Back to directory

Melange Applications

A collection of open-source example applications showing how to build on-device AI apps with full NPU acceleration using ZETIC's Melange model-conversion platform, covering offline translation, voice and vision tasks.

Melange Applications

This repository collects source code for on-device AI applications built with Melange, an NPU-native platform from ZETIC that converts an AI model into a mobile app through automated conversion rather than manual hardware-specific tuning. The example apps cover offline translation, a private AI notepad, conversational AI, time-series forecasting, a multimodal wellbeing screener, camera-based heart-rate detection, voice biomarker analysis, skin image classification, and an AI keyboard, each running fully on-device with no cloud calls.

The project states the apps are tested across many devices and integrate with Android and iOS through a shared API. The repository has moved to a renamed successor repository and is kept here for reference; new issues and pull requests are directed there instead.

It is written mainly in Swift and released under the Apache-2.0 license.

More in Mobile

See all 90
  • Dart and Flutter bindings for the Zvec in-process vector engine, letting mobile and desktop apps run on-device semantic search, retrieval-augmented generation and similarity queries with no separate server.

  • Mobile Claw

    github.com

    A Capacitor plugin that embeds a native Rust AI agent engine directly on Android and iOS, running Claude with file tools, code execution, git and extensible MCP device tools, on-device vector memory and background scheduling, with no cloud relay.

  • A Swift package for streaming, on-device speech recognition on iOS, built on a custom Rust inference runtime (VoxRT) rather than ONNX Runtime or PyTorch Mobile, running a 32-million-parameter FastConformer model with roughly 90ms per-chunk latency on an iPhone 13 Pro Max.