AI Nodes

AI nodes use large language models (LLMs) to generate content, analyze text, classify data, and more. There are 9 AI nodes in total.

Prompt

Icon: MessageSquare

Create a prompt to send to an AI model. Define the system message and user message with dynamic variables from your workflow data.

Inputs: data-in (context data for the prompt)

Outputs: data-out (the constructed prompt)

Config:

System message: Instructions for the AI (e.g., "You are a helpful customer service agent")
User message: The actual prompt, with {{variable}} placeholders for dynamic data
Variables: Map workflow data to prompt variables

Example: Build a prompt that asks the AI to write a personalized follow-up email based on the customer's name and recent booking.

Preview

Icon: Eye

Preview the output of an AI operation before committing it. Shows you what the AI generated so you can inspect it in the workflow editor.

Inputs: data-in (AI-generated content)

Outputs: data-out (same content, passed through)

Config:

Format: How to display the preview (text, markdown, JSON)

Example: After the LLM generates an email draft, Preview lets you see the result in the editor before it gets sent.

LLM

Icon: Sparkles

Run a prompt through a Claude AI model and get the response. This is the core AI execution node that actually calls the language model.

Inputs: data-in (prompt and context)

Outputs: data-out (AI response), error (if generation failed)

Config:

Model: Which AI model to use
Max tokens: Maximum response length
Temperature: Creativity level (0 = precise, 1 = creative)

Example: Pass a customer complaint to the LLM and ask it to draft a professional response.

AI Classifier

Icon: Tags

Classify text into predefined categories. The AI reads the input and decides which category it belongs to, routing data to the matching output.

Inputs: data-in (text to classify)

Outputs: One output per category

Config:

Categories: List of categories with descriptions (e.g., "billing," "technical," "general")
Instructions: Additional context to help the AI classify accurately

Example: Classify incoming support emails by topic — billing questions route to finance, technical issues route to engineering.

AI Summarizer

Icon: FileText

Summarize long text into bullet points, paragraphs, or a brief TL;DR.

Inputs: data-in (text to summarize)

Outputs: data-out (summarized text)

Config:

Format: Bullet points, paragraph, or TL;DR
Max length: Maximum summary length
Focus: Optional topic to focus the summary on

Example: Summarize a long customer conversation thread into a 3-bullet-point recap for the manager.

AI Extractor

Icon: Scan

Extract structured data from unstructured text. Pull out names, emails, phone numbers, addresses, dates, and other fields.

Inputs: data-in (text to extract from)

Outputs: data-out (extracted fields as structured data)

Config:

Fields: Which fields to extract (name, email, phone, address, custom fields)
Instructions: Additional guidance for extraction

Example: Extract the customer's name, email, and requested service from a free-form inquiry message.

AI Sentiment

Icon: Heart

Analyze the emotional tone of text and route it based on sentiment — positive, negative, or neutral.

Inputs: data-in (text to analyze)

Outputs: positive, negative, neutral

Config:

Threshold: How confident the AI must be before routing (0-100%)
Instructions: Additional context for sentiment analysis

Example: Analyze customer reviews. Positive reviews get a thank-you reply. Negative reviews get escalated to the manager.

AI Translator

Icon: Languages

Translate text from one language to another using AI.

Inputs: data-in (text to translate)

Outputs: data-out (translated text)

Config:

Target language: Which language to translate into
Formality: Formal or informal tone
Preserve formatting: Keep the original structure (headers, lists, etc.)

Example: Translate a booking confirmation email into Spanish before sending it to a Spanish-speaking customer.

AI Generator

Icon: Wand2

Generate content for specific purposes — email copy, SMS messages, product descriptions, titles, or custom responses.

Inputs: data-in (context and instructions)

Outputs: data-out (generated content)

Config:

Content type: Email, SMS, description, title, response, or custom
Tone: Professional, friendly, casual, or formal
Length: Short, medium, or long
Instructions: Specific guidance for generation

Example: Generate a professional follow-up email based on the customer's booking details and service history.