What is SillyTavern?
SillyTavern is a power-user chat frontend for roleplay and long-form conversations with many model APIs.
This SillyTavern setup follows its Custom OpenAI-compatible Chat Completion flow. Use the API root ending in /v1, supply the exact model ID when discovery is unavailable, and use Test Message before a long conversation. SillyTavern is a frontend rather than a bundled model service, and long roleplay sessions can accumulate substantial prompt and history tokens. Confirm Custom Chat Completion with Test Message, enter the model manually if discovery fails, and review context limits and usage before starting a long conversation.
TokenHub belongs under SillyTavern's Custom (OpenAI-compatible) Chat Completions source, not Text Completion or a preset for another provider. SillyTavern warns that a custom endpoint is not guaranteed to work with every feature, so verify it with Test Message.
Follow SillyTavern's official Connecting sequence
- Open API Connections.
- Change API type to Chat Completion.
- Choose
Custom (OpenAI-compatible)for Chat Completion Source. - Enter
https://us-api.tokenhub.com/v1for Custom endpoint URL. - Enter your TokenHub API key.
- If TokenHub's
/v1/modelsreturns a list, select the model from the menu. Otherwise enter the complete model ID in the text field. - Click Test Message and send a minimal prompt.
The official guide permits a URL ending in /v1 but says not to append /chat/completions, because SillyTavern adds that path itself.
Use Bypass API status check only after a real request works
If Test Message and an actual chat both return normally but SillyTavern still reports that the endpoint is unavailable, enable Bypass API status check. This suppresses the status warning; it does not fix a 401, 404, timeout, or invalid model. Do not use it to hide a failed request.
Treat Prompt Post-Processing as a SillyTavern-specific control
If an endpoint requires adjacent roles to be merged, a single system message, or a user message first, move through the documented levels from least to most restrictive: None, Merge consecutive messages, Semi-strict, Strict, and Single user message.
For tool calling, do not choose a no tools variant. The no-tools forms of Merge, Semi-strict, and Strict remove tool calls from the prompt. Start with None and basic Chat Completions; tighten processing only after a clear role-format error.
Troubleshoot the fields in SillyTavern's form
- Empty model menu:
/modelsdid not return a usable list, so enter the model ID manually. - 404: make sure the endpoint ends at
/v1. - Test Message succeeds but a status warning remains: only then consider Bypass API status check.
- Tools suddenly disappear: check for a no-tools Prompt Post-Processing option.
- Role-message format error: try Merge, Semi-strict, and Strict one level at a time instead of collapsing everything into one message.
Official reference
Choose a compatible SillyTavern model
SillyTavern custom API FAQ
Which API type should I select?
Choose Chat Completion, then select Custom (OpenAI-compatible) as the source.
What if SillyTavern cannot list my TokenHub model?
Enter the exact model ID manually. The custom endpoint can work even when a /v1/models listing is unavailable.
Should I turn on Bypass API status check?
Only when Test Message works but the status check continues to warn incorrectly. It does not repair a non-working endpoint.
Why do long SillyTavern chats become slow or fail?
Character prompts and chat history consume context continuously. Review the model context limit and token usage, then summarize or trim history when needed.
How do I verify a SillyTavern custom endpoint?
Run Test Message, confirm the selected model remains active, and check for a Chat Completions request in TokenHub logs before starting a long chat.
References
This setup is based on TokenHub and official documentation, last verified on 2026-09-08.
