Structured ExtractionService

Rule-based extraction of named fields from text (not a language model: it returns what the text states explicitly). POST JSON {"text": "...", "schema": ...} to /v1/extract. schema can be a JSON Schema ({"properties": {"total": {"type": "number"}}}), a list of field names (["email", "invoice_number"]) or {"fields": {"name": "<regex>"}}. Built-in fields: email, phone, date, url, money (also total, price, amount, tax), time, company, address, percent, zip, iban, ip, name; any other field name is filled from a "Label: value" line in the text. The response is a JSON object with fields (null when not found; typed per the JSON Schema), matched_by (how each field was found) and schema_mode. Identical input gives identical output on every supplier.