On-device LLM and hybrid routing
Notes on DeepSleep
Live
Overview
Notes on splitting work between on-device inference and cloud models in a local-first sleep and mood app.
Released2025-09-27
StatusLive
Platforms—
Notes
Hybrid design
Several providers are bound to one routing path, with automatic fallback when one fails.
On device
Inference through llama.xcframework reduces network dependency. Sleep routines, mood logs, and presets stay in CoreData.
Usage rules
Tier-based limits and security gates keep usage predictable.