Kenfold 0.1.2

Changes

• Fixed credential redaction at length limits in MCP search inputs and deletion reasons: complete inputs are now redacted before truncation.
• Added dashboard search-query redaction before queries reach embedding and reranking endpoints.
• Redacted deletion and rejection audit reasons across MCP, dashboard, and CLI paths, including a storage-layer safeguard.
• Added 33 end-to-end regression cases covering outgoing model requests, stored audit reasons, Unicode boundaries, ordinary inputs, and placeholders.

Upgrade

git pull && make up

If you use local models, replace make up with make up-embed or make up-extract, whichever you used before.

Use Go 1.27.2 or newer to build the CLI. Race tests, database integration tests and the vulnerability scan passed with Go 1.27.2. No database schema or dependency changes are required.

Known limitations

• Secret detection remains pattern-based. These fixes protect new requests and audit writes; existing records are not rewritten.
• Dashboard queries can still appear in URLs and browser history; this release protects the query payloads sent to model endpoints.
• Raw-LoCoMo results cover 1,402 of 1,540 questions. Extracted-LoCoMo and LongMemEval QA have not been scored.
• Connections from real ChatGPT and claude.ai clients have not been tested.
• Run one server replica.

변경 사항

• MCP 검색 입력과 삭제 사유를 자르기 전에 비밀정보를 가리도록 수정해 길이 제한 경계에서 인증정보 일부가 남는 문제 해결
• 대시보드 검색어를 임베딩·재순위화 API에 보내기 전에 비밀정보를 가리도록 수정
• MCP·대시보드·CLI의 삭제·거절 사유에 포함된 비밀정보가 감사 기록에 저장되지 않도록 저장 계층에서도 보호
• 실제 외부 요청 본문과 감사 기록을 검사하는 회귀 테스트 33개 추가

기존 저장 데이터는 자동으로 정리하지 않습니다. 대시보드 검색어의 URL·브라우저 기록 보존 방식도 변경하지 않습니다. DB 스키마와 의존성 변경은 없습니다.

License: MIT.

Kenfold 0.1.1

Changes

• Added a Korean README, deployment guide and MCP tool reference.
• Fixed KENFOLD_UPDATE_CHECK not reaching the Compose container.
• Updated Git and Docker ignore rules for local environment files and agent working files.
• Removed internal work notes from the benchmark report and corrected the dashboard architecture docs.

Upgrade

git pull && make up

If you use local models, replace make up with make up-embed or make up-extract, whichever you used before.

Use Go 1.27.2 or newer to build the CLI. Race tests, database integration tests and the vulnerability scan passed with Go 1.27.2.

Known limitations

• Raw-LoCoMo results cover 1,402 of 1,540 questions. Extracted-LoCoMo and LongMemEval QA have not been scored.
• Connections from real ChatGPT and claude.ai clients have not been tested.
• Run one server replica.

변경 사항

• 한국어 README, 배포 안내, MCP 명세 추가
• Compose의 KENFOLD_UPDATE_CHECK=false 미적용 문제 수정
• 로컬 환경 파일과 에이전트 작업 파일을 Git과 Docker 빌드에서 제외
• 벤치마크 문서의 내부 작업 기록 정리, 대시보드 설계 설명 수정

License: MIT.

Kenfold 0.1.0

First release for agents on the same machine.

Clone the repository and run make up. A checkout of this tag builds as v0.1.0. The dashboard footer, kenfold version, and the MCP initialize handshake show that version.

A release build checks GitHub for a newer release and reports it. It does not download or restart. Upgrade with git pull && make up. KENFOLD_UPDATE_CHECK=false turns the check off.

This release does not include a verified ChatGPT or claude.ai connection, complete LoCoMo or LongMemEval scores, or support for more than one server replica.

License: MIT.