Where Does KoboldCpp Store Models?
KoboldCpp is a single portable exe with no install. GGUF models live wherever you download them, picked at launch.
Last updated
KoboldCpp installs nothing. One executable (plus GPU variants) runs the launcher GUI, the server, and the browser UI on localhost port 5001. Models ship separately as GGUF files you fetch from Hugging Face or similar repos.
That split defines the file story. The program folder holds the binaries. Your model library is any folder you choose, selected per launch through Browse or a model flag. Stories and settings persist through the browser UI rather than a config file. Image models load through the same picker when enabled.
Where KoboldCpp stores this, by platform
C:\Tools\KoboldCpp\koboldcpp.exe
Portable program with no installer. GPU builds add matching library files beside the exe.
D:\Models\model-Q4_K_M.gguf
Your model library, anywhere you like. Multigigabyte GGUF files belong on a roomy drive, not next to the OS.
/home/<username>/Models/model-Q4_K_M.gguf
Same portable layout on Mac and Linux. Download the matching build and point it at the same model folder.
Frequently asked questions
Where do I get a model for KoboldCpp?
Grab any GGUF file from Hugging Face (Bartowski repacks are beginner friendly). In the launcher click Browse next to Model, select the file, then Launch.
How do I move KoboldCpp to another PC?
Put the exe and the gguf in the same folder and copy both. Settings ride in the launcher, so the pair moves as a unit to any machine.
Notice an outdated path? Let us know.