Understanding AI Nodes and Connections

The LLM Chat node lets you chat with AI models like GPT-4, Claude, or Gemini:

• Model selection: Choose from GPT-4, GPT-3.5, Claude Opus/Sonnet/Haiku, Gemini Pro
• System prompt: Instructions for the AI—use templates with {{variables}} to reference workflow data
• Temperature: Control randomness (0 = deterministic, 2 = very creative)
• Max tokens: Limit how long the response can be
• Outputs: Response text, tokens used, and cost in credits

Connect the LLM Chat node to input nodes (user input, webhook) to receive prompts, and connect its output to other nodes (data processing, output) to use the AI's response in your workflow.

The Agentic AI node is an autonomous AI agent that can use other nodes as tools:

• Tool calling: AI can call any available node as a tool dynamically
• Iterative execution: AI can call tools multiple times in a loop
• Available tools: Specify which nodes the AI can use (database_query, email_send, etc.)
• Max iterations: Limit how many tool calls the AI can make (default 10)
• Outputs: Final response and list of tool calls made

The Agentic AI node connects to other nodes to use them as tools. When you configure available tools, the AI can automatically call those nodes when needed to accomplish its goal.

The Image Generation node creates images using AI models like DALL-E:

• Model selection: DALL-E 3, DALL-E 2, or other image AI models
• Prompt: Description of the image you want to generate
• Size: Image dimensions (1024x1024, 512x512, etc.)
• Quality: Standard or HD quality
• Outputs: Image URL, image data, and metadata

Connect the Image Generation node to input nodes (user input, data) to get prompts, and connect its output to nodes that need images (output, storage, webhook).

The Speech to Text node converts audio to text using Whisper:

• Model selection: Whisper-1 or other speech recognition models
• Audio input: Audio file or audio URL
• Language: Auto-detect or specify language
• Timestamps: Include timestamps in the output
• Outputs: Transcribed text, segments, and confidence scores

Connect the Speech to Text node to nodes that provide audio (file upload, webhook), and connect its output to nodes that process text (LLM Chat, data processing, output).

AI nodes work best when connected properly:

• Input connections: Connect data nodes (user input, webhook, database) to AI nodes to provide context
• Output connections: Connect AI node outputs to processing nodes (logic, data transformation) or output nodes
• Chaining: Chain multiple AI nodes together—use one AI's output as another's input
• Conditional flow: Use condition nodes to route AI responses to different paths
• Error handling: Connect AI nodes to error handling nodes to handle failures gracefully

Each AI node has specific input and output requirements. The system validates connections as you make them, ensuring data types match and connections make sense.