What is a preset in SillyTavern? Presets explained

The kinds of preset

SillyTavern keeps several families of preset, and mixing them up is the most common confusion. Labels can change between versions, but the split has stayed the same:

• Chat Completion presets, in AI Response Configuration (the sliders icon), for APIs that take a list of messages. They hold samplers, Context Size, Max Response Length, streaming, and the prompt manager: main prompt, post-history instructions and utility prompts such as the impersonation prompt. They can also carry connection details like a custom endpoint.
• Text Completion presets, in the same panel when a Text Completion API is selected. These are sampler values only.
• Advanced Formatting templates (the "A" icon): context template, instruct template and system prompt. They tell text-completion models how to wrap each turn in the special tags their family expects.

Character cards, chats, personas and lorebooks are not part of any preset. Switching presets never touches your stories.

What a preset file looks like

A Chat Completion preset is a JSON file. Trimmed down, part of one reads like this:

{ "temperature": 0.9, "top_p": 0.95, "openai_max_context": 56000, "openai_max_tokens": 700, "stream_openai": true, "prompts": [ ... ] }

The field names are SillyTavern's own and may change between versions, so edit presets in the app rather than by hand. The prompts list is the part people forget: a shared preset can arrive with somebody else's main prompt and rules, which change how every character writes.

Importing, saving and switching

Import with the file icon beside the preset dropdown, then choose the preset from the list. If the file carries endpoint settings, SillyTavern asks whether to import them as they are. Moving a slider takes effect immediately, but it only becomes part of the preset when you save it; otherwise the next switch loses your change. Export does the reverse and gives you a file to back up or share.

A tidy habit is one preset per model you use, named after it, so switching models is one click and nothing carries over by accident.

Common mistakes

• Borrowing a preset made for another model. Sampler values are tuned per model; roleplay sampler settings covers what to reset.
• Importing a preset and losing your own prompt blocks without noticing.
• A context size larger than the model allows, which makes long chats fail once they fill up.
• Loading a Text Completion preset and wondering why a Chat Completion connection ignores it.
• Sharing a file with secrets in it. A preset should never contain an API key; check before you post one.

The UnsavedAI preset

The Setup page has a Download SillyTavern preset button for each model that is online. The file fills in UnsavedAI's endpoint URL, the model name and the recommended settings for that model, holds no key, and leaves your own prompts as they are. To use it: set API to Chat Completion in API Connections, open AI Response Configuration, click Import preset and pick the file, choose to import it as is, then paste your key into Custom API Key and click Connect. The SillyTavern guide has the manual route if you prefer typing the values.

Questions people ask