Personal_AI_Infrastructure
Personal_AI_Infrastructure copied to clipboard
feat: Add pai-knowledge bundle for knowledge capture and retrieval
Summary
Knowledge management bundle implementing the first layers of the Knowledge Transformation Stack: CAPTURE, STORE, and CULTIVATE (Sweep practice).
How It Works
📱 Device → Telegram → ingest daemon → Obsidian vault
↓
Claude ← ctx load ← ctx search
CLI Commands
| Command | Purpose |
|---|---|
ingest watch |
Background daemon for Telegram polling |
ctx search |
Semantic search across vault |
ctx load |
Load notes into Claude context |
ctx sweep |
Daily inbox triage |
What This Adds
Bundle (Bundles/pai-knowledge/)
| Component | Purpose |
|---|---|
install.ts |
Interactive installation wizard |
tools/ingest/ |
Telegram capture CLI |
tools/ctx/ |
Search and retrieval CLI |
shortcuts/ |
iOS/macOS shortcut templates |
templates/ |
.env, LaunchAgent plist, profiles |
Skill Definition (.claude/skills/Context/)
- Tag taxonomy (13 dimensions: PARA + Zettelkasten)
- Migration rules for tag normalization
Key Features
- Telegram as universal capture gateway - voice, photo, documents
- Whisper transcription for voice memos
- Vision AI for photo analysis
- Semantic search via embeddings (SQLite)
- Daily sweep practice for inbox triage
- Two-phase retrieval: search → load
Dependencies
-
kai-core(for skill routing) - Bun runtime
- OpenAI API (embeddings, Whisper, Vision)
- Telegram Bot API
Required Configuration
# Add to ~/.claude/.env
TELEGRAM_BOT_TOKEN=...
TELEGRAM_CHANNEL_ID=...
OBSIDIAN_VAULT_PATH=...
OPENAI_API_KEY=...
Test Plan
- [x] Unit tests pass (pai-tooling/ctx-test)
- [x] CLI tests pass (pai-tooling/ingest-test)
- [x] Fresh install tested on Kai baseline
- [ ] Reviewer tests with own config
Related
- Closes #147 (Knowledge Transformation Stack)
- Life OS Vision Discussion