Architecture & Privacy

DuRT is engineered with a local-first, privacy-by-design philosophy.

---

Architecture Principles

┌─────────────────────────────────────────────────────────────┐
│                    DuRT Native macOS App                    │
├──────────────────────────────┬──────────────────────────────┤
│ Audio Ingestion Engine       │ Workflow Orchestrator        │
│ • ScreenCaptureKit (System)  │ • 6-Step Reusable Pipeline   │
│ • AVAudioEngine (Mic + AEC)  │ • State Machine & Subtitles  │
├──────────────────────────────┴──────────────────────────────┤
│                   Service Aggregator (BYOK)                 │
├──────────────────────────────┬──────────────────────────────┤
│ Local / Offline Engines      │ Cloud & Self-Hosted APIs     │
│ • Apple SFSpeechRecognizer   │ • OneASR Gateway             │
│ • Local Ollama LLM           │ • OpenAI / Anthropic / DeepSeek│
│ • Local GRDB SQLite Database │ • Alibaba / Tencent / Custom │
└──────────────────────────────┴──────────────────────────────┘

1. Local-First Processing

• In local mode, voice signals are decoded strictly on macOS on-device speech engines.
• No network requests are sent when using Apple Speech or Local Ollama.

2. Privacy & Data Ownership

• All task records, subtitle histories, and audio clips are persisted in a local SQLite database managed via GRDB on your Mac.
• When connecting to cloud providers (OpenAI, DeepSeek, OneASR, etc.), DuRT connects directly from your Mac to the respective endpoints without any intermediary proxy servers tracking your data.

3. Apple Sandbox Compliance

• The Mac App Store build runs inside Apple's strict App Sandbox, ensuring that DuRT only accesses system resources explicitly granted by you.