Is there any way to use static linking in node?

[NazaRik555]

I'm looking for a way to use static lib instead of dll in node js bindings. Because when I do "npm i onnxruntime-node" it downloads dll

[cookesan]

Not with the published onnxruntime-node npm package. The package is designed around a native .node binding plus ONNX Runtime shared-library binaries for the target platform.

In the current Node package:

• package.json runs node ./script/install on postinstall.
• the install script exists to ensure required platform binaries are present/downloaded.
• js/node/CMakeLists.txt builds onnxruntime_binding as a shared Node addon and copies onnxruntime.dll, libonnxruntime.dylib, or libonnxruntime.so into the package output.

--onnxruntime-node-install=skip only skips extra non-bundled downloads such as CUDA EP binaries; it does not turn the package into a statically linked build.

If you need a static-link experiment, that would be a custom source build of the Node binding and ONNX Runtime, not an npm install onnxruntime-node option.

[NazaRik555]

And what about x32 platforms? I understand that it would be a custom build, but what promblems should I face with to do that?

[cookesan]

If by x32 you mean 32-bit x86 / Node process.arch === "ia32", treat it as an unsupported custom build path rather than something the published onnxruntime-node package can do.

The current package/release packaging is built around bin/napi-v6/{os}/{cpu_arch} and the release pipeline comments list the packaged CPU arch values as x64 and arm64 for win32, darwin, and linux. The install metadata also only knows about win32/x64, win32/arm64, linux/x64, linux/arm64, darwin/x64, and darwin/arm64. There is a local Node build script option for --arch=ia32, but that does not mean the npm package publishes or tests 32-bit binaries.

The main problems I would expect are:

• every native piece must be 32-bit: Node headers/runtime, onnxruntime_binding.node, onnxruntime.dll/.so, and any provider DLLs/shared libraries
• many execution-provider binaries are only packaged for the normal release targets; CUDA/TensorRT are not realistic for 32-bit, and Windows DML/WebGPU dependencies may also be a blocker
• the Node addon and ONNX Runtime shared library must use matching ABI/toolchain/runtime settings
• model size, tensor allocation, and arena behavior can hit 32-bit address-space limits much sooner
• you would need to extend packaging/prepack/install metadata if you want an installable package layout, not only a local build folder

If you mean the Linux x32 ABI specifically rather than normal 32-bit x86, I would be even more conservative: I do not see that ABI in the Node package or release packaging path at all.

[jcchatham8-lgtm]

please help me get this off my phone my ex is running it into the ground with a