Nexa SDK is a developer toolkit from Nexa AI for running AI models locally on-device rather than through a cloud API. Nexa AI, the company behind it, was founded in 2023 and focuses on on-device AI deployment and compact model research.
The SDK, powered by Nexa's own NexaML inference engine, runs large language models, vision-language models, and other model types such as embedding, reranking, speech-to-text, and text-to-speech, across NPU, GPU, and CPU hardware on Android, Windows, and Linux, with day-zero support for new model architectures such as OpenAI's GPT-OSS, IBM's Granite 4, and Qwen3-VL. Because inference happens on the device itself, teams get lower latency and keep data off external servers, a distinction that matters for regulated industries or products running at the edge, and one reason Qualcomm maintains its own fork of the SDK for NPU-accelerated devices.
Nexa SDK suits developers and hardware manufacturers building AI features into PCs, mobile devices, and wearables who need model portability across chip vendors, rather than teams content to call a cloud LLM API. Personal use is free with a license key from the Nexa AI Model Hub covering one device; commercial licensing is arranged directly with the company and pricing is not publicly listed.
