class RubyLLM::ContentFilter

A ContentFilter is one verdict of the provider’s content filtering on a request or its response, such as Azure’s hate or protected material filters. They appear on Message#content_filters for providers that report them, whether or not they stopped the response, so you can see why a reply ended with Message#content_filtered? or what came close.

response = chat.ask "Summarize this paper"
response.content_filtered?                       # => true
response.content_filters.select(&:filtered?).map(&:category) # => ["hate"]

A verdict names a category and a severity, never the text it judged.

Attributes

category [R]

The provider’s name for the category, such as "hate", "jailbreak", or "protected_material_text".

severity [R]

The severity the provider gave, such as "safe", "low", or "medium", or nil for a category judged only as detected or not.

source [R]

Which side was judged: :prompt for the request, :completion for the response.

Public Instance Methods

detected? ()

Source

# File lib/ruby_llm/content_filter.rb, line 55
def detected?
  @detected
end

Returns whether the provider detected what the category looks for, such as a jailbreak attempt or protected material, or nil for a category judged by severity.

filtered? ()

Source

# File lib/ruby_llm/content_filter.rb, line 48
def filtered?
  @filtered
end

Returns whether the filter blocked or cut the text it judged.

to_h ()

Source

# File lib/ruby_llm/content_filter.rb, line 60
def to_h
  { source: source, category: category, severity: severity, filtered: filtered?, detected: @detected }.compact
end

Returns the verdict as a Hash, omitting nil values.