Skip to main content

​
Description

Config overrides allow configuration parameters to be determined dynamically at execution time, rather than being fixed in the YAML. This enables workflows where configuration values depend on runtime inputs or outputs from other nodes.

​
Syntax

Use config_overrides to source configuration values from graph inputs or node outputs: 
search_node:
  type: SearchUnstructuredDataStep
  config:
    filter_retrievals: true
  config_overrides:
    top_k: __inputs__#top_k
  input_mapping:
    query: __inputs__#query

​
Sourcing from Graph Inputs

Override values can come from graph-level inputs: 
inputs:
  query: str
  top_k: int

nodes:
  search:
    type: SearchUnstructuredDataStep
    config_overrides:
      top_k: __inputs__#top_k
    input_mapping:
      query: __inputs__#query

​
Sourcing from Node Outputs

Override values can also come from outputs of other nodes: 
nodes:
  constant_k:
    type: ConstantStep
    config:
      value: 20

  search:
    type: SearchUnstructuredDataStep
    config_overrides:
      top_k: constant_k#output
    input_mapping:
      query: __inputs__#query

​
Example: Dynamic Temperature

nodes:
  # Determine temperature based on query type
  temperature_selector:
    type: LanguageModelStep
    config:
      prompt_template: |
        Analyze this query and return a temperature value between 0 and 1.
        For factual queries, use 0.1. For creative queries, use 0.8.
        Query: {query}
        Return only the number.
      model_id: "vertex_ai/gemini-2.0-flash"
    input_mapping:
      query: __inputs__#query

  generate:
    type: GenerationStep
    config_overrides:
      temperature: temperature_selector#response
    input_mapping:
      message_history: create_message_history#message_history

​
Combining Config and Config Overrides

Static and dynamic configuration can be combined on the same node: 
search:
  type: SearchUnstructuredDataStep
  config:
    # Static configuration
    lexical_alpha: 0.1
    semantic_alpha: 0.9
    reranker: "ctxl-rerank-v2-instruct-multilingual-FP8"
  config_overrides:
    # Dynamic configuration
    top_k: dynamic_k#output
    rerank_top_k: dynamic_rerank_k#output
  input_mapping:
    query: __inputs__#query