google.SafetySetting

One Gemini safety filter setting, in the API's own spelling — pass a list of these as a client's `safety_settings`:

Reference version

Signature

class google.SafetySetting

One Gemini safety filter setting, in the API's own spelling — pass a list of these as a client's safety_settings:

google.GeminiClient.new(
    model = "gemini-2.5-flash",
    safety_settings = [
        google.SafetySetting.new("HARM_CATEGORY_DANGEROUS_CONTENT", "BLOCK_ONLY_HIGH"),
    ],
)

Categories are HARM_CATEGORY_HARASSMENT, HARM_CATEGORY_HATE_SPEECH, HARM_CATEGORY_SEXUALLY_EXPLICIT, HARM_CATEGORY_DANGEROUS_CONTENT and HARM_CATEGORY_CIVIC_INTEGRITY; thresholds are BLOCK_NONE, BLOCK_ONLY_HIGH, BLOCK_MEDIUM_AND_ABOVE, BLOCK_LOW_AND_ABOVE and OFF. The strings go on the wire verbatim, so a category this build has never heard of still works.

Source:<builtin>/google/gemini.bamlbytes 763–1112

Fields

threshold

string

Static methods

function

new

(category: string, threshold: string) -> google.SafetySetting

Creates a safety setting pairing a harm category with a blocking threshold. Both strings go on the wire verbatim.