Library / SDK
Inference & Serving
Nexa SDK
1On-device 2inference 3across 4CPU 5GPU 6NPU
the six Ws · specification
W1
Who
App developers deploying LLMs and multimodal models directly on phones, PCs, and edge devices.
W2
What
Nexa SDK is an on-device inference framework running LLMs, VLMs, ASR, and TTS models across CPU, GPU, and NPU backends with an OpenAI-compatible local API server.
W3
Where
Self-hosted on Android, iOS, Windows, macOS, and Linux/IoT devices via native bindings.
W4
When
Actively maintained by Nexa AI with day-0 support for new model releases through 2026.
W5
Why
It exists to bring efficient, private, quantized model inference to consumer and edge hardware.
W6
With
Supports GGUF, MLX, and Nexa's own quantized model formats.
W7
Watch
Open source SDK with commercial hardware partnerships; self-hosted, runs fully on-device; no data leaves the device.
alternatives