Qualcomm GenieX (formerly Nexa SDK) iconQualcomm GenieX (formerly Nexa SDK)

Open-source on-device runtime for LLMs and VLMs on Snapdragon NPU, GPU and CPU; formerly NexaSDK from Nexa AI, now run by Qualcomm, BSD-3-Clause

8.5K+ GitHub Stars
NPU Hexagon, plus Adreno GPU and CPU
BSD-3 License

Overview

GenieX is Qualcomm's open-source runtime for running language and vision-language models locally on Snapdragon devices, on the Hexagon NPU, Adreno GPU or CPU, with a few lines of code. It started life as NexaSDK from Nexa AI: the nexa.ai domain now redirects to Qualcomm AI Hub, the NexaAI/nexa-sdk repo moved to qualcomm/GenieX, and Qualcomm's developer blog calls it "GenieX (formerly NexaSDK)". Chinese tech press reports that Qualcomm acquired Nexa AI and its small team in March 2026 (Huxiu); Qualcomm has not published deal terms. GenieX runs community GGUF models through llama.cpp and pre-compiled Qualcomm AI Hub models through Qualcomm AI Engine Direct (NPU only), and exposes a CLI, Python SDK, Android SDK (Java and Kotlin), Docker image and an OpenAI-compatible local server. It calls itself the community version of Qualcomm GENIE and is in developer preview.

The Verdict

Who Should Use Qualcomm GenieX?

Best For

  • App developers targeting Snapdragon X Windows laptops or Snapdragon Android phones
  • IoT and robotics teams on Qualcomm Dragonwing boards
  • Former NexaSDK users following the project to its new home
  • Anyone who wants an OpenAI-compatible local server that uses the NPU

Not Ideal For

  • Mac, Intel, AMD or NVIDIA hardware: the SDK now targets Qualcomm only
  • Teams that need a stable, generally available release today
  • Users who want a desktop chat app rather than an SDK

What's Great

  • Permissive BSD-3-Clause license
  • Uses the Hexagon NPU, which most local runtimes ignore
  • Two backends: broad GGUF coverage via llama.cpp and tuned AI Hub models
  • Many interfaces: CLI, Python, Android, Docker, OpenAI-compatible server
  • Very active repo, pushed the day before this page was checked

Watch Out For

  • Snapdragon-only: Windows ARM64, Android and Linux ARM64 on Qualcomm chips
  • Use is also subject to Qualcomm's Terms of Use, on top of the BSD license
  • Still a developer preview, ahead of general availability
  • Rebrand means older NexaSDK guides and links may be out of date

Pricing

View all features & details

Key Features

  • LLM and VLM inference on Hexagon NPU, Adreno GPU or CPU
  • llama.cpp backend for community GGUF models
  • Qualcomm AI Engine Direct backend for AI Hub models
  • CLI, Python SDK, Android SDK, Docker image
  • OpenAI-compatible local server

Supported Platforms

  • Windows ARM64 (Snapdragon X, X Elite)
  • Android (Snapdragon 8 Elite, 8 Elite Gen 5)
  • Linux ARM64 (Dragonwing IoT chipsets)

How It Compares

Feature Qualcomm GenieX Ollama llama.cpp LM Studio
Hardware Snapdragon NPU, GPU, CPU Cross-platform Cross-platform Cross-platform
Uses Qualcomm NPU Yes See vendor See vendor See vendor
Android SDK Yes No Library only No
Form SDK and CLI CLI and app Library and CLI Desktop app

Sources: GenieX repo · GenieX docs · Qualcomm blog (Mar 2026)

Is this your tool? Add a badge