Early Access

Drop-in privacy layer for Korean text pipelines

AegisText detects and redacts Korean PII — names in casual speech, non-standard addresses, resident registration numbers — before your text reaches an LLM, search index, or storage layer. One endpoint. No pipeline changes.

View Quick Startopen_in_new

Generic DLP tools weren't built for Korean text

When a support log says "수진아 내일 봐", generic DLP tools don't recognize "수진" as a person's name. They might flag "이형" from "재훈이형한테" as a separate person. They sometimes miss resident registration numbers entirely. Korean names blend with particles and honorifics. Addresses mix road names with colloquial directions. RRNs and bank accounts follow Korea-specific formats. Rule-only tools miss context. Generic models lack precision.
chat_bubble한국어 인명은 "수진아", "민수한테", "재훈이형" 형태로 쓰입니다
wrong_locationGeneric DLP misses Korean RRNs or flags common nouns as names
integration_instructionsYou don't want to build a detection engine from scratch

Same input, different results

We ran the same Korean text through a generic DLP service and AegisText. Here's what each one caught.

Case 1 / 5

Input

정하늘의 연락처는 010 2222 3333입니다

정하늘 (Name)
Generic❌
Aegis✅
010 2222 3333 (Phone)
Generic❌
Aegis✅

"하늘" overlaps with a common noun. The phone number uses spaces instead of dashes.

Generic DLP
2 / 10
AegisText
10 / 10

Tested with a widely-used generic DLP API service, March 2026. All tests used the same input text and equivalent entity type configurations.

Fits into the pipeline you already have

1
description

Send text

Send support logs, notes, or documents to AegisText

2
security

Get redacted text

Sensitive information comes back replaced with safe placeholders

3
smart_toy

Use downstream

Pass the clean text to your LLM, search index, classifier, or storage. Restore original values when needed

AegisText doesn't replace your pipeline. It's a layer that fits in front of it.

See how Korean PII gets detected and redacted

Paste a customer message, support note, or Korean text sample to preview detected entities and sanitized output.

editTry editing
홍길동의 연락처는 010-1234-5678이고 이메일은 hong@example.com 입니다. 주소는 서울시 강남구 역삼동 123-45이고 주민번호는 900101-1234567입니다.
홍길동의 연락처는 010-1234-5678이고 이메일은 hong@example.com 입니다. 주소는 서울시 강남구 역삼동 123-45이고 주민번호는 900101-1234567입니다.

Sample output only. Hosted API is the fastest way to start. If you need private deployment, tell us in the access form.

Where teams start

headset_mic
Most commonsupport_agent

Support log summarization

Redact customer names, contacts, and account numbers from tickets and chat logs before sending them to an LLM for summarization or classification

neurology
dataset

RAG preprocessing

Remove PII from internal docs and tickets before indexing into your vector store. No personal data in your search chunks

description
edit_note

Internal text sanitization

Clean operations logs, meeting notes, and reports before sharing across teams or sending to external systems

Supported entities

AegisText combines multiple detection modules in a hybrid architecture to achieve higher precision than any single approach alone.

personPerson Namehome_pinAddresscallPhone NumberfingerprintResident Registration NumbermailEmailcredit_cardCredit Card Numberaccount_balanceBank Account

Need a different entity type? Let us know in the access request form.

Frequently asked questions

Is this only for support logs?expand_more
No. AegisText works with any text pipeline — RAG preprocessing, internal memos, operations logs, meeting notes, and more.
Do you store the raw text?expand_more
No. AegisText does not store raw text on its servers. Text is processed and the result is returned immediately. This is a core design principle — we believe a privacy layer should not become another place where sensitive data accumulates.
How is this deployed?expand_more
The Hosted API will be available first as the fastest way to get started. Additional deployment options will be rolled out progressively after that.
How is this different from generic DLP tools?expand_more
AegisText is purpose-built for Korean text. It catches casual-form names ("수진아", "민수한테"), non-standard addresses, and Korea-specific identifiers like RRNs and bank accounts that generic DLP tools often miss or misclassify.
Does AegisText only support Korean?expand_more
Currently, AegisText is optimized for Korean text. We plan to expand to additional languages in the future.
What entities are currently supported?expand_more
Person name, phone number, address, resident registration number, email, credit card number, and Korean bank account — 7 entity types.

Start with the Hosted API — tell us what you're building

We're working with early teams using LLMs on customer or internal text. Share your role, current blocker, and timeline, and we'll follow up with the most relevant starting path.