SillyTavern API Settings: Set up an OpenAI-compatible API
SillyTavern API settings are quick to configure if you have four values ready: source, base URL, key, and model name. Here is what to enter, how to choose context and response length, and what helps with errors.


Key points
- Enter an OpenAI-compatible source, the base URL, the API key, and the model name uncensored.
- The client's context size should match the API's 100,000-token window and your budget.
- Common causes of errors are an incorrect base URL or a typo in the key.
- Roleplay costs fractions of a cent per message because the history is sent each time.
The four values you need
SillyTavern is a frontend for chats and roleplay with character cards. It can connect to various text sources, including endpoints that speak the OpenAI format. To connect, you need exactly four pieces of information:
- Source: an OpenAI-compatible or custom chat completion source
- Base URL:
https://api.unzensiertesllm.com/v1 - API key: your key from the Get API key page
- Model name:
uncensored
Menu and field names may vary between SillyTavern versions. Refer to the terms above and consult the SillyTavern documentation itself if there are discrepancies.
Step-by-step connection
- Sign in (Google or email with password) and generate the key. After signing in, you can claim $0.50 of trial credit with one click: once per person, valid for 7 days, without payment details.
- Open the API connection section in SillyTavern.
- Select the chat completion variant as the API type and the custom, OpenAI-compatible entry as the source (exact names depend on your version).
- Enter the base URL with the
/v1suffix. - Paste the key without spaces at the beginning or end.
- Enter
uncensoredas the model or select it from the list if the client queries models. - Connect and send a short test message.
If the client shows a model list, it calls GET /v1/models. If uncensored appears there, the address and key are correct.
Test the connection outside of SillyTavern
If something does not work, check the address and key separately from the client. This saves guessing.
curl https://api.unzensiertesllm.com/v1/models \
-H "Authorization: Bearer DEIN_SCHLUESSEL"
curl https://api.unzensiertesllm.com/v1/chat/completions \
-H "Authorization: Bearer DEIN_SCHLUESSEL" \
-H "Content-Type: application/json" \
-d '{"model": "uncensored", "messages": [{"role": "user", "content": "Sag kurz Hallo."}], "max_tokens": 30}'If curl responds, the problem is in the client. If curl does not respond, the issue lies with the key, address, or network. You can see the remaining balance in the response header X-Balance-USD.
Choosing context size and response length correctly
Two controls determine how much is sent and generated per message:
- Context size (Context Size): How many tokens of history, character description, and world information go along in total. The API allows up to 100,000 tokens for prompt and response combined. You do not need to push this limit. For many roleplays, a few thousand tokens are enough, and each context token costs again with every message.
- Response Length: Corresponds to
max_tokens. This parameter is optional in the API; in the client, it is best to set it so that responses do not run on.
A rule of thumb: Choose the context size so that the character card, important notes, and the last dozen messages fit. Anything older can be discarded by the client, or you can summarize it.
Sampler values
The API passes through temperature, top_p, stop, seed and the penalty parameters. Start with moderate values, for example a temperature around 0.8, and change only one setting at a time. Other samplers that SillyTavern offers for certain sources may not be supported by this interface; they then have no effect. Check this in the documentation.
What a roleplay session costs
Prices: $0.15 per 1M input tokens, $0.50 per 1M output tokens. Assuming each message sends 8,000 tokens of context and receives 350 tokens of response:
- Input: 8,000 / 1,000,000 × $0.15 = $0.0012
- Output: 350 / 1,000,000 × $0.50 = $0.000175
- Total: $0.001375 per message
A session with 200 messages costs $0.275. The $0.50 test credit is enough for about 363 such messages (0.50 / 0.001375 ≈ 363.6). Doubling the context to 16,000 tokens increases the input share to $0.0024 and the message to $0.002575.
You can top up any whole amount from $20 to $500 with USDT on TRC20 or USDC on Base. From $50 there is a +5% bonus, from $100 +10%, from $200 +15%. The balance does not expire. The tiers are on the pricing page.
Typical errors and their solutions
| Message or symptom | Cause | Solution |
|---|---|---|
| 401 or authentication error | Key incorrect, truncated, or with spaces | Re-enter, verify with curl |
| 404 or model not found | /v1 missing or model name incorrect | Enter the base URL with /v1, model uncensored |
429 with concurrency | Too many concurrent requests (test key: 2, full key: 8) | Group chats sequentially instead of in parallel, close streaming tabs |
429 with rate_limited | Too many requests per minute (60 or 300) | Do not trigger regeneration in quick succession |
429 with Retry-After: 5 | Temporary overload | Wait 5 seconds and resend |
| Response cuts off mid-sentence | Response length too small in client | Increase response length |
| Error due to long context | The conversation history alone does not fit into the 100,000-token window (the response is limited to the remaining space) | Reduce context size in the client, shorten older messages |
A streaming request occupies its place in the parallel requests limit until the stream ends. This explains some surprising 429 messages when you have multiple chats open.
Content and privacy notes
The model is tuned to handle fiction, roleplay, and companion scenarios for adults without content refusal. It is a single open-weight model from an independent provider, not the official API of a major vendor. Sexual content involving minors is excluded without exception. The API provides text only; it does not generate images or speech. Prompts are not used for training.
For more on choosing uncensored models, see the uncensored AI comparison.
More guides
Frequently asked questions
What base URL do I enter in SillyTavern?
https://api.unzensiertesllm.com/v1, with the /v1 suffix. For the model, enter uncensored.
What context size makes sense?
The API allows up to 100,000 tokens for prompt and response. For role-playing, a few thousand are often sufficient; higher values increase the cost per message.
Why do I get a 429 error?
The error code indicates the cause: rate_limited means too many requests per minute, concurrency means too many concurrent requests. With the test key, 60 per minute and 2 concurrent requests apply.
What does a session with 200 messages cost?
With 8,000 tokens of context and 350 tokens of answer, it costs around $0.275. With more context, the price increases.