简体中文
EnglishWhat Is the Difference Between KoboldCPP, KoboldAI Lite, and KoboldAI.net?
Updated
KoboldAI is not one program but a set of components that split the job of running a local AI into two parts: a backend that does the computation and a frontend that you type into. KoboldCPP is the backend (a local LLM API server), KoboldAI Lite is the frontend (a lightweight interface for talking to any compatible API endpoint), and KoboldAI.net is a hosted web version of that same frontend. You can mix them: run KoboldCPP on your own machine and open KoboldAI Lite in a browser, or skip the local install entirely and use KoboldAI.net.
The three components at a glance
| Component | Role | Where it runs | What it gives you |
|---|---|---|---|
| KoboldCPP | Local LLM API server / backend | Your own hardware (Windows, Linux, Mac) | Loads models and serves an API endpoint that frontends connect to |
| KoboldAI Lite | Lightweight user interface / frontend | Your browser | The chat/story interface you actually type in; connects to an API endpoint |
| KoboldAI.net | Hosted web service | The web | KoboldAI Lite delivered as a free web service, with the same flexibility as running it locally |
How they fit together
The relationship is a chain: model → KoboldCPP (serves it as an API) → KoboldAI Lite (talks to that API) → you type in the interface.
- KoboldCPP is described as "our local LLM API server for driving your backend." It is the piece that actually runs the language model on your machine and exposes it as an API endpoint.
- KoboldAI Lite is "our lightweight user-friendly interface for accessing your AI API endpoints." It does not run models itself — it is the window you interact through, and it can point at any compatible endpoint.
- KoboldAI.net is "where we deliver KoboldAI Lite as web service for free with the same flexibilities as running it locally." In other words, it is the Lite interface hosted for you, so you do not have to install or serve the frontend yourself.
Which one do you actually need?
The answer depends on whether you want to run models on your own hardware.
If you want to run models locally
You need KoboldCPP (the backend) plus a frontend. You can use KoboldAI Lite in your browser as that frontend, pointing it at your local KoboldCPP endpoint. This is the fully local path — your hardware does the work.
If you just want to use the interface
Use KoboldAI.net. It serves KoboldAI Lite as a web service, so you get the same interface without installing or hosting it yourself. You still need an API endpoint for it to connect to, since Lite is only the interface.
If you want the interface but not the hosted site
Run KoboldAI Lite yourself and connect it to whatever API endpoint you have — local (KoboldCPP) or otherwise.
Getting started
The official site points to a few concrete entry points:
- Download KoboldCPP: all versions and download links for Windows, Linux, and Mac are on the KoboldCPP official downloads page.
- Guides: the site lists a KoboldAI Quickstart, supported models, AI safety notes, the KoboldCpp API, administration, audio, hardware options, Horde Worker, image generation, networking, Google Colab, and Runpod.
- Community: there is a Discord server for support, discussions, and sharing.
Common points of confusion
- KoboldAI Lite is not a model runner. It is an interface. If nothing is serving an API endpoint, Lite has nothing to talk to.
- KoboldAI.net is not a separate product from Lite. It is Lite delivered as a hosted web service.
- KoboldCPP is the part that touches your hardware. Model loading, GPU/CPU use, and the API all live here, which is why the download and hardware guides center on it.