CUSTOM_TERM, are applied on
/api/v1/mask, /api/v1/redact, and the file-redaction upload, and
always win when they overlap a built-in detection.glossary array of term strings. These apply to this request only;
nothing is stored.1234{ "text": "Project Falcon ships Q3. Contact alias Bluebird.", "glossary": ["Project Falcon", "Bluebird"] }
glossary_id. An explicit glossary_id is
always honoured, even if that glossary is disabled, so you can keep a
glossary on standby and opt into it per request.match_mode that controls how the term is found
in your text:| Mode | Matches | Example term | Matches in text |
word_boundary (default) | The term as a whole word, including inflected forms | patient | patient, patients, patient's |
exact | The term verbatim | POL-987 | POL-987 only |
substring | The term anywhere, even inside another word | acme | acme, acmecorp, myacme |
word_boundary uses the English lemma and an Arabic light-stemmer, so
inflected and pluralised forms of a term still match in both languages.case_sensitive: false); set
case_sensitive: true on an entry when case carries meaning (e.g. a
ticker or code). An entry can be pinned to a locale (ar / en) or left
language-agnostic.glossary (a comma-separated string) and
glossary_id. Your terms are masked in the document just as they are in
text requests.