Presets

Save model settings — temperature, prompts, penalties — and reuse them across chats for a consistent voice.

A preset bundles the model settings and prompts that shape how a chat writes: how creative it is, how it’s prompted, and how it behaves. Save a preset once and apply it to any chat.

Create a preset

Go to Create → Create Preset (or /presets/new).
Set the sampling parameters and prompts.
Save. Presets default to private until you publish them.

Every field explained

Sampling

These control the model’s randomness and word choice:

Temperature — how creative/unpredictable output is. Higher = wilder.
Top P — nucleus sampling; consider only the most-likely tokens summing to P.
Top K — consider only the K most-likely tokens.
Top A — an alternative cutoff based on the top token’s probability.
Min P — drop tokens whose probability is below a fraction of the top token’s.
Repetition penalty — discourage the model from repeating itself.
Frequency penalty — penalize tokens by how often they appear.
Presence penalty — penalize tokens that have appeared at all.
Max output tokens — cap on reply length.

Prompts

System prompt — the master instruction for the model.
Main prompt — additional conversation-level instructions.
Jailbreak prompt — extra directives that loosen default constraints.
Assistant prefill — text the model starts its reply with.
Impersonation prompt — instructions for playing the character’s voice.

Set as default — apply this preset to new chats automatically.
Visibility — public, unlisted, or private.
Description — a note about when to use this preset.

Using presets in chat

In a chat, pick a preset from the settings panel (or rely on your default). Different characters or moods can each have their own preset — a “literary” preset for long-form scenes, a “snappy” one for quick banter.