An AI-powered knowledge management system built on Obsidian and Claude Code.
Turn scattered research, articles, and ideas into an interconnected knowledge graph that evolves with you. Drop files in, get structured insights out.
You read 50 articles a week. You bookmark them. You forget them. Six months later, you can't find the one insight that would connect two ideas you're working on.
Most note-taking systems are filing cabinets — they store information but don't think with you.
This system does three things traditional note apps can't:
-
Ingests and analyzes — Drop a PDF, screenshot, or article into the inbox. The AI reads it, extracts insights, generates WikiLinks to your existing knowledge, and files it into the right category.
-
Self-organizes — Run
sort-gardenand the folder structure reorganizes itself using MECE + Pyramid principles. Categories split, merge, and rebalance as your knowledge grows. No manual filing. -
Thinks with you — Co-create investment theses, stress-test arguments with adversarial sparring, discover cross-domain connections you'd never see manually.
📱 LINE / Mobile 🖥️ Obsidian + Claude Code
│ │
│ Send article URL, │ @Agent run ingest
│ forward PDF, │ @Agent session [topic]
│ share social post │ @Agent sort-garden
│ │ @Agent brief [topic]
▼ ▼
┌──────────┐ ┌──────────────┐ ┌──────────────┐
│ LINE │───▶│ Make.com │───▶│ 00_Inbox/ │
│ Chat │ │ Automation │ │ (raw files) │
└──────────┘ └──────────────┘ └──────────────┘
│ │
Routes by type: AI Agent processes:
• PDF → Drive → Inbox • Extract & analyze
• URL → Gemini → .md • Categorize via taxonomy
• Image → Drive → Inbox • Generate WikiLinks
• File into Garden
│
▼
┌──────────────┐
│ 10_Garden/ │
│ (knowledge │
│ graph) │
└──────────────┘
The fastest path from "I just saw something interesting" to structured knowledge:
LINE → Make.com → Obsidian Inbox
Send anything to a LINE chat — a PDF attachment, a Substack post, a Facebook article, any URL — and a Make.com automation scenario routes it into 00_Inbox/:
| What you send | What happens |
|---|---|
| PDF attachment | Downloaded via LINE API → uploaded to Google Drive → synced to 00_Inbox/ |
| Article URL | Fetched via HTTP → processed by Gemini AI into markdown → saved as .md in 00_Inbox/ |
| Social media post | URL fetched → content extracted → converted to .md in 00_Inbox/ |
Once files land in the Inbox, run @Agent run ingest in Claude Code to process them into structured Garden notes.
Get the Make scenario: Import this blueprint into your Make.com account. You'll need to connect your own LINE, Google Drive, and Gemini API credentials.
Notes are organized into a self-evolving taxonomy:
10_Garden/
├── AI_Hardware/ ← Compute, Memory, Networking, Packaging, Power, Testing
├── AI_Software/ ← Applications, Software Trends, Products
├── Industry_Verticals/ ← Automotive, Clean Energy, Space, Robotics, Materials
├── Investment_Thinking/ ← Mindset, Market Behavior, Frameworks, Trading, Valuation
├── Macro_Markets/ ← Geopolitics, Portfolio Reviews
├── Journal/ ← Personal reflections
└── Parenting/ ← Education, family
This structure is not fixed. Run @Agent sort-garden and the system:
- Scans all notes for semantic clustering
- Proposes folder restructuring based on MECE/Pyramid principles
- Moves files, updates WikiLinks, regenerates indices
- Rewrites
taxonomy.md— which becomes the basis for the next cycle
Every folder has an auto-generated index optimized for both human browsing and AI agent lookup:
- Tier 1 (Parent folders): Lightweight routing index (<3KB) — "which subfolder should I look in?"
- Tier 2 (Leaf folders): Compact table with one row per note — title, key thesis, key entities
Regenerate with: python .agent/generate_indices.py
The Lab is where ideas develop before they become permanent knowledge:
| Command | What it does |
|---|---|
@Agent session [topic] |
AI working session — co-create and spar fluidly |
@Agent spar [topic] |
Adversarial mode — steelman then attack the weakest point |
@Agent learn-style |
Analyze your writing patterns → update style guide |
@Agent draft-post [[Note]] |
Turn a note into a social media draft |
Sessions are logged inside the Lab entry itself (## 工作紀錄), keeping the thinking co-located with the artifact. When an idea matures, @Agent lab-promote moves it to the Garden.
| Command | What it does |
|---|---|
@Agent run ingest |
Process Inbox → structured Garden notes with WikiLinks |
@Agent sort-garden |
Reorganize folder structure using MECE/Pyramid principles |
@Agent process-files |
Convert Office files (.pptx/.docx/.pdf) → markdown |
| Command | What it does |
|---|---|
@Agent recall [topic] |
Quick inventory — what do I know about X? |
@Agent brief [topic] |
Synthesized briefing from all relevant notes |
@Agent gaps |
Knowledge gap analysis + 30-day learning agenda |
@Agent challenge [[Note]] |
Stress-test a thesis with pre-mortem analysis |
@Agent connect [[A]] [[B]] |
Discover non-obvious cross-domain connections |
| Command | What it does |
|---|---|
@Agent promote |
Move durable insights from Lab → Garden |
@Agent garden-review |
Surface stale notes for update or archival |
@Agent memo |
Structure an investment memo from Garden research |
- Obsidian with these plugins: Dataview, Templater, Strange New Worlds
- Claude Code (CLI or VS Code extension)
- Python 3.10+ (for index generation scripts)
- Optional: Make.com account + LINE for mobile capture
-
Clone this repo
git clone https://github.com/Timeverse/My-Brain-System.git cd My-Brain-System -
Open in Obsidian — point Obsidian at the cloned folder as a vault
-
Customize your taxonomy — edit
.agent/taxonomy.mdto match your knowledge domains. The default categories are investment/tech-focused; replace them with whatever you study. -
Drop files into
00_Inbox/and run:@Agent run ingest -
Optional: Set up mobile capture — import the Make.com scenario and connect your LINE + Google Drive
- MECE + Pyramid: Sibling folders don't overlap. Max 3 levels deep. 5-20 notes per leaf.
- Knowledge as Graph: WikiLinks between notes are a first-class success metric. Orphaned notes are incomplete.
- "So What?" Test: Every insight must answer why it matters. No summaries without conclusions.
- Evolution over Perfection: Each
sort-gardencycle makes incremental improvements. The structure is never "done." - Never Delete: Refactor and preserve meaning. Archive, don't destroy.
Every Garden note follows this structure (Traditional Chinese headers, English technical terms):
---
title: [Descriptive Title]
created: 2026-04-03
updated: 2026-04-03
review_by: 2026-10-03
tags: [Category, Subcategory, Keywords]
source_type: pdf_report
source_asset: original_filename.pdf
---
# Title
## 摘要 (Summary)
## 核心發現 (Core Findings)
## 相關概念連結 (Related Concepts) ← WikiLinks to other notes
## 關鍵洞察 (Key Insights) ← Must pass "So What?" test
## 第二層思考 (Second-Level Thinking) ← What are others NOT seeing?
## 第一性原理分析 (First Principles)
## 原始文件 (Original Document)Language note: The default is Traditional Chinese (繁體中文). To use a different language, update the note template in
tasks/ingest.mdand the style guide in.agent/writing_style.md.
MIT — use it, fork it, make it yours.
Built with Claude Code. The system scaffolding is public; personal knowledge notes are excluded via .gitignore.