Skip to content

gauravsurtani/im-watching-you

BranchesTags

Repository files navigation

Lifelogger

Privacy-first personal "second brain" - A local-first life-logging system combining ActivityWatch for activity tracking, Whisper.cpp for audio transcription, LLM processing (local Ollama or free cloud via OpenRouter), and Syncthing for cross-device sync.

Flexible LLM options: Run entirely local for maximum privacy, use free cloud models via OpenRouter (no GPU needed), or hybrid mode that routes sensitive data locally while using cloud for classification.

Architecture

┌─────────────────────────────────────────────────────────────────────────┐
│                         CLIENT DEVICES                                   │
├──────────────────┬──────────────────┬──────────────────────────────────┤
│   Mac Desktop    │  Windows PC      │  Android (Pixel)                 │
│   ActivityWatch  │  ActivityWatch   │  ActivityWatch Android           │
│   Export Script  │  Export Script   │  Audio Capture + Whisper         │
└────────┬─────────┴────────┬─────────┴───────────────┬──────────────────┘
         │                  │                         │
         └──────────────────┼─────────────────────────┘
                            │ Syncthing (P2P encrypted)
                            ▼
┌─────────────────────────────────────────────────────────────────────────┐
│                      HOME SERVER                                         │
├─────────────────────────────────────────────────────────────────────────┤
│  Data Ingest → TimescaleDB → Ollama (LLM) → Notifications (ntfy)        │
└─────────────────────────────────────────────────────────────────────────┘

Features

  • Cross-platform activity tracking via ActivityWatch (Mac, Windows, Android)
  • Audio transcription with Whisper.cpp (on-device for Android)
  • AI-powered daily digests using LLMs (local Ollama or free OpenRouter cloud)
  • Hybrid LLM routing - Cloud for classification, local for sensitive transcripts
  • Push notifications via ntfy (self-hosted)
  • P2P encrypted sync with Syncthing
  • Time-series database with TimescaleDB for efficient queries
  • Web dashboard with search, stats, and event browser
  • Full-text search with fuzzy matching support

Quick Start

1. Server Setup

# Clone the repository
git clone https://github.com/yourusername/im-watching-you.git
cd im-watching-you

# Start Docker services
cd docker
cp .env.example .env
# Edit .env with your settings
docker compose up -d

# Wait for services to be ready
docker compose ps

2. Configure LLM (Choose One)

Option A: OpenRouter Cloud (Free, No GPU)

# Get your free API key at https://openrouter.ai/keys
# Add to .env:
LIFELOGGER_OPENROUTER_API_KEY=your_key_here
LIFELOGGER_LLM_PROVIDER=cloud

Option B: Local Ollama (Maximum Privacy, Requires GPU)

docker exec lifelogger-ollama ollama pull qwen2.5:7b
# In .env:
LIFELOGGER_LLM_PROVIDER=local

Option C: Hybrid (Best of Both)

# Set up both, then in .env:
LIFELOGGER_LLM_PROVIDER=hybrid
# Cloud handles classification, local handles sensitive transcripts

3. Install Python Package

# From repository root
pip install -e .

# Or with dev dependencies
pip install -e ".[dev]"

3. Configure Syncthing

  1. Open Syncthing UI at http://localhost:8384
  2. Add your client devices
  3. Configure shared folders:
    • ~/Syncthing/lifelogger/activity - ActivityWatch exports
    • ~/Syncthing/lifelogger/transcripts - Audio transcripts

4. Client Device Setup

Mac/Linux

# Install ActivityWatch
# Download from https://activitywatch.net

# Install Syncthing
# Download from https://syncthing.net

# Set up export script
cp scripts/clients/export_activitywatch.py ~/bin/
chmod +x ~/bin/export_activitywatch.py

# Add to crontab (hourly export)
crontab -e
# Add: 0 * * * * /usr/bin/python3 ~/bin/export_activitywatch.py

Windows

  1. Install ActivityWatch
  2. Install Syncthing
  3. Copy scripts/clients/export_activitywatch.ps1 to a local folder
  4. Create a Task Scheduler task to run it hourly

5. Set Up Server Cron Jobs

# Edit crontab
crontab -e

# Add:
# Hourly data ingestion
0 * * * * cd /path/to/im-watching-you && python -m lifelogger ingest

# Daily digest at 7 AM
0 7 * * * cd /path/to/im-watching-you && python -m lifelogger digest --send

Claude Desktop Integration (MCP)

Connect Lifelogger to Claude Desktop and ask questions about your life:

  • "What was I working on last Tuesday?"
  • "How much time did I spend coding this week?"
  • "When did I last discuss the budget?"
  • "Am I more productive in mornings or afternoons?"

