Connect SillyTavern to an uncensored API

What you need

• SillyTavern installed and running (a recent release; the menus below match the current layout).
• An API key from your account. Sign up with an email and a password; $0.50 of trial credit is added when you first open the account.
• The base URL https://dolphinllm.com/v1 and the model ID uncensored.

Connect in four steps

1. Open the API Connections panel (the plug icon in the top bar).
2. Set API to Chat Completion, then Chat Completion Source to Custom (OpenAI-compatible).
3. Paste https://dolphinllm.com/v1 into Custom Endpoint (Base URL) and your key into Custom API Key.
4. Click Connect. Pick uncensored from Available Models, or type it into Enter a Model ID. The status turns green when the first test message goes through.

Do not choose Text Completion. This API serves the chat format only (/v1/chat/completions); the model's own chat template is applied on the server, so you do not need an instruct template in SillyTavern for this connection.

Settings that matter for roleplay

Open AI Response Configuration (the sliders icon).

• Context Size (tokens): up to 262,144. Every message sends the character card, the persona, world info and as much chat history as fits, and all of it is billed as input. 16,000 to 32,000 is a good balance between memory and cost for most chats; a full 262,144-token context costs about $0.0393 per message in input. Raise it for long campaigns where the model must remember early events.
• Max Response Length (tokens): 300 to 800 for chat replies. Output is the expensive side at $0.50 per 1M tokens.
• Streaming: on. Text appears as it is written, and you can stop a reply you do not like before it finishes.
• Temperature: 0.7 to 1.0 for creative writing. Top P, frequency and presence penalties and stop strings are passed through to the model.

The model is tuned to answer without content refusals for lawful adult use, so you rarely need jailbreak prompts. A short system prompt that sets the style and perspective works better. The one fixed rule on this API: no sexual content involving minors, whatever the character card says.

Extensions and the trial key

Some SillyTavern extensions send their own requests through the same connection, for example chat summaries or automatic character expressions when they are set to use the main API. On a trial key, which allows 2 parallel requests and 60 a minute, these extra calls can collide with your message and return HTTP 429 with the code concurrency. Turn them off while you test, or point them at a local model. After the first top-up the key allows 8 parallel requests and 300 a minute.

Group chats send one request per character turn, so they use credit faster than one-to-one chats.

What a session costs

Pricing is $0.15 per 1M input tokens and $0.50 per 1M output tokens. A message with 6,000 tokens of card and history and a 400-token reply costs:

• Input: 6,000 × $0.15 / 1,000,000 = $0.0009
• Output: 400 × $0.50 / 1,000,000 = $0.0002
• Total: $0.0011 per message, about $1.10 per 1,000 messages

As the chat grows toward a 32,000-token context, a message costs about $0.005. The $0.50 trial covers roughly 450 messages at the smaller size. Top-ups start at $50; see the pricing page.

Troubleshooting

• No models in the list. Check that the URL ends in /v1 with no trailing slash, then type uncensored into the model ID field by hand.
• 401 invalid_key. The key was copied with a space or was revoked. Create a new one in your account.
• 402 no_credit. The balance is empty or the trial expired. Top up in your account.
• 400 bad_request about context. Prompt plus Max Response Length is above 262,144 tokens. Lower Context Size.
• 503 upstream_busy. The shared servers are full for a moment. Swipe or regenerate after a few seconds; you are not charged for the error.