Skip to content

How-ToGPU

How to Run Local AI Models with KoboldCpp on Windows

How to run local GGUF models with KoboldCpp on Windows: download koboldcpp.exe, the nocuda build or the oldpc build from the official site, allow the unsigned official executable through SmartScreen only if it came from that source, and launch with --model and --launch. Official docs also note about 1GB of temporary disk space.

By , Editorial team

3 min read

How to Run Local AI Models with KoboldCpp on Windows

KoboldCpp is a lightweight, self-contained way to run local language models on a Windows PC. There is no installer: you download one official .exe, point it at a GGUF model file, and it opens a chat in your browser. It is a good fit if you want something smaller than LM Studio or Ollama, or if you already have GGUF files on disk from those apps.

Forge PC has not tested KoboldCpp on our own hardware. The steps below follow the official KoboldCpp documentation for Windows, read on 14 November 2026.

1. Download the right official file

Get the build from the official GitHub releases (the documentation at koboldcpp.net links there). Put it in a folder of its own, for example C:\koboldcpp, using a path with plain letters and no accented characters. The docs say models it downloads will be saved in that folder unless you change Download Dir on the Loaded Files tab.

Pick the file that matches the PC:

Your PC

Official file

NVIDIA graphics card

koboldcpp.exe

AMD, Intel or no discrete GPU

koboldcpp-nocuda.exe

Older CPU or older NVIDIA card

koboldcpp-oldpc.exe

2. Deal with SmartScreen, then open the launcher

Double-click the file. A console window should open, then the launcher. The official Windows guide says the exe is not code-signed, so Windows protected your PC is expected: click More info, then Run anyway. Only do that if the file came from the official GitHub releases, not from a random mirror.

If the window opens and immediately closes, the docs tell you to run it from Command Prompt instead, so the error stays on screen:

cd C:\koboldcpp
koboldcpp.exe

Common official causes include an older CPU that needs koboldcpp-oldpc.exe, a damaged file (sometimes after antivirus scanning), or not enough free space — KoboldCpp says it needs about 1GB of temporary space on the system drive to unpack itself.

3. Load a GGUF model

You need a model file in GGUF format. Use a model you already trust, for example one you have previously downloaded in LM Studio or Ollama and exported, or a file from the model's official Hugging Face page. Our quantization guide and LM Studio model guide explain how to pick a size that fits your graphics card. Do not download weights from an unmarked forum post.

In the launcher, set GGUF Text Model to that file and click Launch. When it has loaded, the official getting-started flow opens a browser on the local interface. If you start from the command line instead, the docs say you should add --launch if you want the browser to open, for example:

koboldcpp.exe --model C:\models\mymodel.gguf --launch

koboldcpp.exe --help lists every flag. You can also save a configuration and reload it with --config.

If you would rather use a more guided app for the first model you ever run, start with our Ollama or LM Studio guide and come back to KoboldCpp when you want a single-exe option.

4. Chat, then connect other tools if you need to

Use the browser UI to chat with the model. You can leave KoboldCpp running and point compatible tools at its local HTTP API, in the same spirit as Ollama's OpenAI-compatible API or LM Studio's server. Check the official flag reference for the port and API options your version exposes, rather than copying an old blog post.

For a more polished chat-only experience, our best free local AI apps guide lists LM Studio and Jan as the usual starting points. KoboldCpp is the one to reach for when you want llama.cpp-style control in a single file.

Our verdict

Download the official Windows exe that matches your GPU, put it in its own folder, dismiss SmartScreen only if the file came from the official release, and load a GGUF model you already trust. NVIDIA users want koboldcpp.exe; AMD and Intel users want koboldcpp-nocuda.exe. Click Launch and chat in the browser, or start it with --model and --launch from a command line. Forge PC has not tested KoboldCpp; this guide follows the official Windows installation documentation.

The bottom line

Download the official Windows exe that matches your GPU, put it in its own folder, dismiss SmartScreen only if the file came from the official release, and load a GGUF model you already trust. NVIDIA users want `koboldcpp.exe`; AMD and Intel users want `koboldcpp-nocuda.exe`. Click Launch and chat in the browser, or start it with `--model` and `--launch` from a command line. Forge PC has not tested KoboldCpp; this guide follows the official Windows installation documentation.

Sources

  1. 01KoboldCpp documentation, Install on Windows (SmartScreen, antivirus)
  2. 02KoboldCpp documentation, Start KoboldCpp for the first time
  3. 03KoboldCpp documentation, Release files and system requirements
  4. 04LostRuins/koboldcpp, official GitHub repository

About the author

Forge PC Staff

Editorial team

The shared byline for articles researched, written and edited by the Forge PC editorial team. How we source, check and correct our work is set out in our editorial policy.

More from Forge PC Staff

Read next

How to Fix Games Crashing on PC: A Step-by-Step Checklist for Windows 11

How-To

How to Fix Games Crashing on PC: A Step-by-Step Checklist for Windows 11

Most PC game crashes come from drivers, corrupted files or unstable overclocks. Update, roll back or clean-install the graphics driver, verify the game's files, run DISM and SFC, and reset GPU and memory tuning to stock. If crashes continue, check temperatures, test RAM with MemTest86 and look at the power supply.

Forge PC Staff3 min

KoboldCpp vs LM Studio: Which Local AI App Should You Use?

Versus

KoboldCpp vs LM Studio: Which Local AI App Should You Use?

KoboldCpp is a single, install-free executable built on llama.cpp, with CUDA and Vulkan builds plus image, speech and agent features and a story-writing UI. LM Studio is a polished app that finds and downloads models for you and serves them over an OpenAI-like API. LM Studio suits most people; KoboldCpp suits tinkerers and writers.

Forge PC Staff3 min

How to Use Radeon Chill: Cut Power, Heat and Fan Noise on AMD Graphics Cards

How-To

How to Use Radeon Chill: Cut Power, Heat and Fan Noise on AMD Graphics Cards

Radeon Chill lowers your frame rate when there's little movement in a game and ramps it up when the action starts, saving power and heat. Turn it on in AMD Software's Global Graphics, set Min and Max FPS, override per game and toggle it with Alt+F11. It disables Anti-Lag and Boost and doesn't work with UWP games.

Forge PC Staff3 min

How to Inpaint in ComfyUI: Fix Part of an AI Image Without Starting Again

How-To

How to Inpaint in ComfyUI: Fix Part of an AI Image Without Starting Again

Inpainting repaints only the masked part of an image. In ComfyUI, load the official inpainting workflow, use an inpainting checkpoint such as 512-inpainting-ema, draw a mask in the Mask Editor, describe the change and queue it. Raise grow_mask_by in VAE Encode (for Inpainting) if you can see seams, and try a few seeds.

Forge PC Staff3 min