Local Processing
Canoryn is designed with privacy-first architecture. All AI processing happens on your Mac.
How It Works
┌─────────────────────────────────────────┐
│ YOUR MAC │
├─────────────────────────────────────────┤
│ Voice → Whisper (local) → Text │
│ Text → Ollama (local) → Response │
│ Response → TTS (local) → Audio │
└─────────────────────────────────────────┘
↑
│ Everything stays here
│
─────┴───── No data leaves ─────────Local AI with Ollama
Canoryn uses Ollama for local LLM inference:
- Models: Llama 3.2, Qwen 2.5, Mistral, etc.
- Speed: Fast inference on Apple Silicon
- Privacy: Zero network calls for AI
Recommended Models
| Model | Size | Best For |
|---|---|---|
llama3.2:3b | 2GB | Fast responses |
qwen2.5:7b | 4GB | Balanced |
llama3.2:8b | 5GB | Complex tasks |
What Stays Local
| Data | Storage |
|---|---|
| Conversations | SQLite (encrypted) |
| Memory | SQLite (encrypted) |
| Blueprints | JSON files |
| Voice recordings | Never stored |
| Screenshots | Processed, not saved |
Optional Cloud Providers
If you prefer, you can use cloud AI:
| Provider | Benefit | Trade-off |
|---|---|---|
| OpenAI | GPT-4 quality | Data leaves device |
| Anthropic | Claude reasoning | Data leaves device |
| Groq | Very fast | Data leaves device |
TIP
All cloud providers are optional. Canoryn works fully offline with Ollama.
Verification
To verify local processing:
- Disconnect from internet
- Use Canoryn normally
- Everything should work (with Ollama)