KoboldCpp Download: One-File Local AI for Chat and Stories
KoboldCpp · for Windows, Linux and macOS
Developer: LostRuins (Concedo) (opens in a new tab)
- Windows
- Linux
- macOS
The download link points to the program’s official source; the file is not hosted on this site. Link address: github.com
Link not working? Let us know
KoboldCpp is a free, open-source, single-file program for running GGUF language models on your own computer, with the bundled KoboldAI Lite web interface and image, video, speech and music generation. It runs on Windows, Linux and macOS and is AGPL-3.0 licensed.
KoboldCpp packs a whole local AI studio into one executable with no installation: text chat and stories, image and video generation, speech recognition and text-to-speech. It runs on the CPU or GPU, with full or partial offloading. Download it only from the official GitHub releases or koboldcpp.net; the project warns that koboldcpp.com is a malicious fake site.
What does KoboldCpp do?
- Text: Chat, adventure, instruct and story modes, with image understanding for compatible vision models.
- KoboldAI Lite UI: Memory, world info, characters and saved stories for writing and roleplay.
- Images and video: Stable Diffusion 1.5, SDXL, SD3 and Flux for images; WAN 2.2 and others for video.
- Speech: Whisper speech-to-text and several text-to-speech models, plus music generation.
- Tools: MCP support, tool calling, web search and simple RAG.
- Compatible APIs: KoboldAI, OpenAI, Anthropic, Ollama, A1111/Forge, ComfyUI and XTTS.
Who uses it?
It suits writers, roleplayers and tinkerers who want an all-in-one local AI app without installing anything.
Which file to download
The Download button opens the official releases page. Models are not included; get a GGUF model separately.
- Windows: koboldcpp.exe (with CUDA), koboldcpp-nocuda.exe (smaller, CPU and Vulkan) or koboldcpp-oldpc.exe for older hardware.
- Linux: koboldcpp-linux-x64, plus nocuda and oldpc variants.
- macOS: koboldcpp-mac-arm64 for Apple Silicon.
On Windows, double-click it to open the launcher, or start it with a model:
koboldcpp.exe --model "C:\Models\model.gguf"
On Linux:
chmod +x koboldcpp-linux-x64
./koboldcpp-linux-x64 --model /path/to/model.gguf
Then open http://localhost:5001 in your browser.
Hardware requirements
- GPU: Optional. NVIDIA via CUDA, AMD/NVIDIA/Intel via Vulkan, Apple via Metal; ROCm with a source build. Use --gpulayers to control how much of the model goes to the GPU.
- Memory: Depends on the model size and context length; start with a smaller model if unsure.
- Older CPUs: Use the oldpc build or the --noavx2 option.
License
KoboldCpp and KoboldAI Lite are free and open source under the AGPL-3.0 license; bundled components such as llama.cpp keep their own (MIT) licenses.
Pros
- Single file, no installation
- Text, images, speech and music in one app
- Compatible with many popular AI APIs
Cons
- Many options can confuse beginners
- Intel Macs need a source build
Frequently Asked Questions
Where is the official KoboldCpp download?
Only from the project’s GitHub releases page or koboldcpp.net. The developers warn that koboldcpp.com is a fake site that is not affiliated with the project.
Do I need to install KoboldCpp?
No. It is a single executable: download it, then run it with a GGUF model. On Windows, running it without arguments opens a launcher window.
KoboldCppOpen source
Download (opens in a new tab)
