Provider specific model name.
If supported by the model.
Variety. Lower = predictable.
Nucleus sampling.
Token shortlist size.
Discourage repeats by count.
Reduce verbatim echoes.
Minimum token prob vs best.
Adaptive nucleus filter.
OpenAI param.
Saved per sune.
Scripts also run. extension.html runs after index.html.
Or you can just prefix names with or:, oai:, or g: to override.
Stored locally.
Used for auto-generating thread titles.
Optional. For future use.