How to save and load a pipeline in Haystack

Haystack pipelines often start as Python objects in notebooks or scripts, but review, deployment, and repeat runs need a portable definition. Saving a pipeline to YAML captures component configuration and graph wiring so the same pipeline shape can be loaded outside the original Python process.

Pipeline.dump() writes the serialized definition to a file-like stream, and Pipeline.load() reads that stream back into a new Pipeline object. The file records component types, initialization parameters, connections, and pipeline metadata rather than generated answers or temporary runtime output.

Secret-bearing components need extra care before serialization. Use environment-backed Secret objects for credentials and review the saved YAML before sharing it, because the pipeline file is source-controlled configuration that another process may later load and execute.

1. Create save_support_pipeline.py with a serializable DocumentCleaner pipeline.

   save_support_pipeline.py

   from pathlib import Path
    
   from haystack import Document, Pipeline
   from haystack.components.preprocessors import DocumentCleaner
    
   pipeline_file = Path("support_cleaner_pipeline.yaml")
    
   pipeline = Pipeline()
   pipeline.add_component(
       "cleaner",
       DocumentCleaner(remove_empty_lines=True, remove_extra_whitespaces=True),
   )
    
   result = pipeline.run(
       {
           "cleaner": {
               "documents": [Document(content=" Reset   your password\n\n")],
           }
       }
   )
    
   with pipeline_file.open("w", encoding="utf-8") as file:
       pipeline.dump(file)
    
   print(f"original_content={result['cleaner']['documents'][0].content}")
   print(f"saved_file={pipeline_file}")

   DocumentCleaner is a built-in component, so the load process can recreate it from the component type stored in the saved YAML.

2. Run the save script.

   $ python save_support_pipeline.py
   original_content=Reset your password
   saved_file=support_cleaner_pipeline.yaml

   Use the same Python environment where haystack-ai is installed.
   Related: How to install Haystack with pip

3. Inspect the saved YAML definition.

   $ cat support_cleaner_pipeline.yaml
   components:
     cleaner:
       init_parameters:
         ascii_only: false
         keep_id: false
         remove_empty_lines: true
         remove_extra_whitespaces: true
         remove_regex: null
         remove_repeated_substrings: false
         remove_substrings: null
         replace_regexes: null
         strip_whitespaces: false
         unicode_normalization: null
       type: haystack.components.preprocessors.document_cleaner.DocumentCleaner
   connection_type_validation: true
   connections: []
   max_runs_per_component: 100
   metadata: {}

   The saved file should show the component type and initialization parameters before it is loaded elsewhere.
   Tool: YAML Validator

4. Create load_support_pipeline.py to load the saved pipeline file.

   load_support_pipeline.py

   from pathlib import Path
    
   from haystack import Document, Pipeline
    
   pipeline_file = Path("support_cleaner_pipeline.yaml")
    
   with pipeline_file.open("r", encoding="utf-8") as file:
       pipeline = Pipeline.load(file)
    
   result = pipeline.run(
       {
           "cleaner": {
               "documents": [Document(content=" Reset   your password\n\n")],
           }
       }
   )
    
   expected_content = "Reset your password"
    
   print(f"loaded_type={type(pipeline).__name__}")
   print(f"loaded_content={result['cleaner']['documents'][0].content}")
   print(f"round_trip_match={result['cleaner']['documents'][0].content == expected_content}")

   Load only pipeline files from trusted project sources. Loading a YAML pipeline asks Haystack to instantiate the component classes named in the file, and project-defined components must be importable in the loading environment.

5. Run the load script from the directory that contains support_cleaner_pipeline.yaml.

   $ python load_support_pipeline.py
   loaded_type=Pipeline
   loaded_content=Reset your password
   round_trip_match=True

   round_trip_match=True confirms that the restored pipeline produced the same cleaned document content as the original pipeline.