terbium-parse 2.0.0

Business documents in. Clean rows out.

Documentation · Quickstart · Changelog · Releases · Trove

terbium parses business documents (vendor catalogues first, plus invoices, receipts, resumes, price lists and decks) into clean rows. It rebuilds structure from the position of every word, scores its own confidence, and only calls an AI model when the algorithm cannot be sure.

Sponsors

Primary sponsor

NodeMaven: the most efficient proxy provider for web scraping and automation, with the highest quality IPs on the market.

Why NodeMaven?

• ZIP targeting
• 99.9% uptime
• IP filtering: every proxy has a fraud score under 97%
• No KYC required
• Free tools: Proxy Bandwidth Checker, Meta Tag Checker, IP Lookup and more

Codes for terbium users: TERBIUM35 for 35% off mobile and residential proxies, TERBIUM40 for 40% off ISP (static) proxies.

Install

pip install terbium-parse

The PyPI name is terbium-parse; you import terbium, and the command is terbium. (The PyPI project named terbium is unrelated.) terbium --version prints the installed version.

Optional AI lanes: pip install "terbium-parse[anthropic]" (or openai, kimi, grok, gemini, or ai for all). OCR of images and image-only pages needs a local tesseract binary. Details: Installation.

Example

Point it at a vendor catalogue and get a table of products, each with its name, SKU, materials or ingredients, dimensions and image. Then a CSV.

import terbium

rows = terbium.build_catalog("vendor_catalogue.pdf", images_dir="images/")
terbium.to_catalog_csv(rows, "catalogue.csv")
# rows: {"sku": "RG-1001", "name": "Anatolia Kilim", "materials": "wool",
#        "dimensions": None, "image": "Anatolia_Kilim.jpeg", "page": 12}

The same engine reads other documents:

doc = terbium.parse("invoice.pdf")                     # line items, totals, vendor
cv = terbium.parse("maria_cv.pdf", doc_type="resume")  # sections, experience, skills
print(doc.stats)   # Stats(total=..., confident=..., ambiguous=...)

terbium catalogue.pdf --csv out.csv                    # product table + photos, no AI
terbium invoice.pdf --csv rows.csv --html report.html  # CSV + self-contained report

Features

• Structure from geometry. Running headers are stripped, two-page spreads split, and columns, rows and matrices rebuilt from word positions. The table detector is content-agnostic.
• Catalog table. Each product photo anchors a row; the name, SKU, materials and dimensions are mined from nearby text, and the photo is extracted and named after the product.
• Many document types. Automatic classification into catalogue, lookbook, transaction, resume, table or deck, with --type to override.
• Formats. PDF, PPTX, XLSX, CSV, and PNG, JPG, WEBP or TIFF images through local OCR.
• Confidence on every record. Each record carries a 0.05 to 1.0 score and the reasons behind it.
• AI only where needed. Hard tables go to a routed model tier (Haiku, Sonnet or Opus class) across Claude (default), GPT, Kimi, Grok and Gemini. With no key, terbium prints which pages need help and which tier it recommends.
• Feeds. Normalized colour, material and dimension fields, plus Shopify CSV and PIM JSON exporters.

Documentation

• Overview
• Installation and Quickstart
• The catalog table and Pulling out images
• Invoices, receipts and resumes
• File formats and Schemas
• Confidence and escalation
• AI lane and routing
• Normalization and feeds
• CLI reference and Python API reference
• Document taxonomy
• Evaluation and Changelog and status

Developer notes stay in this repository: docs/DESIGN.md, docs/ROADMAP.md and docs/TAXONOMY.md.

Contributing

Issues and pull requests are welcome at https://github.com/velofy/terbium. To work on the code:

git clone https://github.com/velofy/terbium.git
cd terbium
pip install -e ".[dev]"
python -m pytest -q
python eval/category_bench.py

License

MIT. Built by anishfyi. See LICENSE.