Output Formats v4.1.0¶

Kreuzberg supports multiple output formats to suit different use cases — from plain text for LLM prompts to structured element arrays for RAG systems to hierarchical document trees for knowledge graphs.

Choose the format that best matches your downstream processing needs:

• Unified (default) — Plain text/markdown output, ideal for LLM prompts and full-text search
• Element-Based — Flat array of typed elements with metadata, suitable for RAG chunking and semantic search
• Document Structure — Hierarchical tree with explicit parent-child references, best for knowledge graphs and structured applications
• PDF Hierarchy — Font size classification into heading levels (H1–H6) for PDFs

Unified Output (Default)¶

By default, Kreuzberg extracts plain text and markdown-formatted content. No configuration is required. The result contains:

• content — Full document text with minimal formatting
• pages — Per-page breakdown for PDFs, DOCX, and PPTX
• tables — Extracted tables in structured format
• images — Image metadata and paths

This format is ideal for:

• Feeding entire documents into LLMs without structured constraints
• Full-text search and indexing
• Simple text processing pipelines

Element-Based Output v4.1.0¶

Segments a document into a flat array of typed elements — titles, paragraphs, tables, list items, code blocks, images, and more. Each element carries a page number and, for text elements in PDFs when hierarchy extraction is enabled, bounding box coordinates.

Use element-based output for RAG chunking, semantic search, or Unstructured.io-compatible pipelines. For hierarchical tree structure, use document structure. For plain text, use the default unified output.

Enable¶

from kreuzberg import extract_file_sync, ExtractionConfig

# Configure element-based output
config = ExtractionConfig(result_format="element_based")

# Extract document
result = extract_file_sync("document.pdf", config=config)

# Access elements
for element in result.elements:
    print(f"Type: {element.element_type}")
    print(f"Text: {element.text[:100]}")

    if element.metadata.page_number:
        print(f"Page: {element.metadata.page_number}")

    if element.metadata.coordinates:
        coords = element.metadata.coordinates
        print(f"Coords: ({coords.left}, {coords.top}) - ({coords.right}, {coords.bottom})")

    print("---")

# Filter by element type
titles = [e for e in result.elements if e.element_type == "title"]
for title in titles:
    level = title.metadata.additional.get("level", "unknown")
    print(f"[{level}] {title.text}")

import { extractFileSync, ExtractionConfig } from '@kreuzberg/node';

// Configure element-based output
const config: ExtractionConfig = {
  outputFormat: "element_based"
};

// Extract document
const result = extractFileSync("document.pdf", null, config);

// Access elements
for (const element of result.elements) {
  console.log(`Type: ${element.elementType}`);
  console.log(`Text: ${element.text.slice(0, 100)}`);

  if (element.metadata.pageNumber) {
    console.log(`Page: ${element.metadata.pageNumber}`);
  }

  if (element.metadata.coordinates) {
    const coords = element.metadata.coordinates;
    console.log(`Coords: (${coords.left}, ${coords.top}) - (${coords.right}, ${coords.bottom})`);
  }

  console.log("---");
}

// Filter by element type
const titles = result.elements.filter(e => e.elementType === "title");
for (const title of titles) {
  const level = title.metadata.additional?.level || "unknown";
  console.log(`[${level}] ${title.text}`);
}

use kreuzberg::{extract_file_sync, ExtractionConfig};
use kreuzberg::types::OutputFormat as ResultFormat;

fn main() -> kreuzberg::Result<()> {
    // Configure element-based output (result_format controls Unified vs ElementBased)
    let config = ExtractionConfig {
        result_format: ResultFormat::ElementBased,
        ..Default::default()
    };

    // Extract document
    let result = extract_file_sync("document.pdf", None, &config)?;

