qualcomm/GenieX

Project overview

Geniex provides a single on-device runtime and SDK for executing LLMs and VLMs locally on Qualcomm Snapdragon hardware. It enables on-device inference across the Hexagon NPU, Adreno GPU, or CPU via CLI, Python, Java, Docker, or an OpenAI-compatible API server.

Project type

Model Runtime

Use cases

Chat Assistants

Deployment

Refer to project documentation

License

BSD-3-Clause

Best for

• Developers building on-device generative AI applications on Qualcomm Snapdragon hardware running Windows ARM64, Linux ARM64, or Android.
• AI engineers needing a unified runtime to test or serve GGUF models or Qualcomm AI Hub bundles locally without external server dependencies.

Key capabilities

• Executes LLMs and VLMs locally on Qualcomm Snapdragon hardware using the Hexagon NPU, Adreno GPU, or CPU.
• Provides a CLI to pull models and run inference with a single command.
• Provides Python bindings exposing a Hugging Face transformers-like interface including from_pretrained and generate functions.
• Serves an OpenAI-compatible API locally to allow existing OpenAI clients to connect without code changes.
• Provides an Android SDK for integrating model execution into mobile applications.
• Provides a single C header underlying all other interfaces for direct integration.

Limitations and risks

• Runs only on Qualcomm Snapdragon platforms and cannot be used on other hardware.
• The project is currently a Developer Preview, which may involve ongoing changes or incomplete production stability.

Getting started

• Install the runtime via the pip install command or a pre-built installer, then run a single command to chat with a model.

Evidence and sources

• GitHub project description: Run frontier LLMs and VLMs locally on Qualcomm devices across NPU, GPU, and CPU with a few lines of code
• README: GenieX is an **on-device Gen AI inference runtime for Qualcomm devices**.
• README: link against the single C header [`sdk/include/geniex.h`](sdk/include/geniex.h); every other interface is a thin wrapper over it.
• README: Bring almost any GGUF model from Hugging Face — or a pre-compiled bundle from [Qualcomm AI Hub](https://aihub.qualcomm.com/models/) — and run it locally on the **Hexagon NPU, Adre…
• README: []

Reviewed at: 2026-07-18T05:28:26Z