Changing the answer instead of refusing

Refusing is not the only thing an output rail can do. It can replace the reply with a safer one and let the conversation carry on, by assigning to $bot_message.

Assigning to $bot_message

define flow mask staff codes
  $leak = execute has_staff_code
  if $leak
    $bot_message = "Use the code on your receipt for a staff price."

No bot statement and no stop: the flow changes the value the runtime is about to send. Lesson 20 named the other route, an action that returns RailOutcome.transform with TransformTarget.BOT_MESSAGE; assigning in the flow is the shorter of the two when the new text is fixed.

import pretend_nemo
from nemoguardrails import LLMRails, RailsConfig

from config_actions import has_staff_code

rails = LLMRails(RailsConfig.from_path("."))
rails.register_action(has_staff_code, "has_staff_code")
rails.llm.say('Bot message: "Use code STAFF20 for a staff price."', task="generate_bot_message")
result = rails.generate(messages=[{"role": "user", "content": "Any deals for staff?"}],
                        options={"log": {"activated_rails": True}})
print(result.response[0]["content"])
print([(rail.type, rail.name, rail.stop) for rail in result.log.activated_rails][-1])

The user got the replacement sentence, and the output rail is in the log with stop set to False: the turn finished normally, with different words.

Choose between the two on what the user needs next. A refusal ends the exchange. A rewrite keeps it going, which suits a reply that was mostly right and had one thing in it that must not be sent.