SillyTavern with a cheaper API: connect UberLLM in one minute
In SillyTavern open API Connections (the plug icon), pick Chat Completion, set Chat Completion Source to Custom (OpenAI-compatible), enter https://api.uberllm.dev/v1 as the base URL and your UberLLM key, type a model id, and press Connect.
Steps
- Make a key on your keys page and add credit (a few dollars lasts a long time on the models below).
- In SillyTavern, click the plug icon at the top (its tooltip says API Connections).
- Set API to Chat Completion.
- Set Chat Completion Source to Custom (OpenAI-compatible).
- Fill in:
- Custom Endpoint (Base URL):
https://api.uberllm.dev/v1 - Custom API Key: your
ull_…key - Enter a Model ID:
deepseek/deepseek-v4.1-flash-uncensored
- Custom Endpoint (Base URL):
- Press Connect. The light turns green and the Available Models list fills in, so you can switch models from there later.
- Optional: press Test Message to check. It is a real request, so it costs a fraction of a cent.
Do not add /chat/completions to the URL. SillyTavern adds it for you; the field's own hint says so.
Settings that matter for roleplay
- Prompt Post-Processing: leave it on None. The "no tools" options turn off tool calling, and the "merge" options are only needed for models that refuse two messages in a row from the same role.
- Streaming: on. Every model here streams, so text appears as it is written.
- Context size: set it to what the model can take, or lower to save money. Each model's page on uberllm.dev/models lists its context length. You pay for the whole prompt on every reply, so a 32k context on a long chat costs about 4 times a 8k one.
- Max response length: 300 to 600 tokens suits most roleplay. Output tokens cost more than input on almost every model.
Why it is cheaper
The same model is sold by several providers at different prices. UberLLM sends each reply to the cheapest one that is up right now and adds no fee per request, so a model like deepseek/deepseek-v4.1-flash-uncensored costs a fraction of its list price. The table below shows the price at the last hourly update.
Troubleshooting
- Status stays red, or the model list is empty: the key is wrong or has a space at the end. Paste it again and press Connect. You can still type the model id by hand; SillyTavern lets a custom endpoint skip the status check.
- "Status check bypassed" but no replies: the base URL is wrong. It must be exactly
https://api.uberllm.dev/v1. - 402 / "insufficient credits": add credit on the credits page.
- 404 model not found: copy the id from the table below, including the part before the slash (
deepseek/…,z-ai/…). - Replies cut off mid-sentence: raise Max response length.
- The model refuses a scene: switch to one of the uncensored models in the first table. Mainstream models (Claude, GPT, Gemini) follow their makers' rules.
Keep reading
- Reference page: SillyTavern doc
- Janitor AI setup (same key works there) · RisuAI doc
- All uncensored and roleplay models with prices
- How UberLLM picks the cheapest provider