    // Access elements
    if let Some(elements) = result.elements {
        for element in &elements {
            println!("Type: {:?}", element.element_type);
            println!("Text: {}", &element.text[..100.min(element.text.len())]);

            if let Some(page) = element.metadata.page_number {
                println!("Page: {}", page);
            }

            if let Some(coords) = &element.metadata.coordinates {
                println!("Coords: ({}, {}) - ({}, {})",
                    coords.x0, coords.y0, coords.x1, coords.y1);
            }

            println!("---");
        }

        // Filter by element type
        let titles: Vec<_> = elements.iter()
            .filter(|e| matches!(e.element_type, kreuzberg::types::ElementType::Title))
            .collect();

        for title in titles {
            let level = title.metadata.additional.get("level")
                .map(|v| v.as_ref())
                .unwrap_or("unknown");
            println!("[{}] {}", level, title.text);
        }
    }

    Ok(())
}

package main

import (
    "fmt"
    "kreuzberg"
)

func main() {
    // Configure element-based output
    config := &kreuzberg.ExtractionConfig{
        OutputFormat: "element_based",
    }

    // Extract document
    result, err := kreuzberg.ExtractFileSync("document.pdf", config)
    if err != nil {
        panic(err)
    }

    // Access elements
    for _, element := range result.Elements {
        fmt.Printf("Type: %s\n", element.ElementType)

        text := element.Text
        if len(text) > 100 {
            text = text[:100]
        }
        fmt.Printf("Text: %s\n", text)

        if element.Metadata.PageNumber != nil {
            fmt.Printf("Page: %d\n", *element.Metadata.PageNumber)
        }

        if element.Metadata.Coordinates != nil {
            coords := element.Metadata.Coordinates
            fmt.Printf("Coords: (%f, %f) - (%f, %f)\n",
                coords.Left, coords.Top, coords.Right, coords.Bottom)
        }

        fmt.Println("---")
    }

    // Filter by element type
    var titles []kreuzberg.Element
    for _, element := range result.Elements {
        if element.ElementType == "title" {
            titles = append(titles, element)
        }
    }

    for _, title := range titles {
        level, ok := title.Metadata.Additional["level"].(string)
        if !ok {
            level = "unknown"
        }
        fmt.Printf("[%s] %s\n", level, title.Text)
    }
}

require 'kreuzberg'

# Configure element-based output
config = Kreuzberg::Config::Extraction.new(output_format: 'element_based')

# Extract document
result = Kreuzberg.extract_file_sync('document.pdf', config: config)

# Access elements
result.elements.each do |element|
  puts "Type: #{element.element_type}"
  puts "Text: #{element.text[0...100]}"

  puts "Page: #{element.metadata.page_number}" if element.metadata.page_number

  if element.metadata.coordinates
    coords = element.metadata.coordinates
    puts "Coords: (#{coords.left}, #{coords.top}) - (#{coords.right}, #{coords.bottom})"
  end

  puts "---"
end

# Filter by element type
titles = result.elements.select { |e| e.element_type == 'title' }
titles.each do |title|
  level = title.metadata.additional['level'] || 'unknown'
  puts "[#{level}] #{title.text}"
end

library(kreuzberg)

config <- extraction_config(
  result_format = "element_based",
  output_format = "markdown"
)

file_path <- "document.pdf"
result <- extract_file_sync(file_path, config = config)

cat(sprintf("Total elements: %d\n\n", length(result$elements)))

for (i in seq_along(result$elements)) {
  element <- result$elements[[i]]
  cat(sprintf("Element %d:\n", i))
  cat(sprintf("  Type: %s\n", element$element_type))
  cat(sprintf("  Content: %s\n\n", substr(element$content, 1, 100)))
}

<?php
use Kreuzberg\ExtractionConfig;
use Kreuzberg\Kreuzberg;

// Configure element-based output
$config = new ExtractionConfig();
$config->setOutputFormat('element_based');

// Extract document
$result = Kreuzberg::extractFileSync('document.pdf', $config);

