openvino_genai_windows 0.1.0
openvino_genai_windows: ^0.1.0 copied to clipboard
Windows x64 implementation of openvino_genai. Downloads the pinned Intel OpenVINO GenAI runtime (SHA-256 verified), compiles the C bridge and bundles the DLLs.
openvino_genai_windows #
The Windows x64 implementation for openvino_genai.
Apps should not depend on this package directly — depend on
openvino_genai and this one comes along automatically.
Unofficial. Not affiliated with, endorsed by, or supported by Intel Corporation. "OpenVINO" is a trademark of Intel Corporation or its subsidiaries.
What it does at build time #
When a Flutter Windows app that (transitively) depends on this package is
built, this package's windows/CMakeLists.txt:
- Downloads the pinned official OpenVINO GenAI runtime archive from
storage.openvinotoolkit.org— ~233 MB, verified against a pinned SHA-256, cached under%LOCALAPPDATA%\openvino_genai_dartso it downloads once per machine, not per build. - Compiles the C bridge (
openvino_genai_bridge.dll) from source against that exact archive's headers and import libraries, so the bridge and the runtime can never be version-skewed. - Bundles the runtime DLL set next to your
.exevia Flutter's standard plugin bundling.
Runtime pin #
| Package version | OpenVINO GenAI runtime | Archive SHA-256 |
|---|---|---|
| 0.1.x | 2026.0.0.0 | 8a6a75eb…beb1fb |
One release, one pin. If your app also uses OpenVINO through another channel, make sure only ONE copy of the runtime DLL set ends up in the process.
Offline / proxy-restricted builds #
Pre-download the archive (any machine), then point the build at it:
set OPENVINO_GENAI_ARCHIVE=C:\downloads\openvino_genai_windows_2026.0.0.0_x86_64.zip
flutter build windows
OPENVINO_GENAI_CACHE_DIR redirects the cache directory. Both are also
available as CMake -D variables.
What gets bundled #
The field-trimmed Release set (~130 MB): core runtime + GenAI + C API + tokenizers (with ICU), the CPU/GPU/NPU device plugins, AUTO/HETERO/BATCH plugins, the IR frontend, and Release TBB.
Not bundled by default: the NPU compiler (72 MB — the NPU plugin alone
cannot compile models for the NPU) and the ONNX/TensorFlow/TFLite/
PyTorch/Paddle frontends (OpenVINO IR models don't need them). Building with
OPENVINO_GENAI_BUNDLE_FULL=ON restores all of them.
Licensing #
This package is Apache-2.0. The OpenVINO runtime it downloads and bundles is
© Intel Corporation, licensed Apache-2.0; the files bundled beside your app
are listed in Intel's redist.txt. The bundled tokenizers pull in ICU, whose
license requires its notice to accompany redistributions (ICU-LICENSE).
Copies of the relevant licenses and third-party notices ship in this
package's third_party/ directory — ship them with your app (e.g. in your
installer or an about screen).