Kreuzberg vs Unstructured: Feature Comparison¶

A comprehensive comparison of Kreuzberg and Unstructured.io for document intelligence workloads.

Executive Summary¶

Feature Matrix¶

Document Processing¶

Output Formats¶

Element Types¶

Supported File Formats¶

Kreuzberg (56+ formats): - Documents: PDF, DOCX, DOC, ODT, RTF, TXT, Markdown, RST, LaTeX, Typst - Presentations: PPTX, PPT, ODP, Keynote - Spreadsheets: XLSX, XLS, ODS, CSV - Web: HTML, XML, EPUB, FictionBook - Code: Jupyter Notebooks, Source code (via tree-sitter) - Data: JSON, YAML, TOML, BibTeX, OPML, OrgMode - Images: PNG, JPEG, TIFF, WebP (via OCR) - Email: EML, MSG

Unstructured (~30 formats): - Documents: PDF, DOCX, DOC, ODT, RTF, TXT - Presentations: PPTX, PPT - Spreadsheets: XLSX, XLS, CSV - Web: HTML, XML, EPUB - Data: JSON, Markdown - Images: PNG, JPEG, TIFF (via OCR) - Email: EML, MSG

Winner: Kreuzberg (broader format coverage)

Metadata Richness¶

Kreuzberg Metadata (format-specific discriminated unions):

{
  "title": "Document Title",
  "authors": ["Author 1", "Author 2"],
  "created_at": "2024-01-15T10:30:00Z",
  "modified_at": "2024-01-20T14:45:00Z",
  "language": "en",
  "format": {
    "format_type": "pdf",
    "page_count": 10,
    "version": "1.7",
    "is_encrypted": false,
    "permissions": {
      "print": true,
      "modify": false
    }
  }
}

Unstructured Metadata:

{
  "filename": "document.pdf",
  "page_number": 1,
  "filetype": "application/pdf"
}

Winner: Kreuzberg (richer, format-specific metadata)

Chunking & Embeddings¶

Language Bindings & Integrations¶

Kreuzberg: - ✅ Python (PyO3) - ✅ TypeScript (NAPI-RS) - ✅ Ruby (Magnus) - ✅ PHP (ext-php-rs) - ✅ Go (cgo FFI) - ✅ Java (JNI FFI) - ✅ C# (P/Invoke FFI) - ✅ Elixir (Rustler NIFs) - ✅ Rust (native) - ✅ WASM (browser/Node/Deno/Workers)

Unstructured: - ✅ Python (native) - ✅ REST API (language-agnostic) - ⚠️ Other languages via API only

Winner: Kreuzberg (native bindings for 10 languages)

Deployment Options¶

Kreuzberg: - ✅ CLI (single binary) - ✅ Self-hosted API (Docker, native) - ✅ Library (embedded in applications) - ✅ WASM (browser-based processing) - ❌ Managed cloud service

Unstructured: - ✅ Managed API (cloud-hosted) - ✅ Self-hosted API (Docker) - ✅ Python library - ❌ CLI - ❌ Browser-based

Cost Analysis¶

Kreuzberg: - License: Apache 2.0 (free, open-source) - Infrastructure: Self-hosted only (compute costs) - Total Cost: Infrastructure + maintenance

Unstructured: - License: Apache 2.0 (free, open-source) - Managed API: Free tier (100 pages/month) + paid plans ($0.01-0.10/page) - Self-hosted: Infrastructure costs only - Total Cost: API fees OR infrastructure + maintenance

Security & Compliance¶

Use Case Recommendations¶

Choose Kreuzberg If:¶

• ✅ You need maximum performance (Rust-based native speed)
• ✅ You're building a polyglot stack (Python, TS, Go, etc.)
• ✅ You require strict data privacy (on-prem processing)
• ✅ You need to process 56+ file formats
• ✅ You want zero API fees (fully self-hosted)
• ✅ You need native bindings for your language
• ✅ You're processing large document volumes (high throughput)
• ✅ You need offline embeddings (no external API calls)

Choose Unstructured If:¶

• ✅ You need ML-based layout detection (GPU-accelerated)
• ✅ You want a managed cloud service (zero ops)
• ✅ You need advanced chunking strategies (by_title, semantic)
• ✅ You're prototyping and want fast setup
• ✅ You need more granular element types (Address, Formula, etc.)
• ✅ You're already using OpenAI/Cohere APIs for embeddings
• ✅ You have low document volume (free tier sufficient)

Migration Path¶

From Unstructured to Kreuzberg: 1. Deploy Kreuzberg API (Docker or native) 2. Update endpoint URLs in your code 3. Add output_format=element_based for Unstructured-compatible output 4. Test with sample documents 5. Optimize with Kreuzberg-specific features (hierarchy, per-page, embeddings)

From Kreuzberg to Unstructured: 1. Sign up for Unstructured API key 2. Update endpoint URLs 3. Remove output_format parameter (element-based is default) 4. Adjust for different metadata structure

Roadmap & Future Features¶

Kreuzberg Planned Features:¶

• ⏳ Enhanced chunking strategies (by_title, semantic)
• ⏳ Layout detection models (optional GPU acceleration)
• ⏳ More element types (Header, Footer, Formula)
• ⏳ Cloud-hosted option (for non-self-hosters)

Unstructured Strengths:¶

• ✅ Mature ML models (layout, tables)
• ✅ Large community & integrations
• ✅ Managed service with SLA

Verdict¶

Kreuzberg excels at: - Performance (Rust native) - Polyglot support (10 language bindings) - Format coverage (56+ formats) - Self-hosted deployments - Cost efficiency (zero API fees)

Unstructured excels at: - ML-powered layout analysis - Managed cloud service - Advanced chunking strategies - Larger ecosystem

Recommendation: - High-volume, polyglot, self-hosted → Kreuzberg - Rapid prototyping, managed service → Unstructured - Hybrid approach: Use both (Kreuzberg for bulk processing, Unstructured for complex layouts)

Further Reading¶

• Migration Guide: Unstructured → Kreuzberg
• Kreuzberg Documentation
• Unstructured Documentation