#2020th of 47 in Model serving
geniex
LLM and VLM inference runtime for Qualcomm Snapdragon NPU, GPU and CPU
- Stars
- 8.5k
- License
- BSD-3-Clause
- Last commit
- Oct 2026
- Last release
- Oct 2026
- Language
- Go
7,783 on 1 Apr 20268,452 on 21 Oct 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
GenieX is an on-device inference runtime for Snapdragon devices, built on one C SDK. It runs GGUF models through llama.cpp on NPU, GPU or CPU, and pre-compiled Qualcomm AI Hub bundles through Qualcomm AI Engine Direct on the NPU. It is exposed as a CLI, Python package, Android Kotlin/Java SDK, Docker image and an OpenAI-compatible server on port 18181. Status is developer preview.
Who it is for: Developers deploying local LLMs on Snapdragon PCs, phones and Linux IoT boards
Strengths
- Runs GGUF models from Hugging Face, ModelScope or Docker Hub on Hexagon NPU, Adreno GPU or CPU
- Pre-compiled AI Hub bundles use the qairt runtime for NPU-only execution
- One C SDK with CLI, Python, Android, Docker and OpenAI-compatible server wrappers
- Python API mirrors transformers from_pretrained() and generate()
Weaknesses
- Runs only on Qualcomm Snapdragon; no x86 or Apple hardware support
- Marked developer preview; interfaces may change between releases
- AI Hub bundles are per-chipset and the available catalogue changes independently
- For llama.cpp on the Hexagon NPU, only the Q4_0 precision is recommended
What it needs
- GPU optional
- Needs Qualcomm Snapdragon device, llama.cpp, Qualcomm AI Engine Direct (QAIRT)
- Models: GGUF, Qualcomm AI Hub bundles, LLM, VLM
- port 18181
Also in Model serving
See all 47| Rank | Project | Score |
|---|---|---|
| 1 | LocalAISelf-hosted inference server with OpenAI-compatible API and pluggable backends | 83 out of 100 |
| 2 | llama.cppC/C++ inference engine serving GGUF models over an OpenAI-compatible API | 81 out of 100 |
| 3 | OllamaLocal model runner with CLI and REST API for open-weight LLMs | 80 out of 100 |
| #4 | strataRuns a 125B-parameter MoE model on one consumer GPU plus system RAM | 79 out of 100 |
| #5 | vLLMInference engine for Hugging Face LLMs with PagedAttention and continuous batching | 77 out of 100 |