pebbles-core 0.2.0

Autonomous content discovery and delivery engine — with primitives for principal-shaped agent capabilities

🪨 Pebbles

A smart content discovery engine that brings you interesting things from across the web — like a penguin bringing pebbles to its mate.

Pebbles monitors multiple sources (Hacker News, Reddit, RSS feeds, YouTube, Letterboxd), matches content against your interests using keyword or semantic matching, and delivers personalized recommendations via Telegram or email. Set daily limits, priority interests, and negative keywords to stay informed without getting overwhelmed.

---

Quick Start

Installation

pip install pebbles-core

Configuration

Create ~/.config/pebbles/config.yaml:

recipients:
  - name: "Jake"
    telegram_id: 123456789
    max_daily_pebbles: 3
    interests:
      - tags: ["surfing", "ocean"]
        keywords: ["surf", "wave", "ocean", "swell"]
        priority: 2
      - tags: ["climbing"]
        keywords: ["climbing", "bouldering", "crag"]
        negative_keywords: ["indoor"]
        priority: 1

sources:
  hackernews:
    enabled: true
  reddit:
    enabled: true
    subreddits: ["surfing", "climbing"]

delivery:
  telegram:
    bot_token: "your-bot-token-here"

matching:
  use_semantic_matching: false
  semantic_threshold: 0.35

Run

# One-time fetch
pebbles run

# Continuous loop (checks every 30 minutes)
pebbles run --loop

---

Configuration Reference

Recipients

Each recipient defines who receives pebbles and what they care about:

recipients:
  - name: "Jake"                    # Display name
    telegram_id: 123456789          # Optional: Telegram user ID
    email: "[e-mail]"       # Optional: Email address
    max_daily_pebbles: 3            # Daily limit (default: 5)
    interests:
      - tags: ["surfing"]           # Category tags
        keywords: ["surf", "wave"]  # Match keywords
        negative_keywords: ["pool"] # Exclude keywords
        priority: 2                 # 1=normal, 2=high, 3=must-have

At least one of telegram_id or email must be provided per recipient.

Sources

Hacker News

sources:
  hackernews:
    enabled: true

Reddit

sources:
  reddit:
    enabled: true
    subreddits:
      - "surfing"
      - "climbing"
      - "technology"

RSS Feeds

sources:
  rss:
    enabled: true
    feeds:
      - "https://example.com/feed.xml"
      - "https://blog.example.com/rss"

YouTube

sources:
  youtube:
    enabled: true
    api_key: "your-youtube-api-key"
    queries:
      - "surf documentary"
      - "climbing training"

Letterboxd

sources:
  letterboxd:
    enabled: true
    usernames:
      - "username1"
      - "username2"

Delivery

Telegram

delivery:
  telegram:
    bot_token: "your-telegram-bot-token"

Email

delivery:
  email:
    smtp_host: "smtp.gmail.com"
    smtp_port: 587
    smtp_user: "[e-mail]"
    smtp_password: "your-app-password"
    smtp_from: "Pebbles <[e-mail]>"  # Optional

Matching

matching:
  use_semantic_matching: false    # Enable AI-powered matching
  semantic_threshold: 0.35        # Similarity threshold (0.0-1.0)

Semantic matching uses sentence-transformers (all-MiniLM-L6-v2) for contextual understanding. Falls back to keyword matching if unavailable.

---

Deployment

Docker

docker run -d \
  --name pebbles \
  -v $(pwd)/config:/root/.config/pebbles \
  -v $(pwd)/data:/root/.local/share/pebbles \
  --env-file .env \
  pebbles-core:latest

Docker Compose

services:
  pebbles:
    image: pebbles-core:latest
    volumes:
      - ./config:/root/.config/pebbles
      - ./data:/root/.local/share/pebbles
    env_file: .env
    restart: unless-stopped

Create .env:

TELEGRAM_BOT_TOKEN=your-token
SMTP_PASSWORD=your-password

---

Roadmap

Pebbles was built in phases:

• Phase 1: Core engine + HackerNews source
• Phase 2: Production hardening (logging, retries, error handling)
• Phase 3: Multi-source expansion (Reddit, RSS, YouTube, Letterboxd)
• Phase 4: Smart matching (semantic understanding, priority scoring, frequency caps)
• Phase 5: Polish + ship (email delivery, Docker deployment, docs)

Future ideas: Mastodon source, Discord delivery, web dashboard, collaborative filtering.

---

Contributing

See CONTRIBUTING.md for how to add sources, delivery adapters, or improve matching.

---

License

MIT