// Access elements
foreach ($result->getElements() as $element) {
    echo "Type: " . $element->getElementType() . "\n";
    echo "Text: " . substr($element->getText(), 0, 100) . "\n";

    if ($element->getMetadata()->getPageNumber()) {
        echo "Page: " . $element->getMetadata()->getPageNumber() . "\n";
    }

    if ($element->getMetadata()->getCoordinates()) {
        $coords = $element->getMetadata()->getCoordinates();
        echo sprintf("Coords: (%s, %s) - (%s, %s)\n",
            $coords->getLeft(), $coords->getTop(),
            $coords->getRight(), $coords->getBottom());
    }

    echo "---\n";
}

// Filter by element type
$titles = array_filter($result->getElements(), function($e) {
    return $e->getElementType() === 'title';
});

foreach ($titles as $title) {
    $level = $title->getMetadata()->getAdditional()['level'] ?? 'unknown';
    echo "[{$level}] {$title->getText()}\n";
}
?>

Elements are in result.elements. Each element has element_id, element_type, text, and metadata.

Element Types¶

Metadata¶

Every element's metadata contains:

PDF coordinates use bottom-left origin in points (1/72 inch).

Example Output¶

{
  "element_id": "elem-a3f2b1c4",
  "element_type": "title",
  "text": "Introduction to Machine Learning",
  "metadata": {
    "page_number": 1,
    "element_index": 0,
    "coordinates": { "x0": 72.0, "y0": 700.0, "x1": 540.0, "y1": 730.0 },
    "additional": { "level": "h1", "font_size": "24" }
  }
}

Filtering Elements¶

config = ExtractionConfig(result_format="element_based")
result = extract_file_sync("document.pdf", config=config)

titles = [e for e in result.elements if e.element_type == "title"]
tables = [e for e in result.elements if e.element_type == "table"]

for title in titles:
    level = title.metadata.additional.get("level", "h1")
    print(f"[{level}] {title.text}")

Migrating from Unstructured.io¶

If you're migrating from Unstructured.io, element-based output follows a similar structure with these key differences:

Document Structure¶

Represents a document as a flat list of nodes with explicit parent-child index references — a traversable tree with heading levels, content layers, inline annotations, and structured table grids.

Use document structure when you need hierarchical relationships between sections. For a flat list of semantic elements, use element-based output. For plain text, use the default unified output.

Comparison¶

Enable¶

from kreuzberg import extract_file_sync, ExtractionConfig

# Enable document structure output
config = ExtractionConfig(include_document_structure=True)

result = extract_file_sync("document.pdf", config=config)

# Access the document tree
if result.document:
    for node in result.document["nodes"]:
        node_type = node["content"]["node_type"]
        text = node["content"].get("text", "")
        print(f"[{node_type}] {text[:80]}")

import { extractFileSync, ExtractionConfig } from '@kreuzberg/node';

const config: ExtractionConfig = {
  includeDocumentStructure: true,
};

const result = extractFileSync("document.pdf", undefined, config);

if (result.document) {
  for (const node of result.document.nodes) {
    console.log(`[${node.content.nodeType}] ${node.content.text ?? ""}`);
  }
}

use kreuzberg::{extract_file_sync, ExtractionConfig};

let config = ExtractionConfig {
    include_document_structure: true,
    ..Default::default()
};

let result = extract_file_sync("document.pdf", None, &config)?;

if let Some(document) = &result.document {
    for node in &document.nodes {
        let text = node.content.text().unwrap_or("");
        println!("[{}] {}", node.content.node_type_str(), &text[..text.len().min(80)]);
    }
}

package main

import (
    "fmt"
    kreuzberg "github.com/kreuzberg-dev/kreuzberg/packages/go/v4"
)

func main() {
    config := kreuzberg.NewExtractionConfig(
        kreuzberg.WithIncludeDocumentStructure(true),
    )

    result, err := kreuzberg.ExtractFileSync("document.pdf", config)
    if err != nil {
        panic(err)
    }

    if result.Document != nil {
        for _, node := range result.Document.Nodes {
            fmt.Printf("[%s]\n", node.Content.NodeType)
        }
    }
}

import dev.kreuzberg.Kreuzberg;
import dev.kreuzberg.ExtractionConfig;
import dev.kreuzberg.ExtractionResult;

