A self-filing encyclopedia of AI/ML terminology. Push terms to queue.txt, and Claude automatically researches, organizes, and connects them. Explore the knowledge graph via interactive visualization, search, and navigation.
┌──────────────────────────────────────────────────┐
│ GitHub Repository │
├──────────────────────────────────────────────────┤
│ │
│ queue.txt ─────────────────┐ │
│ (user edits & pushes) │ │
│ ↓ │
│ ┌─────────────────┐ │
│ │ GitHub Action │ │
│ │ (research.yml) │ │
│ └─────────────────┘ │
│ │ │
│ ↓ │
│ ┌─────────────────┐ │
│ │ Claude API │ │
│ │ (researches) │ │
│ └─────────────────┘ │
│ │ │
│ ↓ │
│ entries.json ◄─────────────┐ │
│ (auto-committed) │ │
│ │ │ │
│ ↓ │ │
│ docs/index.html ◄──────────────────────────┐ │ │
│ (GitHub Pages served) │ │ │
│ ├─ app.jsx │ │ │
│ ├─ graph.jsx │ │ │
│ ├─ research.jsx │ │ │
│ ├─ tweaks-panel.jsx │ │ │
│ └─ reads entries.json ─────────────────────┘ │ │
│ │ │
└──────────────────────────────────────────────────┘
↓
GitHub Pages (https://yourusername.github.io/repo)
-
Edit
queue.txt— Add AI/ML terms you want researched:Transformer Attention Mechanism Retrieval Augmented Generation -
Push to GitHub — Commit and push your changes
-
GitHub Action Triggers — Automatically detects new terms in queue.txt
-
Claude Researches — Each term is researched and structured:
- Term: canonical name
- Category: auto-classified (Model Architectures, Training Methods, etc.)
- TL;DR: one-sentence definition
- Significance: why it matters
- Related: 3-5 connected concepts
-
Results Committed —
entries.jsonis auto-updated and pushed back -
Explore on the Web — Visit GitHub Pages to browse:
- Overview: force-directed graph of all terms and categories
- Categories: card grid view
- Terms: full entry with connections
- Search: find terms or concepts
- Customize: theme, density, font size via tweaks panel
-
Web App — Static React app (no backend)
- Reads entries from
entries.jsonin repo - Renders search, graph, navigation, tweaks
- No input fields (read-only)
- Hosted: GitHub Pages (
docs/folder)
- Reads entries from
-
Research Engine — GitHub Action
- Runs on push to queue.txt
- Calls Claude API with your subscription key
- Parses research results into structured JSON
- Auto-commits
entries.jsonback to repo
-
Storage — GitHub repository
queue.txt— user maintainsentries.json— auto-generated by action- All version-controlled and auditable
git clone <your-repo>
cd ai-wikiCopy the files into your repo:
docs/index.html,docs/*.jsx.github/workflows/research.ymlqueue.txt(start empty or with sample terms)
Get your Claude API key from console.anthropic.com.
In your GitHub repo:
- Settings → Secrets and variables → Actions
- New repository secret
- Name:
CLAUDE_API_KEY - Paste your API key
- Settings → Pages
- Source: Deploy from a branch
- Branch:
main, Folder:/docs - Save
Your site will be live at: https://yourusername.github.io/repo-name
-
Add a term to
queue.txt:Neural Network -
Commit and push
-
Go to Actions tab — watch the workflow run
-
Check
entries.jsonto see the result -
Visit your GitHub Pages site and search for the term
├── .github/
│ └── workflows/
│ └── research.yml # GitHub Action workflow
├── docs/ # GitHub Pages (static site)
│ ├── index.html # Main app
│ ├── app.jsx # Sidebar, views, shell
│ ├── graph.jsx # Force-directed graph
│ ├── research.jsx # Queue parsing (read-only now)
│ ├── tweaks-panel.jsx # Settings UI
│ └── entries.json # Generated by workflow
├── queue.txt # User edits this
├── README.md # This file
└── .gitignore
Edit queue.txt:
# Comments start with #
Transformer
Attention Mechanism
RLHF
# Duplicates are ignored
Transformer # This won't be researched again
Push to GitHub. The Action will:
- Parse new terms
- Research each via Claude
- Skip already-researched ones
- Commit results to
entries.json
- Search: Use the search box in the sidebar
- Navigate: Click categories in the sidebar
- Graph: Drag nodes, hover for context, click to explore
- Customize: Open tweaks panel (bottom-right corner)
- Theme: Aurora, Signal, Ember
- Accent: Violet, Cyan, Amber, Emerald
- Density: Compact, Regular, Comfy
- Font size: 85–120%
Each entry has:
- Category tag — click to see all in category
- TL;DR — crisp definition
- Why it matters — significance and use cases
- Related — linked concepts (click to navigate, or add to queue)
- Metadata — filed date, source
Edit .github/workflows/research.yml — the RESEARCH_PROMPT section controls how Claude researches. Common tweaks:
- Change category list or description
- Adjust the JSON structure
- Set word limits
Edit docs/index.html — CSS tokens at the top:
:root {
--bg: oklch(0.165 0.012 265); /* Background color */
--accent: #a78bff; /* Highlight color */
--radius: 14px; /* Border radius */
--pad: 56px; /* Padding */
}- Check API key is set in Secrets
- Check workflow syntax in
.github/workflows/research.yml - View Actions tab for error logs
- Check
entries.jsonwas created - Verify GitHub Pages is enabled
- Check browser cache (hard refresh: Ctrl+Shift+R)
- Make sure you're editing the root-level
queue.txt, not a copy - The Action watches for changes to
queue.txt— commit and push
- Claude may misclassify categories — edit
entries.jsonmanually to fix - Results are just JSON — you can edit entries directly
- GitHub: Free (repo + Actions + Pages)
- Claude API: Charged per token (your subscription API key)
- Research is usually cheap (~2-5 tokens per entry typically)
- Estimate: ~$0.01 per term
python3 -m http.server 8080 -d docs
# or use VS Code Live ServerOpen http://localhost:8080
- Edit files in
docs/ - Refresh browser
- Commit changes
Changes to HTML/CSS are instant. Changes to entries.json require a GitHub Action run (push to queue.txt) or manual edit.
MIT
- Check GitHub Issues
- See workflow logs in Actions tab for debugging
- Edit
queue.txtto retry research on a term