Integration guide

Connect SillyTavern to OpenLux

SillyTavern supports OpenAI-compatible connections with a custom endpoint. Set that endpoint to OpenLux and character chats run on Claude Opus 5, GPT-6 Astra, Claude Fable 5.1 or any other text model in the catalogue.

Create an API key Read the docs

Abstract orange glass sculpture illustrating the SillyTavern integration with OpenLux — SillyTavern + OpenLux — one key, one endpoint, one bill.

Set it up in four steps

Create an OpenLux key

Generate the key in the console. One key covers every text model, so trying a different persona model costs nothing extra to set up.

Add the connection

In the API Connections panel choose Chat Completion, then OpenAI, and enable the custom endpoint option.

Set the endpoint and key

Enter the OpenLux base URL as the proxy endpoint, paste your key, then click Connect to load the model list.

Pick a model per character

Chats keep their own model selection, so long-form writing can use a different tier than quick conversations.

Configuration

Copy these blocks, replace the placeholder key and you are done.

API Connections → Chat Completion → OpenAI

Custom Endpoint: enabled
Endpoint:        https://openlux.lol/v1
Model:           claude-opus-5
Available:       gpt-6-astra, claude-fable-5-1, gemini-3.1-pro-preview, deepseek-v4

config.yaml (reverse proxy alternative)

openai:
  reverse_proxy: https://openlux.lol/v1
  proxy_password: sk-openlux-...

Models that fit SillyTavern

Claude Opus 5

Anthropic · $0.4412 / 1M tokens

GPT-6 Astra

OpenAI · $0.3677 / 1M tokens

Claude Fable 5.1

Anthropic · $1.765 / 1M tokens

Gemini 3.1 Pro (preview)

Google · $0.1471 / 1M tokens

The full list is in the model catalogue and the use-case collections.

SillyTavern + OpenLux questions

Does SillyTavern work with OpenAI-compatible gateways?

Yes. Enable the custom endpoint on the OpenAI connection and SillyTavern sends its chat-completion requests there instead of to OpenAI directly.

Which model fits long roleplay sessions?

Claude Opus 5 and Claude Fable 5.1 are the strongest for long, character-consistent writing; Gemini 3.1 Pro and DeepSeek V4 keep long sessions affordable.

Do prompts train on my data?

OpenLux is a pass-through gateway: requests go to the upstream model provider under its policy. There is no separate OpenLux training on your conversations, and prompts are not stored as content.

Can I switch models mid-session?

Yes. Change the model in the connection settings — the conversation continues, only the backend model changes, and cost updates per token.