About this project

Turbo AI Chat is a native on-device large model chat application specifically designed for HarmonyOS NEXT. Built using ArkTS, C++ N-API, and MNN Runtime, the project implements a pure native inference chain (ArkTS → N-API → MNN → CPU) that interfaces directly with the HarmonyOS NDK without requiring an Android compatibility layer. Key features include: - Local Inference: Supports running LLMs directly on the device without network dependency, providing streaming output and generation interruption functions. - Model Management: Features a built-in model plaza supporting one-click downloads of preset models from ModelScope (such as Qwen3-4B-Instruct, MiniCPM5-1B, Gemma-4-E2B-it), and supports manual import of compatible MNN model directories via zip or hdc. - Multimodal Capabilities: Supports text dialogue and image understanding (dependent on compatible multimodal MNN models). - Developer Tools: Provides real-time runtime monitoring (CPU, memory, temperature, TTFT/TPOT, etc.) and a LAN OpenAI-compatible API service, allowing external clients (such as Cherry Studio) to call the on-device model. - Interaction Optimization: Adapted for phone, tablet, and PC windows, supporting offline Mandarin voice input and comprehensive Markdown rendering (including code block copying, tables, etc.). This project provides a verifiable engineering baseline for developers exploring "HarmonyOS + Local AI" solutions.