket, Kalshi, Binance, Hyperliquid, Solana DEXs, 5 EVM chains. Scans for edge, executes instantly, manages risk while you sleep. Agent commerce protocol for machine-to-machine payments. Self-hosted. Built on Claude.
- Tencent/teamai-cli – Make Every Team AI Native
- AlexsJones/llmfit – Hundreds of models & providers. One command to find what runs on your hardware.
LLM Wiki 项目深度剖析与商业评估报告
LLM Wiki 项目深度剖析与商业评估报告
1. 核心定位与商业落地价值
LLM Wiki 精准切入了当前 RAG 生态最被忽视的结构性缺陷:传统 RAG 将知识"检索"与"理解"耦合在每次查询时重复执行,导致 token 成本随查询线性增长、跨文档推理能力受限于 top-k 召回、且无法沉淀结构化知识。项目基于 Karpathy 的 llm-wiki.md 模式,将知识库从"检索即席重算"转变为"编译一次、持续维护"的持久化 Wiki,本质上是一次从 Query-time Compute 到 Ingest-time Compute 的成本前置与资产化。
商业落地场景: 企业研发知识中台:将分散的 PRD、技术方案、竞品分析、会议纪要增量编译为带 sources[] 溯源、[[wikilink]] 交叉引用的结构化 Wiki,且天然兼容 Obsidian vault,可直接嵌入现有 Markdown 工作流。 合规与审计场景:log.md 提供可解析的操作时间线,sources[] 实现"每个 Wiki 页面 → 原始来源"的双向可追溯,满足金融/医疗行业的证据链要求。 AI Agent 的本地知识后端:127.0.0.1:19828 的 token-protected HTTP API + 内置 MCP Server,使其成为 Claude Code / Codex 的"私有知识插件",这是相较纯 RAG 库的关键差异化。
竞品护城河:相较 RAGFlow / Dify(Pipeline 编排重、偏服务端)、Obsidian + Copilot(插件碎片化、无图分析),LLM Wiki 的护城河在于"本地优先 + 知识图谱分析 + Agent 原生接口"三位一体。Louvain 社区发现、4-Signal 相关性模型、Graph Insights 的"知识缺口检测"是纯向量库难以复制的结构性能力。但需警惕:GPLv3 许可证对闭源商业集成构成实质障碍。
2. 核心架构设计与技术深度
三层架构:raw/sources(不可变原始层)→ wiki/(LLM 生成层)→ schema.md + purpose.md(规则与意图层)。这一分离是架构上最优雅的决策——原始层不可变保证了幂等重编译的可能性,生成层可随时通过 rebuild index 重建,而 purpose.md 作为"方向性意图"与 schema.md 的"结构性规则"分离,让 LLM 在每次 ingest/query 时都能获得稳定的语义锚点。
技术栈分层:Tauri v2(Rust backend)+ React 19 + Vite 前端,通过 IPC 桥接。关键点在于将重计算下沉至 Rust: Ingest 链路:SHA256 增量缓存 → 两阶段 CoT(Analysis → Generation)→ 持久化队列(串行、崩溃恢复、3 次重试)→ 自动 embedding(LanceDB)。 Query 链路:Tokenized Search(英文分词 + 中文 CJK bigram)→ Vector Search(可选,LanceDB ANN)→ Graph Expansion(2-hop 衰减)→ Budget Control(60/20/5/15 比例分配)→ Context Assembly(编号引用)。
在 README 中展示上榜徽章
[](https://github.fq6825.top/repo/nashsu/llm_wiki)