Yes, docling-serve has S3 support through dedicated batch and regular convert endpoints, depending on your configuration.

S3 Integration via Endpoints#

S3 sources can be used in two contexts:

Regular Convert Endpoint (POST /v1/convert/source)#

• S3 sources are supported only with the KFP engine (DOCLING_SERVE_ENG_KIND=KFP).
• S3 sources require S3Target — presigned URL targets are not supported on this endpoint.
• Use ConvertSourcesRequest with S3SourceRequest and S3Target.

Batch Endpoint (POST /v1/convert/source/batch)#

• S3 sources are available regardless of engine type.
• S3 sources require S3Target — presigned URL targets are not supported.
• Use BatchConvertSourcesRequest with S3SourceRequest and S3Target.
• This endpoint is async-only and returns a TaskStatusResponse.

When using S3 sources:

• The S3SourceRequest requires: endpoint, access_key, secret_key, and bucket.
• Optional: max_num_elements limits the number of S3 objects to iterate for the source. If not specified, all objects matching the criteria will be processed. Must be >= 1 if provided.
• S3-compatible systems like MinIO or IBM COS may work by pointing to a custom endpoint, though this is not explicitly documented.
• Set DOCLING_SERVE_ARTIFACT_STORAGE_VERIFY_SSL=false for local HTTP or self-signed MinIO setups.

Python SDK Support#

The DoclingServiceClient provides a submit_batch() method for S3-based batch processing:

from docling.service_client import DoclingServiceClient
from docling.datamodel.service.requests import S3SourceRequest
from docling.datamodel.service.targets import S3Target

client = DoclingServiceClient(base_url="http://localhost:5000")

sources = [
    S3SourceRequest(
        endpoint="s3.amazonaws.com",
        access_key="your-key",
        secret_key="your-secret",
        bucket="my-bucket",
        key="path/to/document.pdf"
    )
]

job = client.submit_batch(
    sources=sources,
    target=S3Target(
        endpoint="s3.amazonaws.com",
        access_key="your-key",
        secret_key="your-secret",
        bucket="converted-bucket"
    ),
    output_formats=[OutputFormat.DOCLING_V2]
)
result = job.result(timeout=300)

Result Delivery Options#

PresignedUrlTarget#

The PresignedUrlTarget allows docling-serve to upload conversion results directly to S3 using presigned URLs:

• Available on regular convert endpoints (/v1/convert/source, /v1/convert/file) and the batch endpoint with HTTP sources (not S3 sources).
• Requires artifact storage to be enabled on the server (DOCLING_SERVE_ARTIFACT_STORAGE_ENABLED=true).
• The server returns artifact manifests with presigned URLs for each output format.
• You can download conversion results directly from S3 without fetching them through the server.
• More efficient for large documents or batch processing.

The result type depends on your target:

• PresignedUrlTarget → PresignedUrlConvertResponse with per-document artifact manifests
• S3Target → PresignedUrlConvertDocumentResponse with counts only (artifacts delivered remotely)

Workaround: Fetch and Upload#

If you need to process S3 documents outside the batch workflow, you can fetch the document yourself and pass it as a file upload:

file_bytes = s3_client.get_object(...)['Body'].read()
file_stream = BytesIO(file_bytes)
doc_stream = DocumentStream(stream=file_stream, name=filename, format=input_format)
result = doc_converter.convert(doc_stream)