Add PyInstaller Windows release build system.

Frozen builds resolve paths correctly, stage portable onedir output, and zip releases for distribution.

README.md

@@ -67,6 +67,45 @@ You can also launch directly:
 python main.py
 ```
 
+## Build Windows Release (PyInstaller)
+
+Production builds produce a portable **onedir** folder under `dist/GODCWAK/` and a zip under `release/`.
+
+From PowerShell in the project directory:
+
+```powershell
+# One-time: activate venv and install build deps
+.\.venv\Scripts\Activate.ps1
+pip install -r requirements-build.txt
+
+# Build release (no console window)
+python build\build_release.py --clean
+
+# Troubleshooting build (shows console + logs)
+python build\build_release.py --clean --debug
+```
+
+Or double-click `build_release.bat` at the project root.
+
+**Output layout**
+
+```text
+dist/GODCWAK/
+  GODCWAK.exe              Main application
+  Start GODCWAK.bat        Optional launcher
+  config_ollama.json       Writable Ollama settings (seeded on first run)
+  data/                    Passwords, proxies, attempt logs (created at runtime)
+  _internal/               Bundled Python runtime and dependencies
+release/
+  GODCWAK-v1.0-win64-YYYYMMDD.zip
+```
+
+**Notes**
+
+- Default release excludes CuPy/Numba/Playwright to keep the bundle smaller; use `--gpu` if CUDA libraries are installed and you want them bundled.
+- Place `assets/icon.ico` before building to embed a custom application icon.
+- Packaged builds log to `data/godcwak.log` and show error dialogs instead of a console window.
+
 ## GPU Notes
 
 GODCWAK checks for CuPy at startup. With `cupy-cuda12x` installed and a working NVIDIA CUDA-capable GPU, password-combination index generation can use the GPU for larger workloads. Without CUDA, the app falls back to NumPy CPU acceleration or pure Python.