Skip to content
kckoboldcpp.dev
  • What is KoboldCppWhat it does and what it does not
  • DownloadWhich of the seven files you need
  • RequirementsWhat hardware you need
  • First modelFrom download to first reply
  • LauncherEvery tab and field explained
  • Lite UIThe built-in web interface
Most readWhich file do I download?The picker that chooses for your PC
  • WindowsInstall and first run
  • LinuxBinary, drivers and services
  • MacApple Silicon and Metal
  • AndroidRun it in Termux
  • ColabFree cloud notebook
  • Cloud GPUsRunPod, Vast and others
  • DockerContainers and environment settings
  • Build from sourceCompile it yourself
Most readInstall on WindowsOur most-read install guide
  • Best modelsWhat fits your VRAM, tested
  • QuantizationWhat Q4_K_M and friends mean
  • Hugging FaceFind and download GGUF files
  • Chat templatesTemplates and thinking models
Most readBest models by VRAMRecommendations with measured memory
  • GPU layersHow many layers fit your VRAM
  • Context sizeContext, ContextShift, KV cache
  • SpeedMake it faster
  • BackendsCUDA, Vulkan, CPU, Metal
  • NVIDIAOld and new GPUs
  • AMDVulkan and ROCm
  • Multi-GPUSplitting across cards
  • SamplersGeneration settings
  • FlagsEvery command-line option
  • .kcpps filesSaved configs and templates
Most readGPU layers and VRAMThe setting most readers get wrong
  • APIEndpoint explorer
  • OpenAI APIOpenAI-compatible endpoints
  • KoboldAI APIThe native API
  • SillyTavernExact connection settings
  • Other appsJanitor AI, Open WebUI, editors
  • Remote accessPhone or another PC
  • AI HordeShare your model, or use others
  • ImagesImage, video and music
  • VisionLet the model see images
  • VoiceWhisper and text to speech
  • EmbeddingsTextDB, web search, MCP
Most readConnect SillyTavernSettings we verified in a real run
  • TroubleshootingDecision tree for any problem
  • ErrorsEvery error string explained
  • Will not startCloses immediately or crashes
  • Out of memoryCUDA, Vulkan and Metal
  • Not using the GPUHow to tell and how to fix
  • Model will not loadUnknown architecture, bad files
  • Is it safeAntivirus, hashes, fake sites
Most readPaste your logFind the problem in seconds
Tools
Start
  • What is KoboldCpp What it does and what it does not
  • Download Which of the seven files you need
  • Requirements What hardware you need
  • First model From download to first reply
  • Launcher Every tab and field explained
  • Lite UI The built-in web interface
Install
  • Windows Install and first run
  • Linux Binary, drivers and services
  • Mac Apple Silicon and Metal
  • Android Run it in Termux
  • Colab Free cloud notebook
  • Cloud GPUs RunPod, Vast and others
  • Docker Containers and environment settings
  • Build from source Compile it yourself
Models
  • Best models What fits your VRAM, tested
  • Quantization What Q4_K_M and friends mean
  • Hugging Face Find and download GGUF files
  • Chat templates Templates and thinking models
Tune
  • GPU layers How many layers fit your VRAM
  • Context size Context, ContextShift, KV cache
  • Speed Make it faster
  • Backends CUDA, Vulkan, CPU, Metal
  • NVIDIA Old and new GPUs
  • AMD Vulkan and ROCm
  • Multi-GPU Splitting across cards
  • Samplers Generation settings
  • Flags Every command-line option
  • .kcpps files Saved configs and templates
Connect
  • API Endpoint explorer
  • OpenAI API OpenAI-compatible endpoints
  • KoboldAI API The native API
  • SillyTavern Exact connection settings
  • Other apps Janitor AI, Open WebUI, editors
  • Remote access Phone or another PC
  • AI Horde Share your model, or use others
  • Images Image, video and music
  • Vision Let the model see images
  • Voice Whisper and text to speech
  • Embeddings TextDB, web search, MCP
Fix
  • Troubleshooting Decision tree for any problem
  • Errors Every error string explained
  • Will not start Closes immediately or crashes
  • Out of memory CUDA, Vulkan and Metal
  • Not using the GPU How to tell and how to fix
  • Model will not load Unknown architecture, bad files
  • Is it safe Antivirus, hashes, fake sites
Tools
  • VRAM calculator Will it fit?
  • Config builder Commands and .kcpps
  • Log analyzer Decode a log
  • API tester Test and get snippets
Compare
  • Alternatives
  • vs Ollama
  • vs LM Studio
  • vs llama.cpp
  • vs text-generation-webui
  1. Home
  2. Home
  3. About this independent KoboldCpp guide