ExtractionConfig config = ExtractionConfig.builder()
    .includeDocumentStructure(true)
    .build();

ExtractionResult result = Kreuzberg.extractFileSync("document.pdf", config);

if (result.getDocumentStructure().isPresent()) {
    var document = result.getDocumentStructure().get();
    for (var node : document.nodes()) {
        System.out.println("[" + node.content().nodeType() + "]");
    }
}

using Kreuzberg;

var config = new ExtractionConfig
{
    IncludeDocumentStructure = true
};

var result = KreuzbergClient.ExtractFileSync("document.pdf", config);

if (result.Document is not null)
{
    foreach (var node in result.Document.Nodes)
    {
        Console.WriteLine($"[{node.Content.NodeType}]");
    }
}

require 'kreuzberg'

config = Kreuzberg::Config::Extraction.new(include_document_structure: true)

result = Kreuzberg.extract_file_sync('document.pdf', config: config)

if result.document
  result.document['nodes'].each do |node|
    node_type = node['content']['node_type']
    text = node['content']['text'] || ''
    puts "[#{node_type}] #{text[0...80]}"
  end
end

library(kreuzberg)

config <- extraction_config(
  include_document_structure = TRUE,
  output_format = "markdown"
)

file_path <- "document.pdf"
result <- extract_file_sync(file_path, config = config)

cat(sprintf("Total pages: %d\n", length(result$pages)))
cat(sprintf("MIME type: %s\n\n", result$mime_type))

for (i in seq_along(result$pages)) {
  page <- result$pages[[i]]
  cat(sprintf("Page %d structure:\n", i))
  cat(sprintf("  Content: %s\n", substr(page$content, 1, 100)))
  cat("\n")
}

Node Shape¶

Each node in result.document.nodes:

{
  "id": "node-a3f2b1c4",
  "content": { "node_type": "heading", "level": 2, "text": "Supervised Learning" },
  "parent": 0,
  "children": [4, 5, 6],
  "content_layer": "body",
  "page": 5,
  "page_end": null,
  "bbox": { "x0": 72.0, "y0": 600.0, "x1": 400.0, "y1": 620.0 },
  "annotations": []
}

• parent and children are integer indices into the nodes array (null if absent)
• bbox is present when bounding box data is available
• annotations contains inline formatting spans

Node Types¶

Content Layers¶

for node in result.document["nodes"]:
    if node["content_layer"] == "body":
        process_main_content(node)

Text Annotations¶

Paragraphs carry a list of annotations marking character spans:

{ "start": 0, "end": 16, "kind": { "annotation_type": "bold" } }

for node in result.document["nodes"]:
    for ann in node.get("annotations", []):
        text = node["content"].get("text", "")
        span = text[ann["start"]:ann["end"]]
        kind = ann["kind"]["annotation_type"]
        if kind == "link":
            print(f"Link: {span} -> {ann['kind']['url']}")
        else:
            print(f"{kind}: {span}")

Table Grid¶

Table nodes contain a grid with cell-level data:

{
  "rows": 3, "cols": 3,
  "cells": [
    { "content": "Algorithm", "row": 0, "col": 0, "row_span": 1, "col_span": 1, "is_header": true },
    { "content": "Decision Tree", "row": 1, "col": 0, "row_span": 1, "col_span": 1, "is_header": false }
  ]
}

Each cell has row, col, row_span, col_span, is_header, and optionally bbox.

for node in result.document["nodes"]:
    if node["content"]["node_type"] == "table":
        grid = node["content"]["grid"]
        rows, cols = grid["rows"], grid["cols"]
        table = [[None] * cols for _ in range(rows)]
        for cell in grid["cells"]:
            table[cell["row"]][cell["col"]] = cell["content"]
        for row in table:
            print(" | ".join(str(c or "") for c in row))

PDF Hierarchy Detection¶

Classifies text blocks in a PDF into heading levels (H1–H6) and body text based on font size analysis. Uses K-means clustering to group font sizes, then assigns heading levels by rank — largest cluster becomes H1, second-largest becomes H2, and so on.

