# Mobile Deployment Support

WorldDisasterLM mobile support is designed via optimized model artifacts and API-first architecture.

## Strategy

1. Export compact inference artifacts:
- GGUF for on-device CPU inference wrappers
- ONNX for cross-platform runtime support

2. Build mobile clients (Android/iOS) that consume API endpoints:
- `/v1/chat`
- `/v1/risk/score`
- `/v1/incidents/classify`

3. Optionally embed quantized local models:
- Android: ONNX Runtime Mobile / llama.cpp bridges
- iOS: CoreML conversion pipeline or ONNX Runtime

## Recommended Runtime Profiles

- Edge/Offline mode: GGUF 4-bit quantized variants
- Connected mode: FastAPI cloud inference with local fallback

## Security

- Enforce TLS and token-based auth in production
- Cache only non-sensitive incident summaries
- Log consent and audit metadata for high-risk guidance usage