Steps are the building blocks of a tool. Each step does one thing: fetch information, talk to the customer, transform data, or decide where the flow goes next. You combine them on the canvas to build the tool's logic.
The steps
Steps are grouped by what they do.
Reuse
-
Run existing tool runs a tool you already built as a step inside this one. Build something once, use it everywhere.
Get information
-
Search knowledge finds information in your knowledge base.
Get input
-
Ask user asks the customer one question and validates the answer.
-
Prompt agent lets AI collect multiple details from the customer in a natural conversation.
Process data
-
Call API connects to an external system to fetch or send data.
-
Prompt AI model generates, classifies, extracts, or transforms text with an AI model.
-
Run code runs Python for custom logic or calculations.
Control flow
-
If / else branch continues down one of several paths based on a condition.
Finish
-
Return result sends the final output back to the agent.
-
Handover transfers the conversation to a human or another system.
What all steps share
-
Steps that produce a result store it in a variable. Choose where with Select output variable; later steps can then use that value. Most tools only need the default
resultvariable. -
Any text field in a step can contain parameters, variables, and context values. Type / to insert one; the value is filled in when the tool runs.
See Parameters, variables, context and data flow for how data moves through a tool.