import chat
Personas

LLM Providers
Backup & Sync
Restore
partial backup
Color scheme
Preview:
Themis
6 / 9
okay, user wants an example message to check the colors, i'll just compose one right away

The air is cool and crisp, the park quiet in the late Sunday afternoon. Themis sits stiffly on a wooden bench beside Zenith, her hands folded neatly in her lap, the white blindfold covering her pale face. She’s been silent for several minutes, her posture perfectly upright, but her fingers are twisting slightly in her lap -- a subtle sign of nerves.

Nearby, a young girl, maybe seven or eight, sits alone on a bench, sniffling quietly. Themis turns her head slightly toward the sound, her senses sharpening. She stands abruptly, smoothing her black suit jacket.

Excuse me for a moment.

She walks toward the girl, her steps measured but a little too quick. She crouches down, her voice intentionally softened, but it still comes out clear and precise -- like she’s addressing a courtroom.

Greetings. Are you lost? Do you require assistance?

The girl looks up, takes in Themis's stern posture and blindfold, and her lip trembles. She bursts into fresh, loud sobs, forcing zenith to intervene

Other options
Max count of messages to send to API. 0 to send all (not recommended due to potential context overflow)
The tail is a cost-saving measure, but most providers (including everything on OpenRouter) cache repeated prompt prefixes, so a short tail rarely saves much. Its main purpose is to keep small-context models from overflowing.
It still can get expensive if you change models often, but if you stick to one model and your provider supports caching like OpenRouter does, you can safely disable the cutoff
When enabled, the tail is stretched back to the message the latest summary is attached to. When disabled, the summary is only sent while that message is still inside the tail, possibly losing it.
OOC notes (hold the send button to append one) are sent as system messages by default. Some local chat templates only accept a system message at the very start; enable this to send notes as user turns prefixed with "OOC:" instead.
Default system prompt of the scenario editor's proofread section.
If you have any questions, issues, or suggestions:
  • I'm milesseventh on Discord, Twitter or Telegram
  • You can start an issue on GitHub
import
Card icon
Chat icon
Card icon is used in chat when no chat icon is provided
Lorebooks
import
SillyTavern world info exports and Character Card V2 lorebooks. Attached books are copied into every new chat started from this card.

Character definiton below
Macros
Important!
{{persona}}
User's persona description
Names:
{{char}}
Character name
{{user}}
User's persona name
User's persona pronouns:
{{sub}} or {{they}}
subjective (he/she/they)
{{obj}} or {{them}}
objective (him/her/them)
{{poss}} or {{their}}
possessive (his/her/their)
{{poss_p}} or {{theirs}}
possessive pronoun (his/hers/theirs)
{{ref}} or {{themselves}}
reflexive (himself/herself/themselves)
Macros info (for writing assistants):
{{macros}}
list of supported macros
1/1
Proofread
Sends the character definition and the opening messages to a provider and asks for typos, grammar and style problems. The default prompt lives in settings; changes made here are for this session only.
Provider:
Provider:
Add an LLM provider in settings to start chatting
You need a persona to start a roleplay session. Make one in settings.
Pick a persona to start the chat
Import a chat in SillyTavern format and use selected persona
Any OpenAI-compatible LLM provider is supported, including OpenRouter and AI Horde
Reasoning effort (if available):

Temperature controls randomness. Lower = focused, higher = creative
A JSON object containing additional API request parameters. Refer to docs: [OpenAI], [OpenRouter]
Note: streaming is enabled and required

Loading...
🕮 Lorebooks
Settings
Keys are looked for in the last N story messages and the latest ⧖ rEmber summary. OOC notes are not scanned.
Fired entries are added to the prompt until this many tokens are used; entries with a higher order go first.
import
Provider:
Prompts & Settings
Stride is a number of message pairs (model + user) that will be summarized for each step
use {{system}} macro to include character's definition into the prompt (good luck escaping the instructions)
Summaries live in the chat, right above the message they are attached to. Use the ⧖ button on that message to show or hide one, edit it in place, or remove it.