KoboldAI Lite is the web interface built into KoboldCpp. After you click Launch, it opens at http://localhost:5001. You need nothing else to chat, write stories or play text adventures.
Start chatting
Section titled “Start chatting”- Launch KoboldCpp with a model. With Launch Browser ticked (the default in the launcher), Lite opens by itself. Otherwise, open
http://localhost:5001. - Type into the input box at the bottom and send.
- To change how Lite behaves, click Settings in the top menu.
Pick the mode in Settings > General > Usage mode.
| Mode | Use it for |
|---|---|
| Instruct Mode | Giving the AI tasks, ChatGPT-style. |
| Chat Mode | Conversations with a character. |
| Story Mode | Novel-style writing, where you and the AI write the same text. |
| Adventure Mode | Interactive fiction and text RPGs. |
UI styles
Section titled “UI styles”Next to the mode, UI style changes how the text is shown. Some styles are only available in some modes.
| Style | Look |
|---|---|
| Classic Theme | Plain text editor, for writing. |
| Messenger Theme | Chat bubbles. |
| Aesthetic Theme | Roleplay layout. The button next to it opens more display options. |
| Corpo Theme | Assistant-style layout. |
Where the settings are
Section titled “Where the settings are”The launcher decides what the server can do: which model, how much context memory is reserved, which features are loaded. Lite decides what it sends with each request.
Settings has these tabs:
| Tab | What you find there |
|---|---|
| General | Usage mode, UI style, your name and the AI's name. |
| GUI | Appearance. |
| Samplers | Sampler Preset, Context Size, Max Output, Temperature and the other samplers. |
| Media | Image generation, text-to-speech (Audio Output) and voice input (Audio Input). |
| Tokens | Token settings. |
| Tools | MCP tool calling and web search. |
| Misc | Save/load, advanced and experimental settings. |
Context Size in Lite covers the text Lite sends plus the reply (Max Output). The server cuts it down to the context it reserved at launch. If the server reserved 16384 or more, Lite raises its own value to match. A value you lowered yourself stays until the server's context changes. See Context size.
The other top menu items:
- New Session starts over.
- Scenarios loads ready-made stories and characters.
- Save / Load saves and loads your story as a JSON file.
- Context (above the input box) holds memory, world info and the author's note.
- Add File (above the input box) adds images or audio, or generates an image.
Other bundled interfaces
Section titled “Other bundled interfaces”| Address | Interface | Available when |
|---|---|---|
/ | KoboldAI Lite | Always. |
/sdui | StableUI, for images | Always. Generating images needs an image model. See Image generation. |
/musicui | MusicUI, for music and speech | Always. Generating needs a music or TTS model. See Music. |
/lcpp/ | The llama.cpp web UI | Always. It can connect to MCP servers. |
/noscript | A plain page without JavaScript, with story, chat and image modes | Always. For old or restricted browsers. |
/api | The API documentation | Always. |
If you load only an image model, Lite offers to open StableUI. If you load only a music or TTS model, it offers MusicUI.
Lite without KoboldCpp
Section titled “Lite without KoboldCpp”Lite also runs at lite.koboldai.net. That copy can connect to your KoboldCpp, to the AI Horde, or to online AI services.
To use the bundled Lite without a local model, tick Allow Launch Without Models on the Loaded Files tab (--nomodel). Lite then asks whether to connect to an external service.
Advanced
Section titled “Advanced”- TextDB (Context > TextDB) is Lite's document lookup (RAG). Local TextDB searches in the browser. KoboldCpp Embeddings uses the loaded embeddings model.
- Lite loads Tavern character cards, and imports SillyTavern chat exports (JSONL) and several other save formats.
- In Settings > Samplers, Sampler Preset switches between curated sampler sets.