Skip to main content

​
Description

Input mapping defines how nodes connect to each other in an Agent Composer workflow. It specifies which outputs from previous nodes flow into the inputs of subsequent nodes.

​
Syntax

Input mappings use the syntax source_node#output_key: 
input_mapping:
  current_node_input: earlier_node#output_key
This connects the output_key output from earlier_node to the current_node_input input of the current node.

​
Accessing Graph Inputs

Every graph has an implicit input node named __inputs__. Access graph-level inputs via: 
input_mapping:
  query: __inputs__#query

​
Example

nodes:
  # First node receives input from the graph
  search:
    type: SearchUnstructuredDataStep
    input_mapping:
      query: __inputs__#query

  # Second node receives input from the first node
  rerank:
    type: RerankRetrievalsStep
    input_mapping:
      query: __inputs__#query
      retrievals: search#retrievals

  # Third node receives inputs from multiple nodes
  generate:
    type: ResponseGenerationStep
    input_mapping:
      query: __inputs__#query
      retrievals: rerank#retrievals

  # Output node collects final outputs
  __outputs__:
    type: output
    input_mapping:
      response: generate#response

​
Rules

  • All inputs of a node must be connected
  • Unused outputs are allowed
  • A node can receive inputs from multiple source nodes
  • The same output can be connected to multiple downstream nodes

​
Multiple Inputs from Same Source

You can connect multiple inputs from the same source node: 
generate:
  type: ResponseGenerationStep
  input_mapping:
    query: __inputs__#query
    retrievals: search#retrievals
    translate_needed: translate#translate_needed
    detected_language: translate#detected_language