Passing data between nodes
Every node can use the outputs of nodes that run before it. In text fields, type/ to insert a reference to an earlier node’s output, or to one field of it when the output has a schema. References are resolved when the workflow runs.
Workflows run from the Input node through the graph in dependency order. Branches that a Condition doesn’t take are skipped. Cycles aren’t allowed.
Input
The entry point. Define the workflow’s parameters as Input Fields: each has a variable name, type, description and optional default value. When the assistant calls the workflow as a tool, it provides values matching this schema, so describe each field clearly.Output
The exit point. Choose which values from earlier nodes to return. The result is what the assistant receives from the tool call.LLM
Generates text with a model.
Structured output is useful for extraction, classification and reshaping data, not just writing.
Condition
Branches the workflow with if, else if and else. Each condition compares an earlier value using an operator for its type:Tool
Runs a tool from a connector. Pick the tool and write instructions in the message; a model fills in the tool’s parameters from your instructions and the referenced data.HTTP
Calls an external API or webhook.
For security, requests to private or internal network addresses are refused.
Template
Builds text, such as an email or report, by mixing literal text with/ references to earlier outputs.