SillyTavern for beginners: what it is and how to start

What SillyTavern is, and what it is not

SillyTavern is a front end: the window you chat in, plus a lot of control over what gets sent to the model. The model itself lives somewhere else, either on your own computer through a local backend or on a hosted API. That split is why the first screen asks you to connect something before you can talk to anyone.

The upside is freedom. You can switch models without losing characters, edit every prompt, and keep your chats on your own machine.

The five pieces you meet first

• API Connections (the plug icon): which model answers. For a hosted model pick Chat Completion.
• AI Response Configuration (the sliders icon): context size, reply length and samplers.
• Characters: cards you import as PNG or JSON, or write yourself.
• Persona: who you are in the story, sent along with every message.
• World Info: a lorebook of facts that only enter the prompt when a keyword shows up.

You can ignore World Info, extensions and advanced formatting for your first week. They are powerful, and they are also where most beginner confusion comes from.

A good first hour

Install SillyTavern by following its own documentation, open it in your browser, then connect a model. With UnsavedAI that is Chat Completion, source Custom (OpenAI-compatible), our base URL ending in /v1, your key and gemma-4-12b. The SillyTavern guide walks through each field.

Then import one character you like. Any character you make on UnsavedAI can be exported as a Character Card V2 file that SillyTavern reads directly. Play twenty messages before you change any setting, so you know what normal looks like.

Three settings to get right on day one

• Context Size: 56,000 for our Gemma models. Higher fails once the chat fills up.
• Max Response Length: raise it from the default 300 to 600 to 800, or longer replies stop mid-sentence.
• Temperature: 0.9 with Top P 0.95. Leave the rest neutral.

The Setup page also offers a ready-made SillyTavern preset with these values. Import it under AI Response Configuration and it sets them in one go without touching your own prompts.

Common beginner mistakes

Importing a huge preset from a forum on your first day is the big one: it can change prompts, samplers and formatting at once, and you cannot tell which part broke. Others are leaving image sending on (our models are text only, so you get "This model accepts text messages only."), turning on function calling, and clicking swipe repeatedly while a reply is still generating, which queues extra requests.

Questions people ask