In progress2026
Local Edge-AI SDK
An offline-first SDK for running small language models on-device, from Python, C and Flutter.
v0.3.1 · verified on desktop and in CI · not yet run on a physical phone
- year
- 2026
- role
- Builder
- status
- In progress
01what it is
What it does.
A Python pipeline that compresses Hugging Face models to GGUF with llama.cpp, a native C runtime around llama.cpp with token streaming and cancellation, and Dart/Flutter bindings for Android and iOS. No API key, no cloud inference, no telemetry.
Verified so far: Phi-3 Mini Q4_K_M generated and run on Windows, the real runtime on Intel macOS, Android packaging and an emulator run, and the iOS simulator. Not yet: inference on a physical phone, and there is no Neural Engine or NPU support.
02the product
The live product.
The project site. A status page, not a demo: inference runs in the SDK on a device, and the page says so.
03engineering
How it is built.
Model conversion, a native C ABI, Python ctypes and Dart FFI bindings with contract tests, a Flutter plugin with Android and iOS packaging, and model-free CI on Linux, macOS and Windows.
- Python
- C/C++
- Dart
- Flutter
- llama.cpp
04where it stands
Still being built.
In development; the project site is live and the SDK is at version 0.3.1.
more projects
- ResumeForgeIn progressUpload a resume, let an AI agent structure it, adjust the layout in plain English, and export a real PDF — without inventing a single fact.
- CalcBlast GamesShippedA math practice app for iPhone, from early math through AP Calculus — score 80% on a quiz to unlock a round of Block Blast.
- OrblyShippedAsk anything about your Instagram circle — a chat over your own Instagram data export, parsed locally in the browser.