Home
Guides

Static Agent Export and Import

Example usage of Static Agent Export and Import APIs

Suggest Edits

Introduction

This new feature allows users to export an agent from one workflow, and import a static copy of it into a different workflow, on the same cluster or on a different one (e.g. to promote an agent to a higher environment). The imported agent does not need to be re-trained.


Preconditions

  • The feature flag modelExportEnabled must be enabled on the cluster.
  • The user creating the export must have Manager access to the cluster
  • The user importing the agent must have Manage All Data permissions
  • The agent to be exported must successfully trained

Types of Agents Supported

  • The following types of agents may be exported:
  • Image and Document Classification (Custom V1)
  • Document Extraction (Custom V2)
  • Form Extraction

📘

Note: All OCR providers are supported.


Types of Agents Not Supported

  • Image and Document Multi-Classification (Custom V1)
  • Image Extraction (object detection)
  • Unbundling (doc-splitting)
  • GenAI OOTB agents
  • GenAI Extraction agents

Supported Functionality of Imported Agents

  • Submissions to imported agents can be sent and reviewed. Submission result files will contain predictions and labels in the “component_results” section, not the “model_results” section.
  • Filters can be added to imported Classification agents.
  • Linked Labels can be added to imported Document Extraction agents.
  • Imported agents can be removed from a workflow.

Limitations of Imported Agents

  • Imported agents are Static. They cannot be customized or modified.
  • There is no corresponding Agent task page for an imported agent.
  • Imported agents cannot be re-trained.
  • Metrics are not available for imported agents.
  • Fields in an imported agent cannot be deleted, added, or modified.
  • Import and Export of agents is available via API only (not in the UI).

Agent Export

Call the CreateModelExport class to create an export of an agent in a workflow:

def **init**(self, 
             model_id: int,
             wait: bool = True,
             request_interval: int | float = 5) -> None

Create a model export.
Available on 6.14+ only.
Params:  model_id - the model id.
         wait - wait for the export to complete.  Defaults to True.
         request_interval - the interval between requests in seconds.  Defaults to 5.
Copied from: CreateModelExport

Agent export will produce a ZIP file that contains:

  • model.pkl: the agent pickle file
  • Metadata.json: a file of metadata with information about the agent.

🚧

Note that these files should not be modified after export.


{  
   "task_type":"annotation",  
   "model_type":"table",  
   "model_options":{  
      "id":802,  
      "domain":"standard",  
      "high_quality":false,  
      "interlabeler_resolution":"all",  
      "sampling_strategy":"random_oversample",  
      "seed":9823,  
      "test_split":0.2,  
      "weight_by_class_frequency":null,  
      "word_predictor_strength":null,  
      "model_training_options":{  
         "use_partial_data":true,  
         "use_autolabeled_data":true  
      },  
      "predict_options":null  
   },`
`   "labelset":{  
      "id":933,  
      "target_names":[  
         {  
            "created_at":1730079573.410494,  
            "created_by":4,  
            "updated_at":1730079573.410494,  
            "updated_by":null,  
            "id":2930,  
            "labelset_id":933,  
            "name":"Year",  
            "position":0,  
            "active":true,  
            "target_type":"text"  
         },  
         {  
            "created_at":1730079573.410494,  
            "created_by":4,  
            "updated_at":1730079573.410494,  
            "updated_by":null,  
            "id":2931,  
            "labelset_id":933,  
            "name":"TotalExpenses",  
            "position":1,  
            "active":true,  
            "target_type":"text"  
         },`
`         {  
            "created_at":1730079573.410494,  
            "created_by":4,  
            "updated_at":1730079573.410494,  
            "updated_by":null,  
            "id":2932,  
            "labelset_id":933,  
            "name":"TotalIncome",  
            "position":2,  
            "active":true,  
            "target_type":"text"  
         }  
      ],`
`      "fields":\[  
         {  
            "id":218768,  
            "name":"Year",  
            "workflow_id":689,  
            "datatype":"text",  
            "input_config":{  
               "strip_whitespace":false,  
               "strip_punctuation":false  
            },  
            "format_config":{  
            },  
            "validation_config":[  
            ],  
            "multiple":true,  
            "required":true  
         },`
`         {  
            "id":218769,  
            "name":"TotalExpenses",  
            "workflow_id":689,  
            "datatype":"text",  
            "input_config":{  
               "strip_whitespace":false,  
               "strip_punctuation":false  
            },  
            "format_config":{  
            },  
            "validation_config":[  
            ],  
            "multiple":true,  
            "required":true  
         },`
`         {  
            "id":218770,  
            "name":"TotalIncome",  
            "workflow_id":689,  
            "datatype":"text",  
            "input_config":{  
               "strip_whitespace":false,  
               "strip_punctuation":false  
            },  
            "format_config":{  
            },  
            "validation_config":[  
            ],  
            "multiple":true,  
            "required":true  
         }  
      ],`
`      "field_links":[  
         {  
            "id":386768,  
            "workflow_id":689,  
            "component_id":1604,  
            "target_id":2930,  
            "target_name":"Year",  
            "field_id":218768  
         },  
         {  
            "id":386769,  
            "workflow_id":689,  
            "component_id":1604,  
            "target_id":2931,  
            "target_name":"TotalExpenses",  
            "field_id":218769  
         },  
         {  
            "id":386770,  
            "workflow_id":689,  
            "component_id":1604,  
            "target_id":2932,  
            "target_name":"TotalIncome",  
            "field_id":218770  
         }  
      ]  
   },`
`   "dataset":{  
      "id":737,  
      "type":"document",  
      "ocr_engine":"readapi_tables_v1",  
      "configs":{  
         "nest":false,  
         "chars":[  
            "doc_index",  
            "page_num",  
            "page_index",  
            "text",  
            "block_index",  
            "confidence",  
            "position"  
         ],  
         "pages":[  
            "dpi",  
            "image",  
            "thumbnail",  
            "page_num",  
            "doc_offset",  
            "ocr_statistics",  
            "size",  
            "text"  
         ],  
         "blocks":[  
            "doc_offset",  
            "page_num",  
            "page_offset",  
            "text",  
            "block_type",  
            "position"  
         ],`
`         "tokens":[  
            "block_offset",  
            "doc_offset",  
            "page_num",  
            "style",  
            "page_offset",  
            "text",  
            "position"  
         ],  
         "languages":[  
            "ENG"  
         ],  
         "top_level":"page",  
         "ocr_engine":"readapi_tables_v1",  
         "reblocking":[  
            "style",  
            "list",  
            "inline-header"  
         ],  
         "auto_rotate":true,  
         "excel_tables":false,  
         "include_lines":false,  
         "single_column":false,  
         "ocr_statistics":[  
            "mean_confidence",  
            "mode_confidence",  
            "median_confidence"  
         ],`
`         "table_read_order":"row",  
         "spreadsheet_converter_version":4  
      }  
   },  
   "source_host":"<https://clustername.us-east-1.indico-dev.indico.io">,  
   "source_ipa_version":"IPA-6.14.0",  
   "workflow_id":689,  
   "model_group_id":723,  
   "model_id":1324,  
   "component_id":1604,  
   "source_column_id":822,  
   "created_by":4,  
   "created_at":1737746726.994178  
}