Setup

  1. Start the MCP server (to test):

    lifelogger mcp-server

  2. Add to Claude Desktop config (~/.config/claude/claude_desktop_config.json on Linux/Mac, %APPDATA%\Claude\claude_desktop_config.json on Windows):

    {
  "mcpServers": {
    "lifelogger": {
      "command": "lifelogger-mcp",
      "env": {
        "LIFELOGGER_DB_HOST": "localhost",
        "LIFELOGGER_DB_PORT": "5432",
        "LIFELOGGER_DB_NAME": "lifelogger",
        "LIFELOGGER_DB_USER": "lifelogger",
        "LIFELOGGER_DB_PASSWORD": "your_password"
      }
    }
  }
}

  3. Restart Claude Desktop - you'll see "lifelogger" in the MCP tools list

Available Tools (What Claude Can Do)

Tool What It Does
search_activities Search through app usage, websites, window titles
get_daily_stats Get stats for any day (hours, top apps, productivity)
get_productivity_summary Productivity metrics over any time period
get_recent_activities What you've been doing recently
search_transcripts Search through conversation transcripts
get_app_usage Detailed usage stats for any app
compare_periods Compare this week vs last week, etc.
get_context_around_time Activities before/after a specific moment
get_category_breakdown Time by category (Work, Entertainment, etc.)

All data stays local - Claude queries your MCP server, which queries your local database.

CLI Commands

# Run data ingestion
lifelogger ingest

# Generate today's digest
lifelogger digest

# Generate and send digest
lifelogger digest --send

# View activity stats
lifelogger stats --date 2024-01-07

# List registered devices
lifelogger devices

# Send test notification
lifelogger notify "Test message" -t "Test Title"

# Start MCP server for Claude Desktop
lifelogger mcp-server

# Interactive setup guide
lifelogger setup

Configuration

Configuration is via environment variables (or .env file):

# Database
LIFELOGGER_DB_HOST=localhost
LIFELOGGER_DB_PORT=5432
LIFELOGGER_DB_NAME=lifelogger
LIFELOGGER_DB_USER=lifelogger
LIFELOGGER_DB_PASSWORD=your-password

# Ollama
LIFELOGGER_OLLAMA_HOST=localhost
LIFELOGGER_OLLAMA_PORT=11434
LIFELOGGER_OLLAMA_MODEL=qwen2.5:7b

# Syncthing paths
LIFELOGGER_SYNC_BASE_PATH=/home/user/Syncthing/lifelogger

# Notifications
LIFELOGGER_NTFY_SERVER=http://localhost:8080
LIFELOGGER_NTFY_TOPIC=lifelogger

# Data retention (days)
LIFELOGGER_AUDIO_RETENTION_DAYS=30
LIFELOGGER_TRANSCRIPT_RETENTION_DAYS=365

Project Structure

im-watching-you/
├── lifelogger/                 # Python package
│   ├── core/                   # Core modules
│   │   ├── config.py           # Settings management
│   │   ├── database.py         # TimescaleDB operations
│   │   ├── models.py           # Pydantic models
│   │   └── ingest.py           # Data ingestion service
│   ├── sources/                # Data source adapters
│   │   ├── activitywatch.py    # ActivityWatch importer
│   │   ├── transcripts.py      # Transcript importer
│   │   └── youtube.py          # YouTube history importer
│   ├── exporters/              # Output generators
│   │   ├── digest.py           # Daily digest with Ollama
│   │   └── notifications.py    # Notification delivery
│   └── cli.py                  # Command-line interface
├── scripts/
│   └── clients/                # Client-side scripts
│       ├── export_activitywatch.py    # Python export script
│       └── export_activitywatch.ps1   # Windows PowerShell
├── docker/
│   ├── docker-compose.yml      # Server services
│   └── .env.example            # Environment template
├── migrations/
│   └── 001_initial_schema.sql  # Database schema
└── android/                    # Android app (future)

Hardware Requirements

Server

  • CPU: Any modern x86-64
  • RAM: 16GB minimum (32GB recommended for larger models)
  • GPU: NVIDIA GPU with 8GB+ VRAM for Ollama (optional but recommended)
  • Storage: 100GB+ SSD

Model Recommendations

VRAM Recommended Model Performance
8GB Qwen2.5 7B ~40 tok/s
16GB Qwen2.5 14B ~30 tok/s

Extending

Adding New Data Sources

Create a new module in lifelogger/sources/:

# lifelogger/sources/newservice.py
from lifelogger.core.models import ActivityEvent, EventType

class NewServiceSource:
    async def import_data(self, path: Path) -> AsyncIterator[ActivityEvent]:
        # Parse your data format
        yield ActivityEvent(
            timestamp=...,
            device_id="newservice",
            event_type=EventType.CUSTOM,
            data={...}
        )

Custom Notification Channels

Add any Apprise-supported service:

LIFELOGGER_NOTIFICATION_CHANNELS='["ntfy://localhost/alerts", "tgram://bot_token/chat_id"]'

Privacy & Security

  • Local-first architecture - everything runs on your hardware
  • Hybrid LLM routing - sensitive data (transcripts, URLs) stays local even when using cloud classification
  • Syncthing uses TLS encryption for device-to-device sync
  • Database is localhost-only by default
  • Audio files auto-delete after configurable retention period
  • MCP integration is local - Claude queries your local server, data never leaves your machine
  • Configurable privacy settings - control what goes to cloud vs stays local

License

MIT

About

I am watching you.

Resources

Readme
Activity

Stars

1 star

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages