Redact

Filter PII on the device.

Multilingual on-device PII detection and redaction.

• SDKs, install and examples: https://github.com/Desert-Ant-Labs/desert-ant-core/blob/main/docs/models/redact.md
• Website: https://desertant.com/models/redact/

Strip names, addresses, emails, phone numbers, cards, IBANs and national IDs out of text before it leaves the device, across 27 languages (see Languages). A token classifier plus a deterministic layer for structured IDs. The deployable model is 11.6MB (4-bit Core ML on Apple) or 24.5MB (int8 LiteRT .tflite on Android, Linux and the web).

"Call Anna Kovács at anna@example.hu, IBAN GB29NWBK60161331926819" → "Call [GIVEN_NAME] [SURNAME] at [EMAIL], IBAN [BANK_ACCOUNT]"

Try it

• Live demo: desert-ant-labs/redact-demo: paste text and watch PII get highlighted or masked, fully in your browser.

Platforms	iOS, macOS, tvOS, visionOS, Android, Linux, Windows, Browser, Node
Languages	27
Weights	v0.4.0

Install

Swift (requirements)

.package(url: "https://github.com/Desert-Ant-Labs/desert-ant-core.git", from: "3.1.0")

Then add the Redact product to your target.

Kotlin (requirements)

implementation("ai.desertant:redact:3.1.0")

JavaScript (requirements)

npm i @desert-ant-labs/redact @litertjs/core   # browser
npm i @desert-ant-labs/redact                  # Node, prebuilt native core

Taxonomy (20 public labels, plus ORG)

GIVEN_NAME, SURNAME, STREET_NAME, BUILDING_NUMBER, SECONDARY_ADDRESS, CITY, STATE, ZIP_CODE, EMAIL, PHONE, CREDIT_CARD, BANK_ACCOUNT, ROUTING_NUMBER, IP_ADDRESS, URL, GOVERNMENT_ID, PASSPORT, DRIVERS_LICENSE, TAX_ID, SSN.

ORG (organisation / company name) is detected but not redacted by default: a company is not a natural person. It exists so that Silverfin, Odoo or Visma Nova are recognised as organisations instead of being mislabelled as a SURNAME. Opt in by passing it explicitly in the SDK's labels option.

The deterministic layer additionally emits IMEI (device identifier), a deterministic-only label outside the neural head.

How it compares

Every system below was scored by the same harness on the same rows, each at its own operating point, so the comparison measures the models rather than the plumbing.

System	Recall	Precision	Size	Params
redact	88.8	99.6	11.6MB	23M
GLiNER-PII	91.1	90.4	2.3GB	570M
Rampart	61.4	97.2	14.7MB	18.5M
OpenAI privacy filter	60.2	93.5	3GB	1.5B

Recall is the share of personal data fully masked (leak-safe), macro-averaged over WikiANN, MultiNERD and a format-valid structured-PII set across 24 EU languages. Precision is the share of masked spans that were really personal data, on the structured set. Size is the Apple build; the Android and web build is 24.5MB.

Not masking ordinary words matters as much as catching real ones, because a false positive corrupts the text a downstream model receives. On an 11,528-row negative set across 27 languages, built to provoke exactly that (sentence-initial capitals, ALL-CAPS input, month and weekday names, UI vocabulary, bare numbers, company names), 94.1% of rows come back untouched.

AWS Comprehend, English only

Comprehend is the other service teams weigh, and it is not in the table above because its PII API only accepts English, every other language code is refused outright, so there is no way to run it on the other 23. Scored on the same English rows:

System	Names (WikiANN)	Names (MultiNERD)	Structured	English composite
redact	69.5	94.9	95.0	86.5
AWS Comprehend	84.3	98.5	91.9	91.6

Leak-safe recall; precision is the same for both (99.8 against 100.0). On English names Comprehend is ahead of us. It also runs in the cloud, bills per call, and covers one of the 27 languages listed below.

Languages

27 languages: every official EU language, plus 3 more. Latin, Greek and Cyrillic scripts.

The 24 EU languages

Code	Language
bg	Bulgarian
hr	Croatian
cs	Czech
da	Danish
nl	Dutch
en	English
et	Estonian
fi	Finnish
fr	French
de	German
el	Greek
hu	Hungarian
ga	Irish
it	Italian
lv	Latvian
lt	Lithuanian
mt	Maltese
pl	Polish
pt	Portuguese
ro	Romanian
sk	Slovak
sl	Slovenian
es	Spanish
sv	Swedish

Beyond the EU

Code	Language
nb	Norwegian Bokmål
nn	Norwegian Nynorsk
is	Icelandic

Coverage is not uniform: the largest EU languages are the strongest, and Maltese and Irish are the weakest of the 24. The per-language detection numbers are in the benchmark data.

Runtime settings

Recommended defaults: min_score = 0.6, max_length = 256, stride = 64.

License

Desert Ant Labs Source-Available License. Free for most apps, and a commercial license is required at scale. Full terms are at the link. Licensing: licensing@desertant.com.

See THIRD_PARTY_NOTICES.md.

Citation

@software{redact_2026,
  title  = {Redact: Multilingual on-device PII detection and redaction},
  author = {Desert Ant Labs},
  year   = {2026},
  url    = {https://huggingface.co/desert-ant-labs/redact},
}

---

© 2026 Desert Ant Labs · https://desertant.com