Compositev12
forum App guide

Use Composite as your SillyTavern proxy

SillyTavern supports any OpenAI-compatible provider. Point it at Composite's OpenAI-compatible base URL to run community presets, lorebooks and plugins on your own provider key, or on Synth, LuciditySH's creative model, free on Composite, with no key at all, in your existing chats. Using ChubAI instead? See the dedicated ChubAI proxy guide.

What "SillyTavern proxy" means

SillyTavern does not run its own models. Under Chat Completion Source you point it at an OpenAI-compatible API, and every message you send is forwarded there and the reply is shown in your chat. That forwarding service is what people mean by a "proxy" or "reverse proxy" for SillyTavern. It sits between the frontend you use to write and the model actually generating replies. Composite is one such proxy: it authenticates your API key, applies the presets, lorebooks and plugins attached to it, sends the request to your own provider or to the model you picked, and returns a standard OpenAI-formatted response. One Composite key reaches every model on the platform, so there is no separate setup to repeat when you switch models mid-story. The same key and base URL also work in ChubAI, Agnai, and RisuAI, if you use more than one frontend.

Paste values

Copy these directly into the API settings panel in SillyTavern.

forum

SillyTavern

Use these fields in the API settings panel. Choose OpenAI as the connection type.

API TypeOpenAI
Base URLhttps://composite.lucidity.sh/v1
API KeyYOUR_API_KEY
Modelminimax/minimax-m3

Setup steps

forumSillyTavern
Configure the API connection in SillyTavern settings.
expand_more
  1. Open the API settings panel (plug icon in the top bar).
  2. Set API to Chat Completion and Chat Completion Source to OpenAI.
  3. Paste https://composite.lucidity.sh/v1 into the Custom Endpoint (Base URL) field.
  4. Paste your Composite API key into the API Key field.
  5. Click Connect, then select a model from the dropdown or type a model ID from the Models page.
Settings Reference
API:                   Chat Completion
Chat Completion Source: OpenAI
Custom Endpoint:        https://composite.lucidity.sh/v1
API Key:                YOUR_API_KEY
Model:                  minimax/minimax-m3

Using ChubAI, Agnai, or RisuAI instead? Each has its own dedicated setup guide: ChubAI, Agnai, RisuAI.

API Key

You do not need a credit card to get a working SillyTavern connection. Create a Composite account and make a key on the Keys page. Bind it to your own OpenRouter, DeepSeek, Anthropic, OpenAI, Gemini or Mistral key on the API Keys page to use that provider's models at no charge from Composite (the BYOK quickstart shows how), or leave it unbound and use Synth, lucidityai/synth-2.5-flash, which is free on every account. Paste the key into the API Key field alongside the base URL above. Community presets and lorebooks you apply on Composite follow the key into SillyTavern.

Voice (TTS)

SillyTavern's TTS extension can read replies aloud through Composite. Open Extensions, then TTS, set the provider to OpenAI Compatible, and fill in the fields below. List the voices you want in Available Voices, then use the voice map to give each character one.

TTS settings
Provider:          OpenAI Compatible
Provider Endpoint: https://composite.lucidity.sh/v1/audio/speech
API Key:           YOUR_API_KEY
Model:             mistralai/voxtral-mini-tts-2603
Available Voices:  gb_jane_neutral,gb_oliver_neutral,en_paul_neutral

/v1/audio/models lists every speech model with its voices and per-character price. On a key with credits, speech is charged per character of input at the listed price times the listed multiplier; a plan draws the listed price from its pool. Models priced per second or per token of audio are not offered. A key bound to your own provider speaks on that provider, unbilled by Composite, if it serves OpenAI-style speech (OpenRouter and OpenAI do). Anthropic has no speech API.

Troubleshooting

The most common problems when using Composite as a SillyTavern proxy, and how to fix them.

link_offSillyTavern proxy not working
The connection fails or no models appear in the dropdown.
expand_more

Confirm Chat Completion Source is set to OpenAI, the Custom Endpoint (Base URL) is exactly https://composite.lucidity.sh/v1 with nothing appended after it, and the API key was pasted with no leading or trailing spaces. If the model dropdown stays empty after Connect, the endpoint is the most likely cause: SillyTavern needs the base URL here, not the full chat-completions path. Use Test Message to confirm the connection before trusting a longer chat to it.

speed429 errors and rate limits
What a 429 means on Synth and on your own provider.
expand_more

A 429 response on Synth means today's Synth requests are used up: 80 a day, or 300 a day in your first week after joining with a referral code. They reset once a day at the time shown on your account page, and credits or a plan make Synth unlimited. A key bound to your own provider has no Composite daily limit, so a 429 there is your provider's own rate limit.

key_offInvalid API key error
Key format, the Bearer header, and key scoping.
expand_more

Composite API keys start with cmp- and are sent as a bearer token, so the API Key field should hold only the key itself. SillyTavern adds the Bearer header for you. An invalid-key error usually means the key was revoked, mistyped, or deleted; generate a fresh one on the API Keys page. Each key can also be scoped to specific models, so a key that works for one model can still fail on another it is not allowed to use.

auto_awesomeWhich model should I use?
A short starting point for roleplay on SillyTavern.
expand_more
Use caseSuggestion
General roleplay, default choiceminimax/minimax-m3
No provider key yetlucidityai/synth-2.5-flash (Synth), free on every account
Long-running character or campaignA higher context-length model from the Models page, paired with a memory-layer setting on the key
Highest reply quality, budget flexibleCompare current premium models on the Models page

The catalog changes as models are added, so treat the Models page as the source of truth rather than a fixed list.

historyContext window on long roleplays
What happens when a chat outgrows the model's context.
expand_more

Every model has a fixed context window. Once a conversation exceeds it, Composite trims older messages to keep the request within limits, so long-running characters can start losing early details unless the key is set to one of Composite's memory-layer modes, which summarize and retrieve older history instead of dropping it outright. If replies start ignoring established facts in a long chat, check the key's memory-layer setting before assuming the model itself is at fault.

Still stuck after working through the above? The full docs cover account setup, API keys, and model access in more detail, and the feedback page reaches Composite directly if something looks broken rather than misconfigured.