GenieX

GenieX is Qualcomm’s open-source SDK for running frontier large language models and vision-language models locally on Qualcomm-powered devices. It executes models directly on the device’s NPU, GPU and CPU, so AI features work offline, with low latency and without sending user data to the cloud. The project has about 8,400 stars on GitHub and is actively developed in 2026.

Overview

Built on Qualcomm’s AI Engine Direct and AI Hub ecosystem, GenieX lets developers load popular open models — including Llama 3, Qwen3, Gemma and other recent LLM and VLM families — and run conversational or multimodal inference with just a few lines of code. The runtime handles the heterogeneous compute of Snapdragon platforms, distributing work across the Hexagon NPU, Adreno GPU and CPU for the best performance per watt. It suits phones, PCs, IoT devices and other Snapdragon-based hardware GenieX is part of Qualcomm’s broader push to make its silicon a first-class target for open AI models: models are prepared and optimized through Qualcomm AI Hub, then executed locally through the GenieX runtime, which exposes chat-style interfaces for both text-only and multimodal (image plus text) interactions. For end users this means assistant features that respond instantly, keep working on a plane, and never transmit conversations to a server — properties that cloud APIs cannot offer at any price. where on-device AI is a requirement rather than a nice-to-have.

Key Features

• Run frontier LLMs and vision-language models fully on-device
• Executes across Snapdragon NPU, GPU and CPU for performance and efficiency
• Supports popular open model families such as Llama, Qwen and Gemma
• Simple APIs — working local inference in a few lines of code
• Offline, private inference with no cloud round-trips
• Part of Qualcomm’s AI Hub tooling for model optimization and deployment

💳 Pricing

GenieX is free and open source under the BSD-3-Clause license. It targets devices built on Qualcomm Snapdragon platforms.

Who Is It For

GenieX is for mobile, PC and embedded developers building on-device AI features on Snapdragon hardware, where privacy, latency or connectivity rule out cloud inference.