Skip to main content

​
Description

Subgraphs are reusable components that can be referenced in larger workflows. They allow you to define common patterns once and use them multiple times, improving maintainability and reducing duplication.

​
Creating Subgraphs

Define subgraphs at the same level as the main graph with their own inputs, outputs, and nodes: 
# Define a reusable retrieval subgraph
document_retrieval:
  inputs:
    query: str
    threshold: float
  outputs:
    documents: List[str]
    quality_score: float
  nodes:
    search:
      type: SearchUnstructuredDataStep
      input_mapping:
        query: __inputs__#query
    
    quality_check:
      type: QualityCheckStep
      config:
        threshold: 0.7
      input_mapping:
        results: search#retrievals
    
    __outputs__:
      type: output
      input_mapping:
        documents: search#retrievals
        quality_score: quality_check#score

​
Using Subgraphs

Reference subgraphs using the subgraph: prefix in the type field: 
# Main workflow using the subgraph
inputs:
  user_query: str

outputs:
  final_result: str

nodes:
  constant_threshold:
    type: ConstantStep
    config:
      value: 0.8

  retriever:
    type: subgraph:document_retrieval
    input_mapping:
      query: __inputs__#user_query
      threshold: constant_threshold#output

  __outputs__:
    type: output
    input_mapping:
      final_result: retriever#documents

​
Subgraph Reference Syntax

node_name:
  type: subgraph:subgraph_name
  input_mapping:
    subgraph_input: source_node#output

​
Features

FeatureDescription
NestingSubgraphs can contain other subgraphs
ReusabilityOne subgraph can be called multiple times in the same workflow
EncapsulationInternal node names don’t conflict with parent graph
Typed I/OSubgraphs define their own typed inputs and outputs

​
Example: Multiple Subgraph Calls

# Subgraph definition
search_and_rerank:
  inputs:
    query: str
  outputs:
    retrievals: Retrievals
  nodes:
    search:
      type: SearchUnstructuredDataStep
      input_mapping:
        query: __inputs__#query
    rerank:
      type: RerankRetrievalsStep
      input_mapping:
        query: __inputs__#query
        retrievals: search#retrievals
    __outputs__:
      type: output
      input_mapping:
        retrievals: rerank#retrievals

# Main workflow calling the subgraph twice
inputs:
  query: str

outputs:
  response: str

nodes:
  # First call with original query
  primary_search:
    type: subgraph:search_and_rerank
    input_mapping:
      query: __inputs__#query

  # Second call with expanded query
  expanded_search:
    type: subgraph:search_and_rerank
    input_mapping:
      query: query_expansion#expanded_query

  # Merge results
  merge:
    type: MergeRetrievalsStep
    input_mapping:
      retrievals1: primary_search#retrievals
      retrievals2: expanded_search#retrievals