Behavior Tweaking

What TomoriBot is allowed to do and how she generates replies is set in /config under Behavior, Plugins, Channels, and Models. This page covers the most common behavior settings.

Her personality (Multiple Personas) and knowledge (Memory) are covered on their own pages, and admin and cost controls (quotas, whitelists, BYOK) live under Server Moderation.

System Prompt

The system prompt sits above the persona and shapes overall behavior. Everything below is on /config > Behavior > General Behavior:

System Prompt: instructions every persona follows. Set a custom system instruction (up to 16,000 characters), choose from preset system prompts, or reset to the default. The reset confirmation shows the previous prompt so you can restore it if cleared accidentally.
Context Note: a system nudge placed close to the last message, at a depth you choose. Use it for reminders the model tends to forget in long chats.
Response Prefill: text she starts every reply with and continues from. It never shows in Discord. Prefill only applies when the model can continue it: the panel shows whether it is active, and tells you if the model, its Thinking level (use None or Minimal), or Tool Use stops it from being used.

When a SillyTavern preset is active, it controls the system prompt. Turn the preset off to edit the saved server prompt again.

Generation Tuning

/config > Models > Text Samplers & Parameters: sampling parameters such as temperature and top-p. Higher temperature yields more variety.
/config > Behavior > General Behavior: response humanizer. Paced messages adds delays and pauses between streamed messages. Pacing and both split options require live streaming; force lowercase also works without streaming. These settings apply server-wide by default, or to an individual persona. /setup offers the same habits as four ready-made presets.
/config > Behavior > General Behavior: message history limit. Raise it for deeper conversational context, or lower it to save tokens. This is a maximum: recent history grows, then drops older messages in batches. See Inside The Prompt for the window and placement rules.

Capabilities

Two pages under /config > Plugins control what she can do:

/config > Plugins > Available Tools: image generation, sticker usage, thread creation, message management, user blocking, self-teaching, voice messages, and more. Each toggle gates the matching tool (see Tools & Extensions), so turning off Tool Use disables all of them at once.
/config > Plugins > Context Additions: personalization, emojis in replies, and time awareness. These add information to her prompt, so they continue working when Tool Use is disabled.

Automatic short-term memory summarization is toggled in /config > Behavior > Advanced Memory. When a capability is turned off, she cannot perform that action regardless of user requests.

Response drafting

Open /config > Plugins > Response Drafting to turn review On. It starts Off and applies to every persona in that server, including queued replies and generated scenes. Server managers change server settings; your DMs have their own. When On, Tomori holds her reply while a reviewer checks its character voice and scene fit. The reviewer can ask for one complete revision.

Models used, prompts used, criteria, etc. can be completely customized in the same page.

Expressions

Expressions let your personas react with more than words. She can use your server’s emojis and stickers, and you can give her reactions of your own: a favorite GIF, an inside-joke image, or a link to anything on the web. Describe when each one fits, and she sends it when the moment comes.

Members with Manage Server open /expressions manage to browse everything in three tabs: Emojis, Stickers, and Customs.

Your Server’s Emojis and Stickers

Run /expressions initialize and she looks at each emoji and sticker to learn when it fits. Ones added later show as uninitialized in /expressions manage until you run it again. To correct her, select one and choose Edit to change its description and emotion, or Clear Info to erase them.

Custom Expressions

In Customs, open the menu and choose + Add a custom expression. Give it a name, a description of when to use it, an emotion, and either a link or a file. She reads the description to decide when to send it, so be specific: “when chat gets unhinged” works better than “funny”.

A link can point to anything. She posts it exactly as saved, and Discord shows it the way it shows any link: a GIF link from a site like Tenor plays as a GIF, an image link shows the image, and a website shows its preview card. That makes links good for jokes, like a hospital’s website for the moment chat gets unhinged. Links must start with https://.

Files can be PNG, JPEG, WebP, GIF, or MP4, up to 10 MB.

Every persona can use a new custom expression. To keep one for specific personas, select it and use Add Persona. Removing the last persona from that list opens it to everyone again.

Uncensored Output

TomoriBot has no content filter of its own: she adds no moderation layer on top of the model, and replies with whatever the provider generates. /nsfw jailbreaks does not enable hidden bot features; it works around provider-side filters that are stricter than desired.

It toggles three independent techniques (all off by default):

Prompt injection: adds an instruction block to the context to steer the model away from unnecessary refusals.
Unicode spaces: swaps normal spaces for look-alike Unicode spaces so keyword filters do not trigger on phrases, applied to both the prompt and her reply.
Sanitize: obfuscates sensitive words for the same reason, on both requests and replies.