Behavior Tweaking

TomoriBot’s behavior — what she’s allowed to do and how she generates — is controlled by /capabilities and /config, beyond personality (Multiple Personas) and knowledge (Memory). This page is a curated set of the high-value knobs — every command is in the Command Reference.

Capabilities: What She’s Allowed to Do

/capabilities toggles her features on and off — image generation, sticker usage, thread creation, message management, user blocking, self-teaching, voice messages, and more. Each toggle is the feature flag that gates the matching tool (see Tools & Extensions). Turn something off and she simply can’t do it, no matter what a user asks.

Generation Tuning

/model parameters — sampling parameters (temperature, top-p, …): creativity/randomness. Higher temperature is more varied.
/config humanizer — how human-like her responses read. The optional scope option applies the degree server-wide (Global, the default) or to a single persona (Persona), handy when one persona should text casually at degree 3 while another texts like a novel. A persona’s “Inherit” choice clears its override.
/config message-fetch-limit — how many recent messages she pulls as context per trigger. A useful lever: raise it for more conversational awareness, lower it to cut token cost.

System Prompt

The system prompt sits above the persona and shapes overall behavior:

/config system-prompt set — set a custom system instruction (up to 16,000 characters).
/config system-prompt preset — choose from preset system prompts.
/config system-prompt remove — reset to the default.

When a SillyTavern preset is active, the built-in fallback system prompt is replaced — but a custom one you set here is still sent.

Uncensored Output

TomoriBot has no content filter of its own — she is not a moderation harness and adds no safety rails on top of the model. Whatever the underlying provider returns is what she says. /nsfw jailbreaks therefore doesn’t “unlock” anything inside TomoriBot; it exists purely to work around provider-side filters that are stricter than you want.

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

Prompt injection — adds a jailbreak instruction block to the context to steer the model away from unnecessary refusals.
Unicode spaces — swaps normal spaces for a look-alike Unicode space so keyword/token filters don’t match on phrases, on the text sent to the model and on her reply.
Sanitize — obfuscates a set of sensitive words for the same reason, likewise on both the request and the response.

None of these change what the model is capable of — they only reduce how often an over-eager provider filter blocks otherwise-normal output. Some of these options are age-restricted — see Age-Restricted Commands.

Appearance & Time

/persona rename — what she calls herself.
/server timezone — the server timezone, used for time-aware replies and reminders.

Looking for admin/cost controls (quotas, whitelists, BYOK) rather than behavior? Those live under Server Moderation.