Choosing a Configuration

A processing configuration defines what Document AI should do with a document and which structured values it should return.

Configuration choices

Document AI supports three ways to choose processing behavior:

Choice	Best for	Reusable
Platform-managed configuration	Common Shipwell document workflows	Yes
Custom named configuration	A repeatable workflow owned by your company	Yes
One-time custom settings	Testing or a request-specific extraction	No

For uploads, you can also choose Upload only to store files without creating a processing run.

What a configuration controls

A configuration can control:

Extraction schemas — the structured values to extract
Split — whether one PDF containing multiple logical documents should produce separate result slices
Embeddings — whether processed content can be found through semantic search
Auto-description — whether to generate a document description
Key facts — whether the generated description should include high-signal facts

Named configurations

Named configurations are selected with processing_config_id.

The list visible to your account includes:

Platform-managed configurations available to all eligible accounts
Custom configurations owned by your company

Only enabled configurations can be selected for uploads or manual processing.

The selected configuration is used as a complete snapshot for the new run. Document AI does not merge it with another configuration or automatically add extraction schemas.

Use GET /document-store/configs/processing-configs to list the configurations visible to your account.

The is_platform_managed field distinguishes read-only Shipwell configurations from company-owned configurations.

One-time custom settings

One-time settings are supplied as pipeline_config on an upload or reprocessing request. They are useful while testing an extraction schema, but they are not saved for later selection.

Copy

Copied

{
  "pipeline_config": {
    "schemas": [
      {
        "name": "purchase_order_number",
        "prompt": "Extract the purchase order number.",
        "json_schema": {"type": "string"},
        "scope": "document",
        "result_cardinality": "single",
        "include_source_regions": true
      }
    ],
    "split": false,
    "embed": true,
    "describe": {
      "enabled": true,
      "key_facts": true
    }
  }
}

Do not provide pipeline_config and processing_config_id in the same request.

Which option should I use?

Use a platform-managed configuration when its documented purpose matches the documents you process.

Create a custom named configuration when:

Multiple users or integrations need the same extraction behavior
You want to manage the workflow in Shipwell
You need an enabled/disabled lifecycle for the configuration

Use one-time custom settings when:

Prototyping a schema
Running a request-specific extraction
Testing changes before updating a named configuration