Qualcomm AI Engine Direct

source

closed(no published source for the QNN/Qualcomm AI Runtime SDK core engine

license

proprietary(no license to the source since it is not published)

open-wrapper

BSD-3-Clause(qualcomm/qai-appbuilder, formerly quic/ai-engine-direct-helper -- a convenience API layer that requires the proprietary SDK libraries)

Qualcomm publishes no source for the QNN/Qualcomm AI Runtime SDK; it ships only as an EULA-gated binary from the Qualcomm Software Center. A BSD-3-Clause convenience wrapper, qai-appbuilder (formerly ai-engine-direct-helper), still requires libraries from that same proprietary SDK, so it does not open the engine underneath.

• https://github.com/qualcomm/qai-appbuilder recorded 2026-08-13

  The wrapper under its current name: BSD 3-Clause, and "Some libraries from the Qualcomm® AI Runtime SDK are required to use QAI AppBuilder" - an open wrapper over a proprietary SDK that must be downloaded separately from Qualcomm's Software Center.

• https://github.com/quic/ai-engine-direct-helper recorded 2026-08-09

  the former name, kept because the openness note turns on the move; redirects to qualcomm/qai-appbuilder

Qualcomm AI Engine Direct reaches developers through integration rather than direct download: LiteRT, ExecuTorch and ONNX Runtime each ship it as their Qualcomm NPU backend. Qualcomm publishes no developer, device or download figure of its own, so the reading is drawn from that integration breadth rather than a reported number.

• https://ai.google.dev/edge/litert/android/npu/qualcomm recorded 2026-09-29

  The LiteRT Qualcomm NPU page documents "The Qualcomm AI Engine Direct Delegate enables users to run LiteRT models" and lists supported devices including "Snapdragon 8 Elite (SM8750)".

• https://docs.pytorch.org/executorch/stable/backends-qualcomm.html recorded 2026-09-29

  The ExecuTorch Qualcomm backend page says "this ExecuTorch Backend can delegate AI computations to Hexagon processors and Adreno GPU through Qualcomm AI Engine Direct APIs".

• https://onnxruntime.ai/docs/execution-providers/QNN-ExecutionProvider.html recorded 2026-09-29

  The ONNX Runtime QNN page says "The QNN Execution Provider for ONNX Runtime enables hardware accelerated execution on Qualcomm chipsets" and "It uses the Qualcomm AI Engine Direct SDK (QNN SDK) to construct a QNN graph from an ONNX model".

Qualcomm AI Engine Direct provides a unified API across the Hexagon NPU, Adreno GPU and Kryo CPU, with quantized on-device LLM and VLM inference, and LiteRT, ExecuTorch and ONNX Runtime all delegate to it as their Qualcomm NPU backend. It remains Snapdragon-locked and edge-scale, well short of the server-class engines that lead the field.

• https://developers.googleblog.com/unlocking-peak-performance-on-qualcomm-npu-with-litert/ recorded 2026-09-29

  The Google Developers blog says LiteRT models deploy "either ahead-of-time (AOT) or on-device compilation" and reports "up to 100x speedup over CPU" on the Qualcomm NPU, with Gemma and FastVLM as example models.

• https://docs.pytorch.org/executorch/stable/backends-qualcomm.html recorded 2026-09-29

  The ExecuTorch Qualcomm backend page says "Qualcomm AI Engine Direct is designed to provide unified, low-level APIs for AI development" and that developers can interact with "Kryo CPU, Adreno GPU, and Hexagon processors".