Presets
The request recipe: sampling, behavior toggles, and core prompts.
Sampling parameters
Presets carry the full sampler set: temperature, top-p, frequency and presence penalties, max output tokens, and reasoning effort where supported. Conversation-level sliders override the preset for that chat only.
Extra body lets you pass provider-specific JSON fields verbatim; on key clash it beats the preset, and sliders beat both.
Behavior toggles
Toggles adapt the request to strict providers and special workflows:
| Toggle | Effect |
|---|---|
noSystemRole | Rewrites system messages as user messages for endpoints that reject the system role. |
forceAlternateRoles | Merges consecutive same-role messages so roles strictly alternate. |
mustStartWithUserInput | Prepends a stub user turn when the history would otherwise start with the assistant. |
geminiBlockOff | Sets all Gemini safety categories to OFF. |
streamingEnabled | Streams tokens as they generate instead of waiting for the full reply. |
showReasoning | Shows model reasoning blocks in the UI when the model emits them. |
Core prompts
The main prompt opens the request and carries your global instructions. It supports macros, so one preset adapts to any character.
Post-history instructions are injected after the chat history with a configurable role; being last gives them the strongest pull on the next reply.
Prefill seeds the start of the assistant's answer, useful for steering models that support assistant-side continuation.
Default preset
Mark a preset as default for new chats, and pin a different one in the loadout when a scenario needs its own recipe. Duplicate a preset before experimenting; switching back is one click.