[English] | 简体中文
An evolved methodology inspired by Andrej Karpathy, Astro-Han & Rowboat
The Next-Gen Knowledge Orchestration System that Grows with You
We've moved from "Passive Ingestion" to "Active Observation." LLM Wiki OS is now your autonomous research scout.
Introducing the @Observe directive. You can now command your Agent to "keep an eye" on specific entities or topics.
- Autonomous Monitoring: Periodic polling of OpenCLI sources (X, Reddit, arXiv).
- Silent Updates: The OS automatically crystallizes new findings into your wiki.
- Drift Detection: Tracks how concepts evolve over time in your
log.md.
No more hidden settings. Configure your AI's behavior directly inside your notes:
---
status: observing
observe_interval: 7d
unsolved_mysteries: ["How does this theory apply to the new market data?"]
---Simply edit the Markdown file to change how the AI researches that specific topic.
Your log.md is no longer a boring list of "done" tasks.
- Mystery Tracking: AI records "unsolved mysteries" found during ingestion—creating a backlog for future research.
- Audit Trails: Every entry includes a verifiable "AI Audit Block" with confidence scores and human-in-the-loop verification.
- Tier 1: Local Raw (Your private library).
- Tier 2: Structured Memory (NotebookLM, iMA, Feishu, Slack/Discord).
- Tier 3: Global Hunt (OpenCLI 140+ sites).
sync_growth() now not only suggests topics but also manages all "Observing" notes, ensuring your wiki grows organically with the latest global signals.
- Tier 1: Local Raw (Private Truth).
- Tier 2: Structured Relays (NotebookLM, iMA, Feishu, Slack, Discord).
- Tier 3: Live Capture (OpenCLI 140+ sites).
Run sync_growth() to let the OS analyze your intellectual trajectory and suggest the next frontier of your knowledge base.
# 1. Clone & Setup
git clone https://github.com/yuancafe/LLM-Wiki-OS.git ~/.agents/skills/llm-wiki-os
# 2. Re-index your existing vault
python3 scripts/reindex-db.py "~/Documents/MyVault/LLM-Wiki"This version is heavily inspired by:
- Andrej Karpathy: The OG methodology.
- Astro-Han: For the "Reflect" logic and SQLite indexing patterns.
- sdyckjq-lab: For the foundation of the Agent skill.
Tell your AI Agent (Claude Code, Gemini CLI, or OpenClaw):
# 1. Install via git or skill-manager
git clone https://github.com/yuancafe/LLM-Wiki-OS.git ~/.agents/skills/llm-wiki-os
# 2. Sync latest web adapters (140+ sites)
bash scripts/source-registry.sh refresh-opencli
# 3. Initialize your first topic
bash scripts/init-wiki.sh "~/Documents/MyVault/LLM-Wiki/Agent-Architecture" "Agent OS"Includes the Oriental Atlas (数字山水) interactive knowledge graph. A self-contained, offline-first HTML explorer with:
- Hierarchical Layout: Separating entities, sources, and annotations.
- Community Clustering: Visually grouping related concepts.
- Mobile-Responsive: Explore your brain on any device browser.
| Category | Platforms | Method |
|---|---|---|
| Enterprise | Feishu, WeCom, DingTalk, Slack, Discord | Native MCP Connectors |
| Research | NotebookLM, arXiv, Google Docs | Deep Memory Ingestion |
| Social/Web | X (Twitter), Reddit, Bilibili, XHS, Weibo | OpenCLI Dynamic Protocol |
| Personal | iMA, local PDF, Markdown, flomo | iFlow/Local Integration |
This project is an advanced fork and evolution of sdyckjq-lab/llm-wiki-skill.
We express our deepest gratitude to the original creators for their work on the Karpathy methodology. LLM Wiki OS expands on this foundation by adding the 3-Tier Sourcing Engine and the Autonomous Growth Protocol.
- Core Methodology: Andrej Karpathy
- Web Extraction: JimLiu's baoyu-skills & jackwener's OpenCLI
MIT