NeuralAim

Native C++ vision-assist tooling for Windows, built around DirectML and CUDA + TensorRT runtime paths.

Repository  |  Discord Builds  |  Build Guide  |  Config Reference

---

Overview

NeuralAim ships with two Windows runtime options:

Runtime	Best fit	Setup profile
DirectML (DML)	NVIDIA, AMD, Intel, integrated graphics, laptops, and older systems	Universal Windows 10/11 x64 build with no CUDA requirement
CUDA + TensorRT	Supported NVIDIA GPUs where maximum performance matters	Requires a current NVIDIA driver and CUDA 13.1

Precompiled .exe builds are distributed through the pre-releases channel on Discord:

https://discord.gg/qF8h4CDsyY

After downloading a package, unpack it and run ai.exe.

Runtime Support

DirectML

Choose the DML build when compatibility matters more than NVIDIA-specific acceleration.

Requirement	Details
OS	Windows 10/11 x64
GPU	Modern NVIDIA, AMD Radeon, Intel Iris/Xe, or integrated graphics
Extra runtime	None
Recommended for	GTX 10xx/9xx/7xx, AMD GPUs, Intel GPUs, laptops, and office PCs

CUDA + TensorRT

Choose the CUDA build for the fastest supported NVIDIA path.

Requirement	Details
OS	Windows 10/11 x64
GPU	GTX 1660 or RTX 2000/3000/4000/5000
Driver	Latest NVIDIA graphics driver
CUDA	CUDA 13.1
Not supported	GTX 10xx/Pascal and older because of TensorRT limitations
Source-build extra	TensorRT 10.14.1.48 is only needed separately when building from source

The precompiled CUDA release includes the required CUDA + TensorRT runtime files. Before troubleshooting missing CUDA DLLs or launch-close issues, update the NVIDIA driver and install CUDA 13.1.

Quick Start

1. Join the Discord server.
2. Download the DML or CUDA package from the pre-releases channel.
3. Unpack the build.
4. For CUDA, install or update the NVIDIA driver and CUDA 13.1.
5. For DML, skip extra runtime setup.
6. Start ai.exe.
7. Wait for the first model export. The first launch can take up to 5 minutes.
8. Place your .onnx model in the models folder.
9. Press Home, select the model in the overlay, and tune settings.

Controls

Input	Action
Right Mouse Button	Aim at the detected target
F2	Exit
F3	Pause aiming
F4	Reload config
Home	Open or close the overlay and settings

Build From Source

Source builds are for advanced users who want to modify NeuralAim or prepare their own runtime package.

Resource	Link
Build workflow	docs/build.md
CUDA Toolkit	developer.nvidia.com/cuda-downloads
TensorRT docs	docs.nvidia.com/deeplearning/tensorrt

Documentation

Topic	Link
Config reference	docs/config.md
Setup, FAQ, troubleshooting	docs/guides.md
Backend notes	docs/guides/backends.md
Input methods	docs/guides/input-methods.md
Data collection	docs/guides/data-collection.md
Source config files	NeuralAim/config/config.cpp, NeuralAim/config/config.h

References

• TensorRT Documentation
• OpenCV Documentation
• ImGui
• CppWinRT
• GLFW
• WindMouse
• KMBOX
• MAKCU
• depth-anything-tensorrt

License Notes

Dependency	License
OpenCV	Apache License 2.0
ImGui	MIT License

Support And Community

Development is supported through Boosty and Patreon. Supporters get access to improved and better-trained AI models.

For builds, help, discussion, and contribution questions, join the Discord server:

https://discord.gg/qF8h4CDsyY

Star History