Voxta KoboldAI / KoboldCpp
Run a local GGUF model with KoboldCpp and connect Voxta to its API: download, launch with GPU offload, and add the endpoint.
KoboldCpp is a single-executable local LLM runtime built on llama.cpp. You install KoboldCpp separately, point it at a GGUF model, and Voxta connects to its API endpoint.
Setup
Download KoboldCpp
Grab the latest release from github.com/LostRuins/koboldcpp. Single .exe, no installer.
Download a GGUF model
Find a GGUF-format model on Hugging Face that fits your VRAM.
Launch KoboldCpp pointing at the model
Run koboldcpp.exe, point it at your GGUF file, pick GPU offload layers, click Launch. KoboldCpp listens on http://localhost:5001/api by default.
Add to Voxta
Manage Modules → Add Modules → KoboldAI → Install.
Enter the URL (default http://localhost:5001/api), click Save.