Migrate from Unstructured

Why Migrate

Unstructured converts Office documents to PDF before parsing, destroying structural data — track changes, merged cells, comments, headers, footers, text boxes — that exists in the source XML. AILANG Parse reads the XML directly. Deterministic, structural, fast.

On the OfficeDocBench structural benchmark, AILANG Parse scores 93.9% with 100% format coverage, compared to Unstructured's 62.1% composite (38.7% coverage-adjusted — Unstructured only handles 62% of the benchmark files).

Quick Migration

AILANG Parse is a true drop-in replacement. Change the URL — your existing code works unchanged:

curl — One URL change

# Before: Unstructured
curl -X POST "https://api.unstructured.io/general/v0/general" \
  -H "unstructured-api-key: $UNSTRUCTURED_KEY" \
  -F "files=@report.docx"

# After: AILANG Parse (same endpoint, same field name, same header)
curl -X POST "https://docparse.ailang.sunholo.com/general/v0/general" \
  -H "unstructured-api-key: $DOCPARSE_API_KEY" \
  -F "files=@report.docx"

Python

# Before: Unstructured
resp = requests.post(
    "https://api.unstructured.io/general/v0/general",
    headers={"unstructured-api-key": UNSTRUCTURED_KEY},
    files={"files": open("report.docx", "rb")}
)

# After: AILANG Parse (change URL and key — everything else stays the same)
resp = requests.post(
    "https://docparse.ailang.sunholo.com/general/v0/general",
    headers={"unstructured-api-key": DOCPARSE_API_KEY},
    files={"files": open("report.docx", "rb")}
)

Routing Pattern

For mixed-format pipelines, the recommended approach is to route Office formats through AILANG Parse and PDFs through Unstructured. A simple file-extension check is all you need:

import requests
from pathlib import Path

AILANG_PARSE_URL = "https://docparse.ailang.sunholo.com"
UNSTRUCTURED_URL = "https://api.unstructured.io"

OFFICE_EXTENSIONS = {
    ".docx", ".pptx", ".xlsx",
    ".odt", ".odp", ".ods",
    ".csv", ".html", ".md", ".epub"
}

def parse_document(filepath: str) -> dict:
    """Route documents to the best parser for their format."""
    ext = Path(filepath).suffix.lower()

    if ext in OFFICE_EXTENSIONS:
        # Deterministic structural parsing — no AI needed, no per-page billing
        # Same endpoint format as Unstructured — just change the URL
        resp = requests.post(
            f"{AILANG_PARSE_URL}/general/v0/general",
            headers={"unstructured-api-key": DOCPARSE_API_KEY},
            files={"files": open(filepath, "rb")}
        )
    else:
        # PDFs, scanned images — Unstructured ML pipeline
        resp = requests.post(
            f"{UNSTRUCTURED_URL}/general/v0/general",
            headers={"unstructured-api-key": UNSTRUCTURED_KEY},
            files={"files": open(filepath, "rb")}
        )

    return resp.json()

This gives you the best of both worlds: structural fidelity for Office documents and ML-powered extraction for PDFs and scans.

What You Gain

Self-Hosting

Run AILANG Parse locally with Docker — no external dependencies for Office formats:

# Clone and build
git clone https://github.com/sunholo-data/ailang-parse.git
cd ailang-parse
docker build -t docparse .

# Parse a document (mount your files into /data)
docker run -v $(pwd):/data docparse /data/report.docx

# Parse with AI (pass your API key)
docker run -e GOOGLE_API_KEY="your-key" \
  -v $(pwd):/data docparse --ai gemini-2.5-flash /data/document.pdf

The Dockerfile builds AILANG from source and includes all 31 parser modules. Office parsing works immediately with zero configuration. Add a GOOGLE_API_KEY environment variable to enable AI-powered PDF and image parsing.

For detailed configuration options, capability budgets, and AI model setup, see the Self-Host Guide.