Mythosia.Documents.Abstractions 1.2.0

Mythosia.Documents.Abstractions

Core document abstractions for structured document loading and parsing. Framework-agnostic — usable with any RAG pipeline or document processing system.

Installation

dotnet add package Mythosia.Documents.Abstractions

Key Types

DoclingDocument

Unified structured document representation following the docling convention. Content items are stored in flat lists; the tree structure is maintained via body/furniture root nodes.

using Mythosia.Documents;
using Mythosia.Documents.Elements;

var doc = new DoclingDocument
{
 Name = "report",
 Source = "docs/report.pdf",
};

// Builder API
doc.AddTitle("Annual Report");
doc.AddHeading("Revenue", level: 2);
doc.AddParagraph("Total revenue increased by 15%.");
doc.AddCode("var x = 42;", language: "csharp");

// Export to Markdown
string markdown = doc.ToMarkdown();

// Optional: override table rendering strategy
doc.TableSerializer = new SemanticTableSerializer();
string semanticMarkdown = doc.ToMarkdown();

For plain-text content that should be preserved as-is, use RawContent:

var doc = new DoclingDocument
{
 Name = "notes",
 Source = "notes.txt",
 RawContent = rawText, // ToMarkdown() returns this directly
};

Markdown Serialization

DoclingDocument.ToMarkdown() uses MarkdownSerializer to render the body tree. Body text is escaped by default so source text such as *literal*, [brackets], | pipes, and backticks stays literal Markdown content instead of becoming formatting.

using Mythosia.Documents.Elements;

var doc = new DoclingDocument();
doc.AddParagraph("Keep *this* literal and preserve [brackets].");

string safeMarkdown = doc.ToMarkdown();
// Keep \*this\* literal and preserve \[brackets\].

var serializer = new MarkdownSerializer
{
 EscapeText = false,
};

string rawMarkdown = serializer.Serialize(doc);

MarkdownSerializer also clamps heading output to Markdown # through ###### and inserts a blank line when a list is followed by another block element, preventing the next paragraph, heading, table, code block, formula, or image placeholder from being absorbed into the list.

Table Serialization

Table rendering is pluggable via ITableSerializer. The default is GridTableSerializer (standard Markdown pipe table). Switch to SemanticTableSerializer for form-style documents:

using Mythosia.Documents.Elements;

// Default: pipe table
var doc = new DoclingDocument { Name = "report" };
string md = doc.ToMarkdown(); // uses GridTableSerializer

// Semantic: bold group labels for form-style tables
doc.TableSerializer = new SemanticTableSerializer();
string md2 = doc.ToMarkdown(); // uses SemanticTableSerializer

IDocumentLoader

public interface IDocumentLoader
{
 Task<IReadOnlyList<DoclingDocument>> LoadAsync(
 string source, CancellationToken cancellationToken = default);
}

IDocumentParser

public interface IDocumentParser
{
 bool CanParse(string source);
 Task<DoclingDocument> ParseAsync(string source, CancellationToken ct = default);
}

Element Types (Mythosia.Documents.Elements)

Related Packages