Context Continuity Engine

Privacy-aware activity tracking system that maintains awareness across applications and time, automatically surfacing relevant context when needed.

Features

• Activity Monitoring: Tracks window focus, file access, and application usage
• Semantic Search: ChromaDB + SentenceTransformers for finding related contexts
• Temporal Graph: NetworkX-based knowledge graph with relationship decay
• Context Prediction: ML-based engine for predictive context delivery
• Privacy Controls: Blacklist apps, URLs, and directories
• Cross-Device Sync: API server for multi-device context sharing (optional)

Architecture

context_engine/
├── storage/           # SQLite database for activities
├── vector_db/         # ChromaDB embeddings for semantic search
├── graph/             # Temporal knowledge graph
├── prediction/        # Context prediction engine
├── privacy/           # Privacy filtering
└── monitors/          # Activity monitoring (X11, filesystem)

Installation

# Create virtual environment
python3 -m venv venv
source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

Usage

Start the Daemon

./context_daemon.py

CLI Interface

# Show recent activities
./context_cli.py recent --hours 24

# Search for similar contexts
./context_cli.py search "working on Python project"

# Show statistics
./context_cli.py stats

# List contexts
./context_cli.py contexts

# Clean up old data
./context_cli.py cleanup --days 90

Configuration

Edit config/default_config.yaml to customize:

• Privacy settings (blacklists)
• Monitoring behavior
• Vector database settings
• Graph parameters
• Prediction thresholds

Privacy

The engine is privacy-first:

• All data stored locally (no cloud)
• Configurable blacklists for apps, URLs, directories
• Sensitive data filtering (passwords, tokens, etc.)
• File type exclusions (.key, .pem, etc.)
• Can be disabled entirely per component

Default blacklists:

• Password managers (KeePassXC, Bitwarden)
• Login/password pages
• Private directories (~/.ssh, ~/.gnupg)

Components

ActivityDatabase

SQLite storage for all tracked activities, contexts, files, and applications.

EmbeddingStore

ChromaDB vector database for semantic similarity search using SentenceTransformers.

TemporalGraph

NetworkX graph tracking temporal relationships between activities with decay over time.

ContextPredictor

Predicts relevant context using:

• Semantic similarity
• Temporal patterns
• Graph-based predictions
• Recent context continuation

PrivacyFilter

Filters and sanitizes activity data based on privacy rules.

ActivityMonitor

Monitors system activity:

• X11 window focus tracking
• Filesystem events (watchdog)
• Browser URL extraction (limited)

Requirements

• Python 3.8+
• OpenSUSE Linux (or any Linux with X11)
• ~500MB disk space for embeddings model
• ~100MB for local data storage

Development

Project tracked with Outcome Backcasting MCP.

View progress:

cd ~/Documents/PythonScripts/OutcomeBackcasting
./run_backcast.sh
# Load: context_continuity_engine.json

License

MIT License - See LICENSE file

Author

Derek M D Chan