SillyTavern with Zukko: connect Claude and GPT through your own API

SillyTavern is an interface for role-play, interactive stories and character chats. You can connect Claude and GPT through an API in SillyTavern with the Custom (OpenAI-compatible) source, and that's how Zukko is added: the address https://api.zukko.pro/v1 and a fetch_... key. Below is the setup based on the official SillyTavern docs, the settings that drive cost, a calculation for long chats, and troubleshooting.

How to connect SillyTavern to Zukko

Per the SillyTavern docs, connecting to an OpenAI-compatible API looks like this:

1. Open the API Connections panel (the plug icon in the top menu).
2. Set API to Chat Completion.
3. Set Chat Completion Source to Custom (OpenAI-compatible).
4. Enter the endpoint address: https://api.zukko.pro/v1.
5. Enter your fetch_... API key.
6. Choose a model. If the API serves a model list at /v1/models, SillyTavern shows a dropdown; Zukko serves that list. You can also type a model ID into the text field.
7. Click Connect, then Test Message to send a test message.

The SillyTavern docs say explicitly not to add the /chat/completions suffix, and to try adding /v1 to the end of the address if you have connection problems. For Zukko, the correct address is exactly https://api.zukko.pro/v1. The Bypass API status check option is for APIs that don't report status; with Zukko you can leave it off.

For the current model list, call GET https://api.zukko.pro/v1/models (see the models documentation↗) or open the "Models" page in the dashboard.

Settings that drive cost

The response settings panel (AI Response Configuration) has two key parameters:

• Context (tokens) — how many tokens SillyTavern sends to the model in each request: character card, system prompt, chat history and so on;
• Response (tokens) — the maximum answer length.

The key point: on every generation SillyTavern resends the entire context. Once chat history fills the Context window, every request costs about the same, and most of the price is input tokens. Every swipe (regenerating an answer) is a separate full request with the same context.

The cost of a long chat

Say Context = 16,000 tokens and Response = 400. Once the window is full, one generation is 16,400 tokens:

If you raise Context to 32,000, one generation on claude-sonnet-5 costs 32,400 × 0.3 / 1,000,000 = $0.00972, and 200 generations cost $1.944. So context size is your main lever for saving money: keep it as large as the character and story actually need.

A note on caching. Vendors may read resent context from the prompt cache at roughly 10% of the input price, while Zukko bills cached tokens at the regular input price. In long chats with a large repeated context, the real difference from official prices is therefore smaller than a straight price-list comparison suggests. More in the key limits and cache docs↗.

Temperature and Top P for Claude models

SillyTavern sends sampling parameters with every request. claude-sonnet-5 and claude-opus-5-5 may reject a request with a 400 error if temperature is anything other than 1 or top_p is below 0.99. For these models, keep Temperature = 1 and Top P = 1. With Claude, steer variety through the prompt and the character card instead. For GPT and GLM models, tune samplers as usual and check the results on test messages.

How to choose a model

Nobody can tell you in advance which model fits your style and characters best, so test it. Take one character card and 5–10 identical opening situations, run them on two or three models (you can switch the model in API Connections and regenerate), and compare the answers. Spending per model is shown in the dashboard's "Usage" section.

Common errors

The connection won't establish

Check the address: https://api.zukko.pro/v1, with no /chat/completions at the end. The key must start with fetch_ and be copied in full.

400 on Claude models

Most likely Temperature or Top P was changed. Set Temperature = 1 and Top P = 1 again (see the section above).

A message format error

Some APIs require, for example, a single system message or strictly alternating roles. SillyTavern has a Prompt Post-Processing setting for this, with modes from Merge to Single user message. Start with the least strict mode.

401, 402, 403

401 means an invalid key, or the key was deleted in the dashboard. 402 means the balance ran out. 403 model_not_allowed_for_key means the model isn't on the key's allowed list.

413 or answers get cut off

413 means the request is too large: lower Context. If answers are cut off, raise Response (tokens).

FAQ

Why does spending grow even though my messages are short?

The price isn't set by the length of your message but by the whole context sent with each request: card, history, system prompt. A short line in a long chat costs nearly as much as a long one.

Can I switch models within one chat?

Yes, the model is chosen in API Connections and applies to the next generation. The chat history stays as it is.

How do I cap spending?

Set daily and monthly limits on the key in the dashboard: once a limit is reached, requests stop going through. Also keep Context (tokens) reasonable and watch how many swipes you make.

Where do I get a key?

Create one in the Zukko dashboard↗ and paste it into API Connections. If you're new to APIs, start with the quick start guide↗.