A sophisticated personal AI assistant designed to operate entirely on local hardware. No cloud dependencies - your data stays yours.
Overview
Friday is my personal AI assistant that runs on my homelab. It leverages the Hermes-4-14B model via vLLM and provides proactive awareness capabilities through multi-channel communication and extensive tool integration.
Repository: github.com/arturgoms/friday
Features
Communication & Interaction
- Telegram bot interface with full conversation history
- Interactive CLI with 79 integrated tools across 13 modules
- Multi-channel session management
- Channel-agnostic session tracking with persistent SQLite storage
Intelligence Capabilities
- Local LLM inference without cloud dependencies
- Context-aware responses using conversation history
- Tool chaining for complex multi-step queries
- Auto-snapshot system for data collection
Proactive Awareness
- Health monitoring integration (Garmin/InfluxDB)
- Calendar-based awareness via CalDAV
- Portfolio tracking via DLP API
- Weather and environmental sensing
- Threshold-based notifications with quiet hour respect
Specialized Systems
- Daily journal threads with voice transcription (Whisper)
- Scheduled reports (briefings, journal notes)
- Markdown formatting for Telegram compatibility
- Obsidian vault integration
Tech Stack
- Python 3.12+ with Pydantic-AI framework
- vLLM for model serving
- SQLAlchemy with SQLite for persistence
- Typer for CLI
- CalDAV for calendar sync
- InfluxDB for health data
- Whisper for voice transcription
- systemd for service management
Why Local?
Running an AI assistant locally means:
- Complete privacy - no data leaves my network
- No API costs or rate limits
- Full control over the model and its behavior
- Works offline (except for external integrations)
The trade-off is hardware requirements (RTX 3090 in my case), but for a personal assistant that knows my schedule, health data, and daily routines, privacy is worth it.