Reasoning effort, personalities, and the sampling knobs that shape replies.
3 answers
It sets how hard the model thinks before it answers. Low is quicker and right for everyday questions, Medium sits in between, and High works through multi-step problems most carefully — and most slowly.
High spends noticeably more of your daily allowance per turn, so ROK asks you to confirm the first time you pick it. Reach for it on the hard ones, then drop back down.
Named system prompts you control. In Settings → Personality, create one with a name and a system prompt — a writing style, a persona, a set of rules — then pick which one is active. ROK follows it from your next message, and you can keep several and switch between them.
ROK Hermes Spicy's persona isn't editable: it's a fixed server-side prompt, which is what makes that mode behave consistently.
They're the sampling knobs behind every reply. Temperature controls how varied the wording is — lower is more predictable, higher is more adventurous. Top P and Top K narrow the pool of candidate words the model picks from, and Max tokens caps how long a reply can get.
The defaults suit most conversations; change one at a time if you're tuning.
Didn't find it? Tell us what you were doing and we'll answer — and probably add it here.
[email protected]