GenieX in qualcomm/geniex is a generative AI inference runtime that runs large language models (LLMs) and vision-language models (VLMs) locally on Snapdragon devices. It supports Hexagon NPU, Adreno GPU, and CPU execution paths. The latest release listed in the repository is v0.9.3, but the provided information does not list the changes in this version.
Models can be obtained in GGUF format from Hugging Face, ModelScope, and Docker Hub, or as precompiled bundles from Qualcomm AI Hub. A single C SDK is usable from a CLI, Python, Kotlin/Java, Docker, and an OpenAI-compatible server, and it supports Windows ARM64, Android, and Linux ARM64. It is written in Go and is the Qualcomm GENIE community edition under the BSD-3-Clause license. It is close to a tool for Qualcomm devices that runs models locally via CLI or Python, or that starts an API server.
The provided information contains no per-hardware inference speed or accuracy comparisons. So while the supported paths and platforms can be confirmed, there is not enough basis to judge performance on a specific device or the suitability of individual models.