\n<\/p>\n
GenieX is an on-device Gen AI inference runtime for Qualcomm devices. Bring almost any GGUF model from Hugging Face \u2014 or a pre-compiled bundle from Qualcomm AI Hub \u2014 and run it locally on the Hexagon NPU, Adreno GPU, or CPU in a few lines of code. One C SDK underneath, exposed through a CLI, Python, Kotlin\/Java, Docker, and an OpenAI-compatible server. It is the community version of Qualcomm GENIE.<\/p>\n
GenieX runs only on Qualcomm Snapdragon. Find your platform, then jump straight to the interface you want to use.<\/p>\n
Platform
\nExample devices
\nJump to a quickstart<\/p>\n
\ud83e\ude9f Windows ARM64 (Compute)
\nSnapdragon X \u00b7 X Elite
\nCLI \u00b7 Python \u00b7 Local server<\/p>\n
\ud83e\udd16 Android (Mobile)
\nSnapdragon 8 Elite \u00b7 8 Elite Gen 5
\nAndroid SDK<\/p>\n
\ud83d\udc27 Linux ARM64 (IoT)
\nDragonwing QCS9075
\nCLI \u00b7 Docker \u00b7 Python<\/p>\n
No device on hand? Spin up a remote session on Qualcomm Device Cloud.<\/p>\n
Pick your interface below. Each one follows the same three steps \u2014 Install, Run, and Docs \u2014 and shows both runtimes: a GGUF model from Hugging Face (llama_cpp) and a pre-compiled bundle from Qualcomm AI Hub (qairt, NPU).<\/p>\n
Install<\/p>\n
Windows ARM64 \u2014 download the installer, run it, then open a new terminal.
\nLinux ARM64 \u2014 one line, no sudo:
\ncurl -fsSL https:\/\/qaihub-public-assets.s3.us-west-2.amazonaws.com\/qai-hub-geniex\/install.sh | sh<\/p>\n
Run \u2014 chat with any model in one line (drag in an image for VLMs):
\n# GGUF from Hugging Face \u2192 llama.cpp (NPU \/ GPU \/ CPU)
\ngeniex infer google\/gemma-4-E4B-it-qat-q4_0-gguf<\/p>\n
# Pre-compiled bundle from Qualcomm AI Hub \u2192 Qualcomm AI Engine Direct (NPU)
\ngeniex infer ai-hub-models\/Qwen2.5-VL-7B-Instruct
\n\ud83d\udcd6 Docs \u2014 Install \u00b7 Quickstart \u00b7 Command reference<\/p>\n
Install<\/p>\n
Run \u2014 mirrors Hugging Face transformers (from_pretrained() \u2192 .generate()):
\n# GGUF from Hugging Face \u2192 llama.cpp
\nfrom geniex import AutoModelForCausalLM<\/p>\n
model = AutoModelForCausalLM.from_pretrained(“unsloth\/Qwen3.5-2B-GGUF”, precision=”Q4_0″)<\/p>\n
messages = ({“role”: “user”, “content”: “What is 2+2?”})
\nprompt = model.tokenizer.apply_chat_template(messages, add_generation_prompt=True)<\/p>\n
for chunk in model.generate(prompt, max_new_tokens=256, stream=True):
\n print(chunk, end=””, flush=True)<\/p>\n
model.close()
\n# Pre-compiled bundle from Qualcomm AI Hub \u2192 Qualcomm AI Engine Direct (NPU)
\nfrom geniex import AutoModelForCausalLM<\/p>\n
model = AutoModelForCausalLM.from_pretrained(“ai-hub-models\/Qwen3-4B”)<\/p>\n
messages = ({“role”: “user”, “content”: “What is 2+2?”})
\nprompt = model.tokenizer.apply_chat_template(messages, add_generation_prompt=True)<\/p>\n
for chunk in model.generate(prompt, max_new_tokens=256, stream=True):
\n print(chunk, end=””, flush=True)<\/p>\n
model.close()
\n\ud83d\udcd6 Docs \u2014 Install \u00b7 Quickstart \u00b7 API reference<\/p>\n
Install \u2014 ships with the CLI (install above).
\nRun \u2014 pull any model (GGUF or Qualcomm AI Hub bundle), then serve an OpenAI-compatible API:
\ngeniex pull ai-hub-models\/Qwen3-4B-Instruct-2507
\ngeniex serve # serves http:\/\/127.0.0.1:18181\/v1
\ncurl http:\/\/127.0.0.1:18181\/v1\/chat\/completions \\
\n -H “Content-Type: application\/json” \\
\n -d ‘{
\n “model”: “ai-hub-models\/Qwen3-4B-Instruct-2507”,
\n “messages”: ({“role”: “user”, “content”: “Hello!”})
\n }’
\nPoint any OpenAI client at http:\/\/127.0.0.1:18181\/v1 \u2014 no code changes.
\n\ud83d\udcd6 Docs \u2014 Local server guide<\/p>\n
Install \u2014 add the SDK to your app module’s build.gradle.kts:
\ndependencies {
\n implementation(“com.qualcomm.qti:geniex-android:0.3.1”)
\n}
\nRun \u2014 fastest path is the sample app (chat UI, model picker for GGUF + Qualcomm AI Hub bundles, VLM support):
\nThe Android demo app lives in qualcomm\/ai-hub-apps. Clone it, open the sample app in Android Studio, and hit Run.
\n\ud83d\udcd6 Docs \u2014 Install \u00b7 Quickstart \u00b7 API reference<\/p>\n
Install
\ndocker pull docker.io\/qualcomm\/geniex:latest
\nRun \u2014 the container wraps the CLI, so geniex infer \u2026 works exactly as above.
\n\ud83d\udcd6 Docs \u2014 Docker guide<\/p>\n
Install \u2014 link against the single C header sdk\/include\/geniex.h; every other interface is a thin wrapper over it.
\n\ud83d\udcd6 Docs \u2014 sdk\/README.md \u00b7 notes\/build.md<\/p>\n
GenieX has two runtimes so you get broad model coverage and peak Snapdragon performance in one stack. Both LLMs and VLMs are supported.<\/p>\n
llama.cpp (llama_cpp)
\nQualcomm AI Engine Direct (qairt)<\/p>\n
Get models from
\nHugging Face (any GGUF)
\nQualcomm AI Hub (pre-compiled)<\/p>\n
Format
\nGGUF
\nPer-chipset bundle<\/p>\n
Compute units
\nNPU \u00b7 GPU \u00b7 CPU
\nNPU only<\/p>\n
Best for
\nBringing your own GGUF
\nHighest NPU performance<\/p>\n
For llama.cpp, pick the Q4_0 precision when prompted \u2014 it has the best Hexagon NPU support. See the Models guide \u2192 for the full list, precisions, and how to run a local model.<\/p>\n
Contributions are welcome! Before opening a PR, please read CONTRIBUTING.md for branch naming, commit \/ PR title format, pre-commit checks, and the FFI-update rule for public SDK headers.<\/p>\n
Questions, ideas, or want to show off what you built? Come say hi.<\/p>\n
\ud83d\udcac Slack \u2014 ask questions and chat with the community in real time.
\n\ud83d\udc1b GitHub Issues \u2014 report a bug or request a feature.
\n\ud83d\udd17 LinkedIn \u2014 follow Qualcomm AI Hub for news and updates.<\/p>\n
Thanks to everyone building GenieX \ud83d\udc99<\/p>\n