Should we redact Tool content

[stevepolitodesign]

RubyLLM offers an API for Tools. At the moment, I think the current implementation would redact content when those are called, but I need to investigate. If this is the case, does this make sense?

[stevepolitodesign]

I can confirmed the following works:

• When we did not wrap the calls in RubyLLM::TopSecret.with_filtering I could confirm we sent raw data to the provider, and the response was in the expected shape.
• When we did wrap the calls in RubyLLM::TopSecret.with_filtering I could confirm we sent filtered data to the provider, and the response was restored.

class ProfileLookup < RubyLLM::Tool
  description "Looks up a user profile by email and returns their account details"

  param :email, desc: "The user's email address"

  def execute(email:)
    {
      email: email,
      plan: "Premium",
      member_since: "2021-03-15"
    }
  end
end

# Should not filter
puts "\n--- Tool call without filtering ---"
chat = RubyLLM.chat.with_tool(ProfileLookup)
response = chat.ask "Look up the profile for user@example.com"
puts response.content

# Should filter
puts "\n--- Tool call with filtering ---"
RubyLLM::TopSecret.with_filtering do
  chat = RubyLLM.chat.with_tool(ProfileLookup)
  response = chat.ask "Look up the profile for user@example.com"
  puts response.content
end