{"slug": "qualcomm-acquires-nexa-ai-open-sources-genai-runtime-for-hexagon-npus", "title": "Qualcomm acquires Nexa AI, open-sources GenAI runtime for Hexagon NPUs", "summary": "Qualcomm acquired Nexa AI and open-sourced GenieX, an on-device GenAI inference runtime for its Hexagon NPUs, Adreno GPUs, and CPUs. The runtime supports GGUF models from Hugging Face and pre-compiled bundles from Qualcomm AI Hub, enabling local AI inference on Snapdragon devices via CLI, Python, Kotlin/Java, Docker, and an OpenAI-compatible server.", "body_md": "GenieX is an **on-device Gen AI inference runtime for Qualcomm devices**. 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, 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.\n\nGenieX runs **only on Qualcomm Snapdragon**. Find your platform, then jump straight to the interface you want to use.\n\n| Platform | Example devices | Jump to a quickstart |\n|---|---|---|\n🪟 Windows ARM64 (Compute) |\nSnapdragon X · X Elite |\n|\n\n**Android***(Mobile)*[Android SDK](#android-kotlin--java)**Linux ARM64***(IoT)*[CLI](#cli)·[Docker](#docker)·[Python](#python)No device on hand? Spin up a remote session on\n\n[Qualcomm Device Cloud].\n\nPick your interface below. Each one follows the same three steps — **Install**, **Run**, and **Docs** — and shows both runtimes: a **GGUF** model from Hugging Face (`llama_cpp`\n\n) and a **pre-compiled bundle** from Qualcomm AI Hub (`qairt`\n\n, NPU).\n\n**Install**\n\n**Windows ARM64**—[download the installer](https://github.com/qualcomm/GenieX/releases), run it, then open a new terminal.** Linux ARM64**— one line, no`sudo`\n\n:\n\n```\ncurl -fsSL https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-geniex/install.sh | sh\n```\n\n**Run** — chat with any model in one line (drag in an image for VLMs):\n\n```\n# GGUF from Hugging Face → llama.cpp (NPU / GPU / CPU)\ngeniex infer google/gemma-4-E4B-it-qat-q4_0-gguf\n\n# Pre-compiled bundle from Qualcomm AI Hub → Qualcomm AI Engine Direct (NPU)\ngeniex infer ai-hub-models/Qwen2.5-VL-7B-Instruct\n```\n\n📖 **Docs** — [Install](https://geniex.aihub.qualcomm.com/en/run/cli/install) · [Quickstart](https://geniex.aihub.qualcomm.com/en/run/cli/quickstart) · [Command reference](https://geniex.aihub.qualcomm.com/en/run/cli/reference)\n\n**Install**\n\n```\npip install geniex\n```\n\n**Run** — mirrors Hugging Face `transformers`\n\n(`from_pretrained()`\n\n→ `.generate()`\n\n):\n\n``` python\n# GGUF from Hugging Face → llama.cpp\nfrom geniex import AutoModelForCausalLM\n\nmodel = AutoModelForCausalLM.from_pretrained(\"unsloth/Qwen3.5-2B-GGUF\", precision=\"Q4_0\")\n\nmessages = [{\"role\": \"user\", \"content\": \"What is 2+2?\"}]\nprompt = model.tokenizer.apply_chat_template(messages, add_generation_prompt=True)\n\nfor chunk in model.generate(prompt, max_new_tokens=256, stream=True):\n    print(chunk, end=\"\", flush=True)\n\nmodel.close()\n# Pre-compiled bundle from Qualcomm AI Hub → Qualcomm AI Engine Direct (NPU)\nfrom geniex import AutoModelForCausalLM\n\nmodel = AutoModelForCausalLM.from_pretrained(\"ai-hub-models/Qwen3-4B\")\n\nmessages = [{\"role\": \"user\", \"content\": \"What is 2+2?\"}]\nprompt = model.tokenizer.apply_chat_template(messages, add_generation_prompt=True)\n\nfor chunk in model.generate(prompt, max_new_tokens=256, stream=True):\n    print(chunk, end=\"\", flush=True)\n\nmodel.close()\n```\n\n📖 **Docs** — [Install](https://geniex.aihub.qualcomm.com/en/run/python/install) · [Quickstart](https://geniex.aihub.qualcomm.com/en/run/python/quickstart) · [API reference](https://geniex.aihub.qualcomm.com/en/run/python/api-reference)\n\n**Install** — ships with the CLI ([install above](#cli)).\n\n**Run** — pull any model (GGUF or Qualcomm AI Hub bundle), then serve an OpenAI-compatible API:\n\n```\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  }'\n```\n\nPoint any OpenAI client at `http://127.0.0.1:18181/v1`\n\n— no code changes.