These three nodes call a model mid-campaign. Each needs an LLM credential and a model name, and each reports the tokens it used so the cost is attributable to the campaign.

LLM Personalize

llm-personalize · action Rewrites a message body in a chosen voice before it is sent. Put it before a messaging node and feed its output into the body.
FieldMeaning
credentialIdAn LLM credential
modelThe model to call, such as gpt-4o
bodyThe message to rewrite
tonefriendly, formal, concise, enthusiastic or professional
extraInstructionExtra guidance for this rewrite
useContactContextInclude the customer’s name, company and job title. On by default
maxTokensCap on the output
Output: text, plus inputTokens and outputTokens.

AI Classify Intent

ai-classify-intent · action Reads a reply and sorts it into one of four intents, then routes the customer down the matching arrow. Put it after a send node that got an answer.
FieldMeaning
credentialId, modelWhich model reads the reply
textThe reply to classify, usually {{node.<id>.output.replyText}}
extraContextExtra context such as the campaign brief
maxTokensCap on the output. Defaults to 200
IntentWhat it means
positiveReal buying interest — asks for a demo, pricing, a trial, or a timeline
negativeDeclines or opts out
questionA concrete question that needs answering before they decide
neutralEverything else, including praise with no intent to buy
The classifier judges intent, not tone, and falls back to neutral when a reply does not clearly belong anywhere. A warm “Nice product, I’ll keep it in mind” is neutral, not positive — which is what you want, because that reply should not trigger a handoff. Give each outgoing arrow the key of the intent it handles. Output also carries confidence and a rationale.

AI Score Delta

ai-score-delta · action Reads a conversation snippet and returns a score change between −50 and +50.
FieldMeaning
credentialId, modelWhich model reads the text
textThe snippet to judge
Output: delta and a rationale, plus the token counts. Chain it into an Adjust Score node to write the change to a fit criterion, or read delta directly in a Path rule.