A single self-contained executable for local story writing, chat, and roleplay with GGUF models built on llama.cpp, with the classic KoboldAI Lite interface built in.
Download koboldcpp.exeOpens the official LostRuins/koboldcpp releases page · Windows, no installation required
Maintainer
LostRuins
Cost
Free
Install
None — single .exe
Format
GGUF
The full KoboldAI Lite interface — memory, world info, author's note, characters, and scenarios.
Also runs Stable Diffusion 1.5, SDXL, and other image models from the same executable.
Speaks KoboldCpp, OpenAI, Ollama, and A1111 APIs — plugs into other tools that expect those formats.
NVIDIA users get CUDA acceleration out of the box with the right build.
Download koboldcpp.exe
The button above opens the real releases page — grab the plain koboldcpp.exe, or the CUDA-specific build if you have an NVIDIA GPU.
Get a GGUF model
Weights aren't bundled — download a GGUF-format model separately, for example from Hugging Face.
Drag the model onto the .exe
Or double-click koboldcpp.exe on its own to open a settings GUI first, pick your model, and hit Launch.
Open the local interface
By default it's available at http://localhost:5001 in your browser.
On Windows with an NVIDIA card, add the --usecublas flag (or set it in the GUI) and make sure you downloaded the CUDA-enabled .exe variant — this offloads model layers to your GPU for a large speed boost over CPU-only.