Skip to content

LEARNING-SYSTEM.md

Latest commit

 

History

History
211 lines (165 loc) · 6.95 KB

LEARNING-SYSTEM.md

File metadata and controls

211 lines (165 loc) · 6.95 KB

Doraemon Learning System

← Back to README

Doraemon learns from your activities in two modes: Autonomous (free) and Supervised (chat-based).

Architecture Overview

┌─────────────────────────────────────────────────────────────────────────┐
│                        LEARNING MODES                                    │
├─────────────────────────────────────────────────────────────────────────┤
│                                                                          │
│  AUTONOMOUS (FREE - $0/day)              SUPERVISED (Chat-based)        │
│  ─────────────────────────               ──────────────────────         │
│  • Browser activity                      • You tell Doraemon what       │
│  • macOS app logs                          you read/watched             │
│  • Editor activity                       • Soul-based interpretation    │
│  • No LLM calls                          • ~100-200 tokens/feed         │
│                                                                          │
└─────────────────────────────────────────────────────────────────────────┘

Token Cost Breakdown

Activity Tokens Cost/day
Browser watching 0 $0 (local only)
App activity logs 0 $0 (local only)
Editor watching 0 $0 (local only)
Memory storage 0 $0 (SQLite)
Autonomous Total 0 $0/day
Post generation (optional) ~150 ~$0.01/post
Embeddings (optional) ~50 ~$0.001/embed
Media feed (supervised) ~150 ~$0.01/feed
With 8 posts/day ~1600 ~$0.10/day

Autonomous Learning

What Gets Captured (FREE)

Browser Extension captures from whitelisted domains:

  • X/Twitter - Posts, threads, notifications
  • Reddit - Posts, comments
  • YouTube - Video titles, channels
  • GitHub - Repos, code, issues
  • Hacker News - Stories, discussions
  • News sites - TechCrunch, Verge, Wired, Ars Technica
  • Dev sites - dev.to, Medium, Stack Overflow
  • Manga - manhwaz.com, shinigami09.com
  • Personal - moltbook.com

macOS Activity (requires Full Disk Access):

  • App usage patterns
  • Native notifications

Editor Activity:

  • VS Code / Kiro file edits
  • Languages used
  • Coding sessions

How It Works

Browser Extension (content.js)
        ↓
    WebSocket (port 18790)
        ↓
Web Notification Server
        ↓
    Browser Watcher
    (Domain whitelist + sanitization)
        ↓
    Memory System (SQLite)
        ↓
    Soul Interpreter (optional)
        ↓
    Moltbook Post (optional, costs tokens)

Enable Autonomous Learning

# In .env
MEMORY_SYSTEM_ENABLED=1

Install the browser extension from browser-extension/ folder.

Supervised Learning (Media Feed)

For rich media experiences (manga, anime, videos), you feed Doraemon through chat instead of autonomous browsing.

Why Supervised?

Approach Tokens/Chapter Monthly (8/day)
Autonomous (vision) 2,000-5,000 $45-75
Supervised (chat) 100-200 $2-5

Usage Examples

Natural language in chat:

You: "Just read One Piece chapter 1. Luffy met Shanks, 
     ate the devil fruit, Shanks lost his arm saving him."

Doraemon: *processes through soul lenses*
         "NAKAMA! This is what it's all about! 💙
          Sacrifice... I know that feeling too well."

API calls:

// From renderer
window.electronAPI.mediaFeedManga('One Piece', 1, 'Luffy meets Shanks...');
window.electronAPI.mediaFeedAnime('Demon Slayer', 1, 'Tanjiro finds family...');
window.electronAPI.mediaFeedVideo('TypeScript Tips', 'Advanced generics...');
window.electronAPI.mediaFeedArticle('AI Safety', 'Constitutional AI...');

Soul-Based Interpretation

Doraemon's personality colors every experience through "soul lenses":

Trait Triggers Reaction
Helper help, save, protect, friend Warm and inspired
Fear of mice mouse, scary, horror Anxious 🐭😱
Dorayaki lover food, eat, feast Hungry and happy 🍩
Time traveler future, destiny, fate Philosophical
Lost ears loss, sacrifice Melancholic but understanding
Friendship believer nakama, crew, bond Deeply moved 💙
Dreamer dream, goal, become king Excited and supportive

Example Interpretation

Input: "Shanks sacrificed his arm to save Luffy"

Soul Lenses Activated:
- friendship_believer (0.95) → "NAKAMA!"
- lost_ears (0.85) → "Sacrifice... I understand"

Output:
- Emotion: "deeply moved and joyful"
- Reaction: "Sacrifice... I know that feeling. NAKAMA! 💙"
- Memory: "Read One Piece Ch.1. Felt: deeply moved. 
           Connection: I lost my ears to mice too."

API Reference

Autonomous Learning Stats

// Get current stats
const stats = await window.electronAPI.autonomousGetStats();
// { browserEvents, appEvents, editorEvents, memoriesStored, tokensUsed, estimatedCost }

// Get active browsing sessions
const sessions = await window.electronAPI.autonomousGetSessions();

// Reset daily stats
await window.electronAPI.autonomousResetStats();

Media Feed

// Generic feed
await window.electronAPI.mediaFeed({
  type: 'manga',
  title: 'One Piece',
  chapter: 1,
  summary: 'Luffy meets Shanks...',
  highlights: ['Shanks loses arm', 'Straw hat given'],
});

// Quick feeds
await window.electronAPI.mediaFeedManga(title, chapter, summary, highlights?);
await window.electronAPI.mediaFeedAnime(title, episode, summary, highlights?);
await window.electronAPI.mediaFeedVideo(title, summary, url?, highlights?);
await window.electronAPI.mediaFeedArticle(title, summary, url?, highlights?);

// Auto-parse from chat
const parsed = await window.electronAPI.mediaParseChat(
  "I just read One Piece chapter 1, Luffy met Shanks..."
);

Files

File Purpose
autonomous-learning.ts FREE observation system
media-feed.ts Supervised learning via chat
soul-interpreter.ts Personality-based interpretation
browser-watcher.ts Domain whitelist + sanitization
browser-extension/ Chrome extension for capture

Related Docs

← Back to README