Agent Import

Call the AddStaticModelComponent class to import the model pickle file into a workflow:

def __init__(self,
             workflow_id: int,
             after_component_id: int | None = None,
             after_component_link_id: int | None = None,
             static_component_config: dict[str, Any] | None = None,
             auto_process: bool = False,
             export_file: str | None = None) -> None

Add a static model component to a workflow.
Available on 6.14+ only.
Copied from: AddStaticModelComponent

Example Script

This is a sample script to export an agent from one workflow and import it into another. This script uses the Python SDK .

from indico import IndicoClient, IndicoConfig
from indico.queries import AddStaticModelComponent, CreateModelExport, GetWorkflow
from indico.types.model_export import ModelExport
import os
import requests
from pathlib import Path

# Constants for workflow and model IDs
EXPORT_WORKFLOW_ID = 689
IMPORT_WORKFLOW_ID = 690
MODEL_ID = 1324

# Configuration for the Indico client
my_config = IndicoConfig(host="cluster.us-east-1.indico-dev.indico.io",
                         api_token_path="./indico_api_token.txt")

# Instantiate the Indico client
client = IndicoClient(config=my_config)

def download_export(export: ModelExport) -> Path:
   """
   Download the model export file from the provided signed URL.

   Args:
       export (ModelExport): The model export object containing the signed URL and file path.

   Returns:
       Path: The path to the downloaded file.
   """
   file_name = os.path.join(os.getcwd(), Path(export.file_path).name.replace(".indico", ".zip"))
   response = requests.get(export.signed_url, stream=True)
   with open(file_name, "wb") as dl_path:
       for chunk in response.iter_content(chunk_size=8192):
           dl_path.write(chunk)
   return Path(file_name)

def main():
   """
   Main function to add a static model component to an existing workflow.
   """
   # Retrieve the workflow details
   workflow = client.call(GetWorkflow(workflow_id=EXPORT_WORKFLOW_ID))

   # Create a model export from one workflow
   model_export = client.call(CreateModelExport(model_id=MODEL_ID))

   # Download the exported model file
   dled_export = download_export(model_export[0])

   # Add the static model component to another workflow
   client.call(
       AddStaticModelComponent(
           workflow_id=IMPORT_WORKFLOW_ID,           after_component_id=workflow.component_by_type("INPUT_OCR_EXTRACTION").id,
           auto_process=True,
           component_name="imported_model",
           export_file=str(dled_export),
       )
   )

if __name__ == "__main__":
   main()

Common Errors

Export Errors

  • If the agent is not trained: “Only successfully trained models can be exported”.
  • If an agent fails to download: “Model failed to download.”
  • If an agent fails to export: “Model failed to export”
    • If the agent export times out (takes more than 15 minutes): “Model export timed out”.
    • If the agent ID does not exist: “The model ID does not exist or you do not have access to this model”.
    • If the user does not have manager access to the agent: “The model ID does not exist or you do not have access to this model”.

Import Errors

  • Incompatible with valid actions: “Model type must conform to valid actions.”
  • If version incompatibility: “Versions are incompatible.”
  • If dataset incompatibility: “Datasets must match between model types.” Includes dataset settings or dataset provider incompatibility.
  • If file type is incorrect: “File type unsupported.”
  • If a generic failure to upload error: “Model failed to upload.”
  • If the OCR engine is not available: “OCR engine not supported.”

Updated 6 days ago