This guide helps you migrate your code from previous versions of the Aurelio SDK to v0.0.19, which introduces significant changes to the extract endpoint.

​
Key Changes

​
Deprecated: quality parameter

The quality parameter previously used with values "low" and "high" for both PDF and MP4 file extraction has been deprecated.

​
New: model parameter

The new model parameter replaces quality and provides more granular control over the extraction process with different model options.

​
New: Enhanced processing options for video files

For MP4 files, chunking preferences previously set through the quality parameter are now explicitly configured in the processing_options parameter.

​
Migration Table

File TypeOld ApproachNew ApproachNotes
PDFquality="low"model="aurelio-base"Fastest, cheapest option for clean PDFs
PDFquality="high"model="docling-base"Code-based OCR method for high precision
PDF-model="gemini-2-flash-lite"New! State-of-the-art VLM-based extraction
MP4quality="low", chunk=Truemodel="aurelio-base", processing_options={"chunking": {"chunker_type": "regex"}}Basic chunking for videos
MP4quality="high", chunk=Truemodel="aurelio-base", processing_options={"chunking": {"chunker_type": "semantic"}}Semantic chunking for videos

​
Code Examples

​
PDF Extraction

​
Before (pre-v0.0.19):

# Fast extraction for simple PDFs
response = client.extract_file(
    file_path="document.pdf",
    quality="low",
    chunk=True,
    wait=30
)

# Higher quality extraction for complex PDFs
response = client.extract_file(
    file_path="complex_document.pdf",
    quality="high",
    chunk=True,
    wait=30
)

​
After (v0.0.19+):

# Fast extraction for simple PDFs
response = client.extract_file(
    file_path="document.pdf",
    model="aurelio-base",  # Equivalent to the old quality="low"
    chunk=True,
    wait=30
)

# Higher quality extraction for complex PDFs
response = client.extract_file(
    file_path="complex_document.pdf",
    model="docling-base",  # Equivalent to the old quality="high"
    chunk=True,
    wait=30
)

# NEW: State-of-the-art extraction using VLM
response = client.extract_file(
    file_path="document.pdf",
    model="gemini-2-flash-lite",  # New option not available before
    chunk=True,
    wait=30
)

​
Video Extraction

​
Before (pre-v0.0.19):

# Basic video extraction with regex chunking
response = client.extract_file(
    file_path="video.mp4",
    quality="low",
    chunk=True,
    wait=-1
)

# Video extraction with semantic chunking
response = client.extract_file(
    file_path="video.mp4",
    quality="high",
    chunk=True,
    wait=-1
)

​
After (v0.0.19+):

# Basic video extraction with regex chunking
response = client.extract_file(
    file_path="video.mp4",
    model="aurelio-base",  # Only supported model for videos
    chunk=True,
    wait=-1,
    processing_options={
        "chunking": {
            "chunker_type": "regex"  # Equivalent to old quality="low"
        }
    }
)

# Video extraction with semantic chunking
response = client.extract_file(
    file_path="video.mp4",
    model="aurelio-base",  # Only supported model for videos
    chunk=True,
    wait=-1,
    processing_options={
        "chunking": {
            "chunker_type": "semantic"  # Equivalent to old quality="high"
        }
    }
)

​
URL Extraction

The changes for extract_url are identical to those for extract_file - replace the quality parameter with the appropriate model parameter, and for videos, specify chunking preferences in processing_options.

​
VLM-based Extraction (New Feature)

The new gemini-2-flash-lite model uses a Vision Language Model to process PDF content, offering state-of-the-art accuracy. This can be especially valuable for:

  • Scanned documents with complex layouts
  • Documents with tables, charts, and diagrams
  • Documents where context and visual understanding are important
response = client.extract_file(
    file_path="complex_scanned_document.pdf",
    model="gemini-2-flash-lite",
    chunk=True,
    wait=60  # May require longer processing time
)

Note: As mentioned in the OCR in Large Multimodal Models paper, VLMs like Gemini can occasionally hallucinate content. While hallucinations are rare, the model is designed for high-recall, potentially with lower precision than code-based OCR methods.

​
Pricing Considerations

  • aurelio-base pricing is equivalent to the old low quality setting
  • Both docling-base and gemini-2-flash-lite are priced equivalent to the old high quality setting

​
Transition Period

The quality parameter will continue to work during a transition period but will be removed in a future release. We recommend updating your code to use the new model parameter as soon as possible.