\n\n📖 **Docs** — [Local server guide](https://geniex.aihub.qualcomm.com/en/run/cli/local-server)\n\n**Install** — add the SDK to your app module's `build.gradle.kts`\n\n:\n\n```\ndependencies {\n    implementation(\"com.qualcomm.qti:geniex-android:0.3.1\")\n}\n```\n\n**Run** — fastest path is the sample app (chat UI, model picker for GGUF + Qualcomm AI Hub bundles, VLM support):\n\nThe Android demo app lives in [ qualcomm/ai-hub-apps](https://github.com/qualcomm/ai-hub-apps/blob/release/geniex_chat_android/README.md). Clone it, open the sample app in Android Studio, and hit\n\n**Run**.\n\n📖 **Docs** — [Install](https://geniex.aihub.qualcomm.com/en/run/android/install) · [Quickstart](https://geniex.aihub.qualcomm.com/en/run/android/quickstart) · [API reference](https://geniex.aihub.qualcomm.com/en/run/android/api-reference)\n\n**Install**\n\n```\ndocker pull docker.io/qualcomm/geniex:latest\n```\n\n**Run** — the container wraps the CLI, so `geniex infer …`\n\nworks exactly as above.\n\n📖 **Docs** — [Docker guide](https://geniex.aihub.qualcomm.com/en/run/linux/install)\n\n**Install** — link against the single C header [ sdk/include/geniex.h](/qualcomm/GenieX/blob/main/sdk/include/geniex.h); every other interface is a thin wrapper over it.\n\n📖 **Docs** — [sdk/README.md](/qualcomm/GenieX/blob/main/sdk/README.md) · [notes/build.md](/qualcomm/GenieX/blob/main/notes/build.md)\n\nGenieX has two runtimes so you get **broad model coverage** *and* **peak Snapdragon performance** in one stack. Both LLMs and VLMs are supported.\n\nllama.cpp (`llama_cpp` ) |\nQualcomm AI Engine Direct (`qairt` ) |\n|\n|---|---|---|\nGet models from |\n|\n\n[Qualcomm AI Hub](https://aihub.qualcomm.com/models/)(pre-compiled)**Format****Compute units****Best for** For llama.cpp, pick the\n\nprecision when prompted — it has the best Hexagon NPU support. See the`Q4_0`\n\n[Models guide →]for the full list, precisions, and how to run a local model.\n\nContributions 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.\n\n🏗️ Build the CLI, SDK, or Python bindings |\n|\n\n**Run**& select compute units / pull models[notes/run.md](/qualcomm/GenieX/blob/main/notes/run.md)** Release**— SemVer tags, channels, HTP signing[notes/release.md](/qualcomm/GenieX/blob/main/notes/release.md)** All developer docs**[docs/README.md](/qualcomm/GenieX/blob/main/docs/README.md)Questions, ideas, or want to show off what you built? Come say hi.\n\n- 💬\n— ask questions and chat with the community in real time.**Slack** - 🐛\n— report a bug or request a feature.**GitHub Issues** - 🔗\n— follow Qualcomm AI Hub for news and updates.**LinkedIn**\n\nThanks to everyone building GenieX 💙\n\nBSD 3-Clause — see [LICENSE](/qualcomm/GenieX/blob/main/LICENSE) and [NOTICE](/qualcomm/GenieX/blob/main/NOTICE).\n\nUse of this project is also subject to Qualcomm's [Terms of Use](https://www.qualcomm.com/site/terms-of-use).", "url": "https://wpnews.pro/news/qualcomm-acquires-nexa-ai-open-sources-genai-runtime-for-hexagon-npus", "canonical_source": "https://github.com/qualcomm/GenieX", "published_at": "2026-07-07 22:44:11+00:00", "updated_at": "2026-07-07 22:59:26.613873+00:00", "lang": "en", "topics": ["artificial-intelligence", "generative-ai", "ai-tools", "ai-infrastructure", "ai-products"], "entities": ["Qualcomm", "Nexa AI", "GenieX", "Hexagon NPU", "Adreno GPU", "Snapdragon", "Hugging Face", "Qualcomm AI Hub"], "alternates": {"html": "https://wpnews.pro/news/qualcomm-acquires-nexa-ai-open-sources-genai-runtime-for-hexagon-npus", "markdown": "https://wpnews.pro/news/qualcomm-acquires-nexa-ai-open-sources-genai-runtime-for-hexagon-npus.md", "text": "https://wpnews.pro/news/qualcomm-acquires-nexa-ai-open-sources-genai-runtime-for-hexagon-npus.txt", "jsonld": "https://wpnews.pro/news/qualcomm-acquires-nexa-ai-open-sources-genai-runtime-for-hexagon-npus.jsonld"}}