Docling Document Serializer

[az-ihsan]

Hi,

I am working on safety and manual documents which has the same format on each page with a header. basic question, can we customize docling markdown serializer given the DoclingDocument?

Or what is the best practice to have the docling document for a specific style of document?

any lead would be appreciated.

Cheers
Ahm

[dosubot[bot]]

Hi @az-ihsan! I'm Dosu and I’m helping the docling team.

Yes, you can customize the Docling markdown serializer for your documents. The recommended approach is to subclass MarkdownTableSerializer and override its serialize method to control how table content is rendered. You can then inject your custom serializer into your workflow by subclassing MarkdownDocSerializer or by providing it via a serializer provider mechanism. Example code and details are available here and here.

For documents with consistent headers (like safety/manual docs), Docling stores headers and footers as "furniture" items. By default, these are omitted during iteration, but you can include them by specifying included_content_layers={ContentLayer.BODY, ContentLayer.FURNITURE} when iterating over your DoclingDocument items. See usage details here.

For advanced customization and examples, check out the advanced chunking and serialization documentation.

To reply, just mention @dosu.

---

Share context across your team and agents. Try Dosu.

[az-ihsan]

Hi, actually the header is not missing, I can see in the doctags file yielded from the docling document with the tag of , I just need a hint in order to post-process this doctags?

do you have any idea?

cheers,
Ahm

[dosubot[bot]]

You can post-process <otsl> doctags (which represent tables) using Docling's built-in parsing utilities. The recommended approach is to use the functions in docling_core/types/doc/utils.py—specifically, otsl_extract_tokens_and_text to extract tokens and text from the OTSL string, and otsl_parse_texts to convert those into table cells, handling cell spans and merges. For a higher-level interface, use parse_otsl_table_content(otsl_content) to get a TableData object with rows, columns, and cell content, which you can then manipulate or export as needed (e.g., to markdown or HTML) using DoclingDocument methods. This ensures robust and lossless conversion from doctags to structured Python objects for further processing. Example and details here.

If you want a Python code sketch or have a specific post-processing goal (extract, modify, transform), let me know your environment and what you want to achieve!

To reply, just mention @dosu.

---

Share context across your team and agents. Try Dosu.

[az-ihsan]

OK great, tthanks! I just want to make it clear for my next steps

Could you maybe sketch a post-processing from the doctags into markdown with a particular header of my document is transformed into a text or a metadata section in my markdown?

Cheers,

[dosubot[bot]]

Here's a Python sketch for post-processing doctags into markdown, with your header table (from <otsl>) transformed into either plain text or a YAML metadata section at the top of your markdown file.

from docling_core.types.doc.document import DoclingDocument, DocItemLabel, ContentLayer
import yaml

# Load doctags into DoclingDocument
doc = DoclingDocument.load_from_doctags(doctags_document)

# Find header tables (commonly in ContentLayer.FURNITURE or first table per page)
header_tables = []
for item, _ in doc.iterate_items(included_content_layers={ContentLayer.FURNITURE, ContentLayer.BODY}):
    if getattr(item, "label", None) == DocItemLabel.TABLE:
        # You may want to refine this filter for your specific header location
        header_tables.append(item)

# Option 1: Convert header table to plain text
if header_tables:
    header_table = header_tables[0]
    header_text = "\n".join(
        [" | ".join(cell.text for cell in row) for row in header_table.data.grid]
    )

# Option 2: Convert header table to YAML metadata (frontmatter)
if header_tables:
    header_table = header_tables[0]
    # Assuming first row is keys, second row is values
    keys = [cell.text for cell in header_table.data.grid[0]]
    values = [cell.text for cell in header_table.data.grid[1]]
    metadata = dict(zip(keys, values))
    yaml_frontmatter = "---\n" + yaml.dump(metadata) + "---\n"

# Serialize the rest of the document to markdown (excluding header if needed)
# You can filter out header tables by their self_ref or by skipping the first table
markdown_body = doc.export_to_markdown(
    included_content_layers={ContentLayer.BODY},
    # Optionally, filter out header table refs
)

# Combine header and body
markdown_output = f"{yaml_frontmatter}\n{markdown_body}"  # or use header_text

# Save to file
with open("output.md", "w", encoding="utf-8") as f:
    f.write(markdown_output)

You can adapt the header extraction logic depending on your document's structure. The key steps are: load doctags, identify header tables, extract their content, format as text or YAML, and prepend to your markdown output. For more details on parsing <otsl> and working with TableData, see the docling-core utils.py and document.py.

Let me know if you want to tweak the extraction logic or need help with a specific header format!

To reply, just mention @dosu.

---

Docs are dead. Just use Dosu.