Connect SillyTavern and other apps

KoboldCpp has its own chat interface, KoboldAI Lite, at http://localhost:5001. You do not need another app. Connect one only if you prefer its interface or features.

Which address to use

Start KoboldCpp with a model first. Then enter the address that matches the connection type you pick in the app:

Connection type in the app	Address
KoboldAI or KoboldCpp	http://localhost:5001
OpenAI-compatible	http://localhost:5001/v1
Ollama	http://localhost:11434 (start KoboldCpp on port 11434, see below)

• If an app asks for the full KoboldAI API address, use http://localhost:5001/api.
• localhost only works on the PC that runs KoboldCpp. For a phone or another PC, see Remote access.
• If you changed Port: on the Network tab (--port), use your port in place of 5001.

When KoboldCpp is ready, the console shows the addresses:

Console

Starting Kobold API on port 5001 at http://localhost:5001/api/
Starting OpenAI Compatible API on port 5001 at http://localhost:5001/v1/
Starting llama.cpp secondary WebUI at http://localhost:5001/lcpp/

API key

• No password set: if the app requires an API key, enter any text.
• Password set: enter your password as the API key. See Passwords and security.

Model name

OpenAI-style apps often ask for a model name. KoboldCpp ignores it and always uses the model it has loaded, so any name works. http://localhost:5001/v1/models lists the loaded model.

The exception is router mode, where the model name selects which model to load. See Admin mode.

Reply length

Set the response length (max tokens) in the app. If the app does not send one, KoboldCpp uses Default Gen Amt: on the Context tab (--defaultgenamt, default 2048).

SillyTavern

SillyTavern has its own connection type for KoboldCpp:

1. In SillyTavern, click API Connections (the plug icon in the top bar).
2. Set API to Text Completion and API Type to KoboldCpp.
3. Enter http://localhost:5001 in API URL.
4. If you set a password, enter it in koboldcpp API key (optional).
5. Click Connect.

KoboldCpp understands the SillyTavern-specific sampler fields, such as banned strings.

SillyTavern can also use KoboldCpp for images when an image model is loaded, through the A1111/Forge or ComfyUI APIs. Use the same http://localhost:5001 address. See Image generation.

Ollama apps

Apps that only speak the Ollama API look for it on port 11434.

1. On the Network tab, set Port: to 11434. On the command line, add --port 11434.
2. Launch. The console prints Ollama Emulation is now available at port 11434.
3. In the app, use http://localhost:11434.

If Ollama itself is running, it already uses port 11434 and KoboldCpp cannot start on it. Close Ollama first.

Anthropic-API apps

KoboldCpp serves the Anthropic Messages API at /v1/messages. With a password, the app must send it as Authorization: Bearer <password>. KoboldCpp does not accept the x-api-key header that Anthropic clients normally use. Configure such apps to send a Bearer token. If an app can't, run KoboldCpp without a password only where no one else can reach it, for example with Host: set to 127.0.0.1.

Other apps and tools

• KoboldAI Lite online: the hosted copy at lite.koboldai.net can connect to your local KoboldCpp.
• KoboldAI United and KoboldAI Client: enter the KoboldAI API address.
• Bundled interfaces: besides KoboldAI Lite, KoboldCpp serves the llama.cpp WebUI at /lcpp/, the image interface at /sdui and the music and speech interface at /musicui. The last two only generate when the matching model is loaded. See KoboldAI Lite.
• KoboldCpp Agent: the bundled coding agent connects to http://127.0.0.1:5001/v1 by default. See Agent.
• GPTLocalhost: a Microsoft Word add-in that uses KoboldCpp.
• Speech-to-text clients: KoboldCpp accepts WAV, MP3 and FLAC audio. WebM recordings are not supported. See Speech-to-text.

Go deeper

• API overview
• Passwords and security
• Remote access
• Streaming and multiple users