LLM Configuration

OpenCyvis is model-agnostic. Bring your own AI account, connect a private service, or run a local model.

In-App Setup

Go to Settings → LLM Provider and configure:

Field	Description
Provider type	openai (OpenAI-compatible), anthropic, or ollama
Base URL	API endpoint (e.g. https://api.openai.com/v1)
API Key	Your key (not needed for Ollama)
Model	Model name (e.g. gpt-4o, claude-opus-4-20250514)

Deeplink Configuration

You can also configure via ADB deeplink:

# Local Ollama (fully private)
adb shell am start -a android.intent.action.VIEW \
  -d "opencyvis://config?provider=ollama&base_url=http://localhost:11434&model=gemma4:26b"

# Cloud API (OpenAI-compatible)
adb shell am start -a android.intent.action.VIEW \
  -d "opencyvis://config?provider=openai&base_url=https://api.example.com/v1&api_key=YOUR_KEY&model=qwen-vl-max"

Share Config via QR Code

Need to set up OpenCyvis on another phone? Export your current LLM configuration as a QR code, then scan it on the target device to import everything — provider type, base URL, API key, and model name — in one step.

Go to Settings → LLM Provider → Share to generate a QR code. On the receiving device, tap Scan QR Code in the setup wizard.

⚠️ Warning QR codes encode your API key. Only share them with people you trust, and over private channels.

Provider Profiles

Save multiple configurations (e.g. "Work Qwen", "Home Gemma", "Claude") and switch with one tap. See Provider Profiles.

Cloud vs Local

	Cloud Models	Local Models (Ollama)
Privacy	Data sent to API	Nothing leaves device
Speed	2-8s per step	7-10s per step (with GPU)
Quality	Highest reasoning	Good, improving fast
Cost	Pay per token	Free (hardware cost)
Offline	❌	✅

💡 Tip **Recommended cloud:** Qwen 3.5 Plus — stable, fast, affordable. **Recommended local:** Gemma 4 26B-A4B — best balance of speed, quality, and VRAM (17 GB).

Supported Models

See the full benchmark table at Supported Models.