Quick Start¶

from kreuzberg import extract_file_sync, ExtractionConfig, PdfConfig, HierarchyConfig

config: ExtractionConfig = ExtractionConfig(
    pdf_options=PdfConfig(
        extract_metadata=True,
        hierarchy=HierarchyConfig(
            enabled=True,
            k_clusters=6,
            include_bbox=True,
            ocr_coverage_threshold=0.8
        )
    )
)

result = extract_file_sync("document.pdf", config=config)

# Access hierarchy information
for page in result.pages or []:
    print(f"Page {page.page_number}:")
    print(f"  Content: {page.content[:100]}...")

import { extractFile } from '@kreuzberg/node';

const config = {
    pdfOptions: {
        extractMetadata: true,
        hierarchy: {
            enabled: true,
            kClusters: 6,
            includeBbox: true,
            ocrCoverageThreshold: 0.8,
        },
    },
};

const result = await extractFile('document.pdf', null, config);
if (result.pages) {
    result.pages.forEach((page) => {
        console.log(`Page ${page.pageNumber}:`);
        console.log(`  Content: ${page.content.substring(0, 100)}...`);
    });
}

use kreuzberg::{extract_file_sync, ExtractionConfig, PdfConfig, HierarchyConfig};

fn main() -> kreuzberg::Result<()> {
    let config = ExtractionConfig {
        pdf_options: Some(PdfConfig {
            hierarchy: Some(HierarchyConfig {
                enabled: true,
                detection_threshold: Some(0.75),
                ocr_coverage_threshold: Some(0.8),
                min_level: Some(1),
                max_level: Some(5),
            }),
            ..Default::default()
        }),
        ..Default::default()
    };

    let result = extract_file_sync("document.pdf", None::<&str>, &config)?;
    println!("Hierarchy levels: {}", result.hierarchy.len());
    Ok(())
}

package main

import "github.com/kreuzberg-dev/kreuzberg/packages/go/v4"

func main() {
    // Basic hierarchy configuration
    config := &kreuzberg.ExtractionConfig{
        PdfOptions: &kreuzberg.PdfConfig{
            ExtractImages: true,
            Hierarchy: &kreuzberg.HierarchyConfig{
                Enabled:               kreuzberg.BoolPtr(true),
                KClusters:             kreuzberg.IntPtr(6),
                IncludeBbox:           kreuzberg.BoolPtr(true),
                OcrCoverageThreshold:  kreuzberg.Float64Ptr(0.8),
            },
        },
    }

    // Advanced hierarchy configuration with more clusters
    advancedConfig := &kreuzberg.ExtractionConfig{
        PdfOptions: &kreuzberg.PdfConfig{
            ExtractImages: true,
            Hierarchy: &kreuzberg.HierarchyConfig{
                Enabled:               kreuzberg.BoolPtr(true),
                KClusters:             kreuzberg.IntPtr(12),
                IncludeBbox:           kreuzberg.BoolPtr(true),
                OcrCoverageThreshold:  kreuzberg.Float64Ptr(0.8),
            },
        },
    }

    _ = config
    _ = advancedConfig
}

import dev.kreuzberg.config.ExtractionConfig;
import dev.kreuzberg.config.PdfConfig;
import dev.kreuzberg.config.HierarchyConfig;

ExtractionConfig config = ExtractionConfig.builder()
    .pdfOptions(PdfConfig.builder()
        .hierarchyConfig(HierarchyConfig.builder()
            .enabled(true)
            .detectionThreshold(0.75)
            .ocrCoverageThreshold(0.8)
            .minLevel(1)
            .maxLevel(5)
            .build())
        .build())
    .build();

using Kreuzberg;

