Before you start
What happened
When running the FreeToken Windows desktop installer and attempting to change the target installation directory from the default system path to a custom path on a secondary drive (e.g., D:\CustomFolder\FreeToken), the installation fails or the resulting application build breaks/fails to resolve paths correctly upon launch.
To Reproduce:
Download the latest FreeToken Windows desktop installer.
Launch the installer.
At the destination directory prompt, click Browse and select a custom subfolder on a secondary drive (e.g., D:...).
Complete the installation process.
Attempt to launch the application.
Expected Behavior:
resolving the latest engine wheels …
powershell.exe -File "C:\Users\Novicracker\AppData\Local\FreeToken Desktop\eng
ine\install.ps1"
[warn] console mode unavailable (spawn: CreateProcessW "powershell.exe -Executi onPolicy Bypass -NoProfile -NonInteractive -File \"C:\\Users\\Novicracker\\AppDa ta\\Local\\FreeToken Desktop\\engine\\install.ps1\"\0" in cwd Some("C:\\Users\ \Novicracker\0") failed: The system cannot find the file specified. (os error 2
)); showing plain output
Could not start Windows PowerShell to run the installer: program not found
Environment:
OS: Windows 11 Pro (UEFI)
Installation Target Attempted: D: Drive subfolder
Desktop app version
v0.2.0-beta.20
OS
Windows 11
OS details
Installation Target Attempted: D: Drive subfolder
GPU and driver
NO NVIDIA CARD INSTALLED
CPU and system RAM
i7-7700K, DDR4 32GB RAM
Checkpoint
qwen2.5-1.5b-instruct-q4_k_m.gguf
Model settings
I have no spec or LLM I just through that one in there to fill the gaps. We never made it to the LLM stage.
Engine log
> resolving the latest engine wheels …
> powershell.exe -File "C:\Users\Novicracker\AppData\Local\FreeToken Desktop\eng
ine\install.ps1"
[warn] console mode unavailable (spawn: CreateProcessW `"powershell.exe -Executi
onPolicy Bypass -NoProfile -NonInteractive -File \"C:\\Users\\Novicracker\\AppDa
ta\\Local\\FreeToken Desktop\\engine\\install.ps1\"\0"` in cwd `Some("C:\\Users\
\Novicracker\0")` failed: The system cannot find the file specified. (os error 2
)); showing plain output
Could not start Windows PowerShell to run the installer: program not found
Anything else
Well... yeah. Not having a translation layer for CUDA-to-ROCm or CUDA-to-Vulkan is severely limiting this project's reach and user adoption. Sure, targeting the existing industry standard is easy, but the people pushing the envelope to keep local AI accessible and affordable for everyone rely on an even hardware playing field.
CUDA is only the dominant standard because NVIDIA got to the table first with a mature software lock-in, not because hardware alternatives are incapable. If we want to prevent a permanent hardware monopoly where local AI is locked behind expensive green cards, the software layer has to support cross-vendor ecosystems natively. Think about it: true engineering progress happens when we break down artificial vendor boundaries, not when we build around them.
Before you start
What happened
When running the FreeToken Windows desktop installer and attempting to change the target installation directory from the default system path to a custom path on a secondary drive (e.g., D:\CustomFolder\FreeToken), the installation fails or the resulting application build breaks/fails to resolve paths correctly upon launch.
To Reproduce:
Download the latest FreeToken Windows desktop installer.
Launch the installer.
At the destination directory prompt, click Browse and select a custom subfolder on a secondary drive (e.g., D:...).
Complete the installation process.
Attempt to launch the application.
Expected Behavior:
Environment:
OS: Windows 11 Pro (UEFI)
Installation Target Attempted: D: Drive subfolder
Desktop app version
v0.2.0-beta.20
OS
Windows 11
OS details
Installation Target Attempted: D: Drive subfolder
GPU and driver
NO NVIDIA CARD INSTALLED
CPU and system RAM
i7-7700K, DDR4 32GB RAM
Checkpoint
qwen2.5-1.5b-instruct-q4_k_m.gguf
Model settings
I have no spec or LLM I just through that one in there to fill the gaps. We never made it to the LLM stage.
Engine log
Anything else
Well... yeah. Not having a translation layer for CUDA-to-ROCm or CUDA-to-Vulkan is severely limiting this project's reach and user adoption. Sure, targeting the existing industry standard is easy, but the people pushing the envelope to keep local AI accessible and affordable for everyone rely on an even hardware playing field.
CUDA is only the dominant standard because NVIDIA got to the table first with a mature software lock-in, not because hardware alternatives are incapable. If we want to prevent a permanent hardware monopoly where local AI is locked behind expensive green cards, the software layer has to support cross-vendor ecosystems natively. Think about it: true engineering progress happens when we break down artificial vendor boundaries, not when we build around them.