# qualcomm/GenieX

Run frontier LLMs and VLMs locally on Qualcomm devices across NPU, GPU, and CPU with a few lines of code

Repository: https://github.com/qualcomm/GenieX
Canonical: https://ross.abutalabs.com/products/geniex
Homepage: https://geniex.aihub.qualcomm.com/en/get-started/what-is-geniex
Language: Rust
License: BSD-3-Clause
License Family: permissive
Topics: llm, on-device-ai, sdk, vlm, go, gpt-oss, llama3, qwen3, granite4, qwen3vl, gemma4, hexagon, local-ai, qualcomm, snapdragon
Last push: 2026-08-26T15:04:29+00:00

## Health v2 (maintenance only)
Score: 85/100 (v2, computed 2026-09-03T02:39:23.370411+00:00)
- activity 99, release rhythm 86, longevity 53
- inputs: {"age_days": 747, "days_push": 7, "days_rel": 12, "gap_med": 2, "n_releases_24m": 64}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 8326, forks 1049 (observed 2026-08-28T04:10:20.315005+00:00)

## What it is
GenieX is an on-device generative AI inference runtime for Qualcomm Snapdragon devices, running frontier LLMs and VLMs on the Hexagon NPU, Adreno GPU, or CPU. It exposes a single C SDK through a CLI, Python and Kotlin/Java SDKs, Docker, and an OpenAI-compatible server, dispatching to llama.cpp (GGUF) or Qualcomm AI Engine Direct runtimes.

## Use cases
- run llama 3 or qwen3 locally on a snapdragon laptop npu
- run gguf models from hugging face on hexagon npu
- embed on-device llm inference in an android app
- serve an openai-compatible local llm api on windows arm64
- deploy vision-language models on qualcomm dragonwing iot devices
- run llms offline on-device without cloud
- benchmark npu vs gpu vs cpu inference on snapdragon

## When to choose
- you target Qualcomm Snapdragon hardware (Snapdragon X, 8 Elite, Dragonwing) and want NPU-accelerated local LLM/VLM inference
- you want to run GGUF models from Hugging Face with minimal setup
- you need an OpenAI-compatible local server or mobile (Kotlin) SDK for on-device AI
- you want pre-compiled, chipset-optimized models from Qualcomm AI Hub

## When to avoid
- you run on non-Qualcomm hardware (Intel, AMD, Apple Silicon) - GenieX only works on Snapdragon
- you need server-scale multi-GPU inference or high-concurrency production serving
- you need non-GGUF model formats without a pre-compiled AI Hub bundle
- you require a fully stable release - it is currently in developer preview

## Facets
- artifact type: library
- maturity: active
- function: llm-inference, sdk, cli, http-server, machine-learning
- domain: large-language-models, artificial-intelligence, machine-learning, developer-tools, mobile-development, iot
- platform: windows, python, cli
- tags: on-device-ai, qualcomm, snapdragon, hexagon-npu, gguf, llama-cpp, openai-compatible, vlm, edge-ai, rust, android, linux, docker, arm64

## Member repositories
- qualcomm/GenieX (main) score 85

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:10:20.315005+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:27:45.986526+00:00, confidence not recorded.
  - readme: https://github.com/qualcomm/GenieX (fetched 2026-08-28T04:10:20.315005+00:00, sha 42bf9c35f07c)
  - homepage: https://geniex.aihub.qualcomm.com/en/get-started/what-is-geniex (fetched 2026-08-29T08:27:39.325315+00:00, sha 27b1826e38e7)
  - site_page: https://geniex.aihub.qualcomm.com/en/get-started/quickstart (fetched 2026-08-29T08:27:39.334933+00:00, sha aa50bb96cdd3)
  - site_page: https://geniex.aihub.qualcomm.com/en/resources/faq (fetched 2026-08-29T08:27:39.337020+00:00, sha 926c74f36986)
- Data as of 2026-08-30T08:39:29.467469+00:00.