jev extract

Pull values out of text without letting the model invent them. Regexes find every candidate span in code, Jev picks the one that matches the field's meaning, code normalizes the verbatim value.

jev extract [text] --want <field> [--want <field>...] [--context <text>] [options]

When to use

Invoices, emails, forms, logs: anything where you need specific values and cannot afford a hallucinated one. If no regex matches, the field is none and no API call is made.

Fields

Form	Example	Meaning
builtin	--want email,phone,date	email, phone, url, amount, date, percent, number
custom regex	--want invoice=/INV-\d+/	Your pattern, named
custom regex with meaning	--want po=/PO\s?\d{6}/:the purchase order number	The description drives the question
aliased builtin	--want "sender=email:the sender's address" (quote descriptions with spaces)	Same pattern, different meaning, so two emails can be told apart

Options

Option	Meaning	Default
[text]	Text, @file, or -	stdin
-w, --want <field>	A field, repeatable; comma-separate builtins	required
-c, --context <text>	What the document is, e.g. "supplier invoice"
--min-confidence <p>	Below this confidence the value is review	0.6
--fail-on <list>	Exit 2 when any field is review or missing	none

Output

Per field: value (verbatim), normalized, confidence, action (auto, review, none), candidates (how many spans matched).

Builtin	Normalized form
email	lowercased
phone	digits only, leading + kept
url	trailing punctuation stripped
amount	{"value": 1250, "currency": "USD"}
date	YYYY-MM-DD
percent, number	number

Example

jev extract @invoice.txt --want amount,date --want invoice=/INV-\d+/ --context "supplier invoice"

Field    Value          Normalized                        Conf  Action  Cands
-------  -------------  --------------------------------  ----  ------  -----
amount   \$1,250.00      {"value":1250,"currency":"USD"}   0.97  auto    3
date     Sept 30, 2026  2026-09-30                        0.94  auto    2
invoice  INV-2231                                         1.00  auto    1

Patterns: pre-parsed value extraction, date extraction.