Custom Behavior

Set up a custom AI behavior when you want more personalized results. Changing these settings can improve the quality of responses. Identifying the correct settings for your particular use case may take some experimentation.

Have a look at the LLM models we provide.

Settings

Below are common settings shared across LLM providers.

Temperature - it increases or decreases the confidence in response. Use lower for more factual and concise responses, and higher for more creative tasks.

Top P - it is an alternative to temperature. Keep it low for exact and factual answers, and increase for more diverse responses. We recommend using Top P or temperature, but not both.

Frequency Penalty - higher frequency penalty makes a word less likely to appear again. This setting helps reduce word repetition in the response by penalizing words that appear more often.

Presence Penalty - it stops the model from repeating words too much, regardless of how often they appear (unlike Frequency Penalty). If you want more diverse text, use a higher presence penalty. If you need the model to stay on topic, use a lower presence penalty.