SillyTavern gives you a lot of control over characters, lore, prompts, and formatting. That freedom also means the best SillyTavern proxy is personal. A service that is perfect for a short chat may not be the one you want for a slow-burn campaign with a large lorebook.

Composite accepts the OpenAI-compatible format SillyTavern supports and adds model selection, long-context handling, reusable prompts, lorebooks, and workflow plugins. SillyTavern remains your workspace; Composite sits behind the connection.

What is a SillyTavern proxy?

In this context, a proxy or provider is the API service SillyTavern calls to generate replies. SillyTavern's documentation describes a Custom (OpenAI-compatible) Chat Completion source for services that expose the generic OpenAI API schema. That matters because it lets you choose a provider based on model access, price, and roleplay tools instead of being locked to one built-in option.

How to choose the best SillyTavern proxy

Start with compatibility. A provider should support chat completions, API-key authentication, streaming if you use it, and a model list that SillyTavern can read or a model ID you can enter manually. SillyTavern recommends using the Custom OpenAI-compatible source for this kind of endpoint and notes that a connection URL may need a /v1 suffix. Composite supplies that exact base URL.

  • Model choice: Use different models for different genres and budgets instead of forcing every character through one default.
  • Context support: Long chats need space for character cards, lore, recent messages, and the response itself.
  • Prompt control: Check whether you can preserve your preferred formatting and avoid a provider that rewrites every request in ways your setup does not expect.
  • Cost visibility: Read model pricing and set limits before a long session. Cheap input is useful, but output and context length also affect the bill.
  • Connection health: Choose a service with clear model availability and test tools so you can separate an endpoint issue from a model issue.

Composite vs. NanoGPT, OpenRouter, LiteRouter, and VoidAI

All five services can work with an OpenAI-compatible client, so the decision is less about basic connectivity and more about the layer around the model. Prices, catalogues, limits, and availability change often; verify those details on each provider's site before loading a long conversation.

ProviderPublished focusWhat that means for SillyTavern
CompositeRoleplay configuration around the API key: prompts, lorebooks, workflow plugins, and context-management choices.Use it when you want the provider layer to preserve a repeatable story setup across chats.
NanoGPTA multi-model service with an API that documents basic OpenAI compatibility.Worth testing if you want another model catalogue; confirm the selected model's context and request behavior with your character card.
OpenRouterA broad model marketplace with model metadata, parameter support, and routing options.Useful when model-by-model choice is the priority and you are comfortable managing roleplay prompts in SillyTavern itself.
LiteRouterAn OpenAI-compatible proxy that advertises multi-provider routing, fallback, and context summarization.Consider it when automatic routing is attractive; test its summarization and fallback behavior against your formatting expectations.
VoidAIA unified OpenAI-compatible gateway that also documents image, audio, video, embedding, and moderation endpoints.Potentially handy if one account needs several AI capabilities; for text roleplay, compare the chat model, price, and context limits first.

For a fair test, send the same character card, lorebook, opening, generation settings, and model where possible. Judge the result after several turns, not just the first reply.

Why use Composite with SillyTavern?

Composite lets you create a key for a specific setup and use the models allowed by that key. You can attach reusable prompts, lorebooks, and plugins without rebuilding them in every new chat. For established stories, Composite's context-management options help keep important details in play as conversations grow. It works with your own provider key at no charge from Composite, and Synth, its free model, lets you try a setup before connecting a provider or buying credits.

Model choice changes with the character, genre, and budget. Start with a scene you know well, compare a few candidates with the same temperature and prompt settings, then decide from several replies instead of someone else's favorite model name.

Connect Composite to SillyTavern

  1. Create a Composite account and API key.
  2. Open SillyTavern's API Connections panel.
  3. Choose Chat Completion, then choose Custom (OpenAI-compatible) as the source.
  4. Set the endpoint or base URL to https://composite.lucidity.sh/v1.
  5. Paste your Composite API key, select a model, and use Test Message to confirm the connection.

Do not append /chat/completions in SillyTavern when it asks for the base URL. SillyTavern's own connection guide makes the same distinction for custom endpoints. For the current Composite values, troubleshooting, and screenshots-free setup notes, see the SillyTavern proxy guide and the SillyTavern Custom OpenAI-compatible guide.

Fixing the usual connection mistakes

An invalid-key message means the credential needs attention. A missing model list may mean the endpoint is not set to /v1, or that you should type an allowed model ID manually. If replies arrive but the formatting looks wrong, inspect SillyTavern's prompt post-processing setting. Its documentation explains that restrictive modes can merge or alter messages for endpoints with stricter requirements. Composite uses the standard chat-completions format, so begin with the least restrictive setting that works for your character setup.

Our recommendation

Choose Composite when you want a simple OpenAI-compatible setup and want prompts, lorebooks, plugins, and context options attached to the connection. Make one API key, test it with a familiar character, then keep the model and settings that suit the way you write.