Orvyn

What it is

Orvyn is a personal AI assistant and thinking partner built for my own daily workflow rather than as a commercial product. It handles agenda tracking, reminders, morning and evening briefings, and remote commands through Telegram, wrapped in a minimal dark web UI with a command palette and a thin Electron shell.

How it works

The backend follows a layered Domain-Driven Design architecture separated into domain, application, infrastructure, and interfaces layers. Long-term associative memory is stored in a SQLite relation graph with automated entity extraction and a nightly “gardener” job that consolidates and prunes connections. Voice interaction runs through a full-duplex streaming pipeline using local speech-to-text (faster-whisper) and local text-to-speech. The system runs on a single RTX 3070 (8 GB VRAM) machine using a hybrid local and cloud model routing setup.

Where it’s going

The long-term goal for Orvyn is to bridge software and physical hardware—extending the same memory and task architecture so the assistant can observe and act on real-world devices and workshop tools alongside digital tasks.