# PurifySec (Purify Solution Co., Ltd.)

> PurifySec is the AI division and product of Purify Solution Co., Ltd. (주식회사 퓨리파이솔루션), a Korean corporation incorporated on 2026-03-13 whose existing business is water purification solutions. PurifySec is a defensive, human-in-the-loop security triage assistant for SOC teams, built on Anthropic's Claude API. Stage: working prototype. No customers yet, no outside funding, not affiliated with Anthropic.

## Company facts
- Legal name (English): Purify Solution Co., Ltd.
- Legal name (Korean): 주식회사 퓨리파이솔루션
- Type / country: corporation (주식회사), Republic of Korea
- Incorporated: 2026-03-13
- Business Registration No. (사업자등록번호): 323-86-03748
- CEO (대표자): Sangheon Lee (이상헌)
- Head of AI (leads PurifySec): Jisub Lee (이지섭), JisubLee@purifysec.com, https://www.linkedin.com/in/jisub-lee-188634442
- Address: Building A, 154 Pokpo-ro, Hwado-eup, Namyangju-si, Gyeonggi-do, Republic of Korea
- Address (Korean): 경기도 남양주시 화도읍 폭포로 154, 가동
- Existing business: water purification solutions
- AI division: PurifySec (website https://purifysec.com, email domain @purifysec.com)
- General contact: hello@purifysec.com

## Product (built on the Claude API)
- What it is: a security triage assistant for SOC teams. Defensive use only, on systems customers own or are authorized to manage.
- How Claude is used (design; the working prototype runs on the Claude API, some parts are still being built): the application calls Claude through Anthropic's Messages API for each alert. Claude uses read-only tools (for example search_events, get_host_context, get_user_context) to query SIEM/EDR data, and returns a schema-validated JSON triage note (suggested verdict and priority, evidence references, MITRE ATT&CK candidates, next steps). It is also designed to draft incident summaries and code-remediation reports.
- Human in the loop: an analyst approves, edits or rejects every output; no autonomous actions on customer systems.
- Design elements (some still being built): prompt caching for long log context, routing between a smaller and a larger Claude model, an offline evaluation harness, prompt-injection defenses that treat log text as untrusted data, audit logging.
- Status: working prototype. Planned next step: early-access pilots with a small number of SOC teams (not started).

## Pages
- [Home](https://purifysec.com/): product summary and company details
- [Technology](https://purifysec.com/technology): architecture and Claude API design
- [Product](https://purifysec.com/product): features
- [About](https://purifysec.com/about): company information and leadership
- [Responsible use](https://purifysec.com/responsible-use): defensive-use policy
- [Contact](https://purifysec.com/contact)