demystify-extract

Python client SDK for dmstfy-extract — schema-first document extraction with generated Pydantic v2 models for the starter schema library (invoice, bank statement, resume, contract, email).

Not yet published to PyPI. Until then, use it via the monorepo (see the module README's "Use as your own repo" section) or an editable/path install.

pip install demystify-extract      # once published — httpx + pydantic v2

What you get

• Generated Pydantic models for each library schema version (InvoiceV1, …) that validate the API result shape, including the {raw, value} normalization envelope.
• Envelope accessors — field_value / field_raw / is_normalized_field.

The normalization envelope

Fields whose schema carries x-demystify-normalize: date|amount|name come back as {"raw": ..., "value": ...}. raw is the text exactly as printed; value is canonical (ISO 8601 date / integer minor units / cleaned name). Amount envelopes add currency (ISO 4217 or None) and exponent (minor-unit scale — INR/USD 2, JPY 0, BHD 3), so ¥1234 → {"raw": "¥1234", "value": 1234, "currency": "JPY", "exponent": 0}.

from demystify_extract.generated import InvoiceV1
from demystify_extract import field_value

invoice = InvoiceV1.model_validate(result["data"])
field_value(invoice.total_amount)  # integer minor units

Results carry a top-level result_format: "normalized@1" version tag.

See the module README for the full API, offline quickstart, and architecture.

License

MIT