// Basic hierarchy configuration with properties
var config = new ExtractionConfig
{
    PdfOptions = new PdfConfig
    {
        ExtractImages = true,
        Hierarchy = new HierarchyConfig
        {
            Enabled = true,
            KClusters = 6,
            IncludeBbox = true,
            OcrCoverageThreshold = 0.8f
        }
    }
};

var result = await KreuzbergClient.ExtractFileAsync("document.pdf", config);
Console.WriteLine($"Content length: {result.Content.Length}");

// Advanced hierarchy detection with custom parameters
var advancedConfig = new ExtractionConfig
{
    PdfOptions = new PdfConfig
    {
        ExtractImages = true,
        Hierarchy = new HierarchyConfig
        {
            Enabled = true,
            KClusters = 12,           // More clusters for detailed hierarchy
            IncludeBbox = true,       // Include bounding box coordinates
            OcrCoverageThreshold = 0.7f  // Higher OCR threshold for stricter detection
        }
    }
};

var result = await KreuzbergClient.ExtractFileAsync("complex_document.pdf", advancedConfig);
Console.WriteLine($"Advanced hierarchy detection completed: {result.Content.Length} chars");

// Minimal configuration with only enabled flag
var minimalConfig = new ExtractionConfig
{
    PdfOptions = new PdfConfig
    {
        Hierarchy = new HierarchyConfig
        {
            Enabled = true;
            // Other properties use defaults:
            // KClusters = 6
            // IncludeBbox = true
        }
    }
};

var result = await KreuzbergClient.ExtractFileAsync("document.pdf", minimalConfig);
Console.WriteLine("Extraction with default hierarchy settings complete");

// Disabling hierarchy detection
var noHierarchyConfig = new ExtractionConfig
{
    PdfOptions = new PdfConfig
    {
        Hierarchy = new HierarchyConfig
        {
            Enabled = false
        }
    }
};

var result = await KreuzbergClient.ExtractFileAsync("document.pdf", noHierarchyConfig);
Console.WriteLine("Extraction without hierarchy detection complete");

require 'kreuzberg'

# Using keyword arguments with defaults
config = Kreuzberg::Config::Extraction.new(
  pdf_options: Kreuzberg::Config::PDF.new(
    extract_images: true,
    hierarchy: Kreuzberg::Config::Hierarchy.new(
      enabled: true,
      k_clusters: 6,
      include_bbox: true,
      ocr_coverage_threshold: 0.8
    )
  )
)

# Using hash syntax alternative
config = Kreuzberg::Config::Extraction.new(
  pdf_options: Kreuzberg::Config::PDF.new(
    extract_images: true,
    hierarchy: {
      enabled: true,
      k_clusters: 6,
      include_bbox: true,
      ocr_coverage_threshold: 0.8
    }
  )
)

Output¶

Hierarchy data is in result.pages[n].hierarchy. Each page has a blocks list:

{
  "block_count": 4,
  "blocks": [
    { "text": "Chapter 1: Introduction", "level": "h1", "font_size": 24.0, "bbox": [50.0, 100.0, 400.0, 125.0] },
    { "text": "Background", "level": "h2", "font_size": 18.0, "bbox": [50.0, 150.0, 300.0, 168.0] },
    { "text": "This chapter provides...", "level": "body", "font_size": 12.0, "bbox": [50.0, 200.0, 550.0, 450.0] }
  ]
}

• bbox: [left, top, right, bottom] in PDF points (present when include_bbox=True). This is the only way to obtain bounding box coordinates for text elements — they are not included by default.
• level: "h1" – "h6" or "body"

Configuration¶

Choosing k_clusters¶

Ocr_coverage_threshold¶

Requires an OCR backend to be configured separately.

Troubleshooting¶

• hierarchy is None — Check hierarchy.enabled is True. If the PDF is image-only, enable OCR. If fewer text blocks than k_clusters, reduce k_clusters.
• Most blocks classified as body — Document may use uniform font sizes. Reduce k_clusters (try 3–4).
• Heading levels don't match visual inspection — Levels are assigned by font size rank, not absolute size. Filter on block.font_size directly for absolute thresholds.

See the HierarchyConfig reference for the full parameter list.