Last updated 13 Nov 2026

About this independent KoboldCpp guide

Not the official site

KoboldCpp is developed by LostRuins (known as Concedo) and contributors at github.com/LostRuins/koboldcpp. The project’s own website is koboldcpp.net. koboldcpp.dev is not affiliated with either. We never host KoboldCpp or any model: every download link on this site goes to the project’s GitHub releases or to the original Hugging Face page. We say this loudly because lookalike sites exist, and the project’s README warns that koboldcpp.com is malicious (how to spot fake sites).

Why this site exists

The project documents itself well, and we do not copy that. We add what an independent guide can: tested evidence for what we claim, tools that run in your browser, depth on error messages and hardware, honest comparisons, and a focus on safety. Everything we say we tested is listed with its raw data on how we test.

Who we are

The koboldcpp.dev editorial team. We do not invent people or credentials, and we do not publish a legal entity name or personal biographies on this page. If you need to reach us, see contact.

How we work

  • Every claim carries one of three labels: tested by us, upstream says it, or users report it.
  • Our method, machines and every finding are public: how we test.
  • Mistakes are fixed and logged in public: corrections.
  • Drafting uses AI assistance, and every number and instruction on a page is checked against our recorded data or the project’s own documentation before it is published.
  • We have no ties to hardware makers or to other model runners we mention, and we do not take payment for placement.

Money

There are no ads and no affiliate links at launch. If ads are ever added, they will go in reserved places and never inside our tools or on legal pages, and the privacy page will be updated first.

Contact

Email hello@koboldcpp.dev, or see the contact page for what to send. To report a page that looks wrong, use the same address. Our terms explain what this site is and is not.

Frequently asked questions

Is this the official KoboldCpp website?

No. The project's own site is koboldcpp.net, and its code and downloads are on GitHub at LostRuins/koboldcpp. koboldcpp.dev is an independent guide.

Who runs it?

The koboldcpp.dev editorial team. We do not publish personal details; what matters is on the page: our method, our data and a log of our corrections.

Is it affiliated with KoboldAI?

No. KoboldAI, KoboldCpp, KoboldAI Lite and the AI Horde are separate projects, and we are not part of any of them.

Related

  • GuideHow we testHow we test, and every finding.
  • FixIs it safeIs KoboldCpp safe, and how to spot fake sites.

Start

  • What is KoboldCpp
  • Download
  • Requirements
  • First model
  • Launcher
  • Lite UI

Guides

  • Windows
  • Linux
  • Mac
  • Android
  • Colab
  • Cloud GPUs
  • Docker
  • Build from source
  • Best models

Connect

  • API
  • OpenAI API
  • KoboldAI API
  • SillyTavern
  • Other apps
  • Remote access
  • AI Horde
  • Images
  • Vision

Fix

  • Troubleshooting
  • Errors
  • Will not start
  • Out of memory
  • Not using the GPU
  • Model will not load
  • Is it safe

Project

  • About
  • How we test
  • Corrections
  • Contact
  • Privacy
  • Terms
  • FAQ
  • KoboldAI family
  • Releases
  • hello@koboldcpp.dev

koboldcpp.dev is an independent guide. KoboldCpp is developed by LostRuins (Concedo) and contributors at github.com/LostRuins/koboldcpp. The project's own site is koboldcpp.net. We are not affiliated, and we never host KoboldCpp or models: every download link goes to GitHub.

Checked against KoboldCpp 1.124.2 · 2026-11-13

    No results. Try the troubleshooting tree or paste your log into the log analyzer.