Selvedge
Developed by masondelan and released on July 10, 2026, Selvedge is an open-source, local decision-tracking tool that records the underlying reasoning process and evidence when an AI coding agent modifies code. This tool operates similarly to a digital black box flight recorder, intervening in real-time at the moment the agent changes the source code to capture the AI's thought process and preserve it in the local environment. Unlike traditional version control systems that only record the final modified state of the source code, Selvedge records why the AI modified a specific function and which libraries it used.
Selvedge, released on July 10, 2026, by developer masondelan, is an open-source, local decision-tracking tool that records the underlying reasoning and evidence when an AI coding agent modifies code. This tool functions similarly to a digital black box flight recorder, intervening in real-time when the agent changes source code to capture the AI's thought process and preserve it in a local environment. Unlike traditional version control systems that only record the final modified state of the source code, Selvedge records the logical reasons and causal relationships behind why the AI modified a specific function or added a library, helping developers clearly track this information later.
Existing collaboration and version control tools, such as git blame, only record the physical change history and author of the code, failing to explain the actual reasons or deeper intentions behind the code changes. In particular, when pair programming with an AI agent, a common problem is the expiration of the context window, which leads to the chronic limitation of losing the agent's long-term memory of why it designed the implementation in a certain way. To bridge this gap, Selvedge provides a local repository for recording the thoughts of the AI agent, offering a powerful solution that not only preserves the framework of the code written by the agent but also the decision-making process behind it.
This tool collects metadata about the entire coding activity, as if the AI agent were generating text, and stores it discreetly and quickly in a lightweight SQLite database file in a local directory within the project. It is a completely offline structure that does not transmit any personal information or source code to external servers, ensuring strong data security and confidentiality. Developers can easily integrate it with various intelligent agents that support the command-line interface (CLI) and Model Context Protocol, allowing them to clearly visualize and maintain the entire lifecycle of research source code that requires complex design rationale, such as genomic data processing pipelines or automated protein structure prediction scripts.
💻 System Requirements
"Not required (CPU-based ultra-lightweight SQLite and CLI operation)"
"Package installation space within 50MB, history database size at the level of tens of MB"
⚡ Installation
4-1. Quick Start
pip install selvedge && cd your-project && selvedge setup
4-2. Detailed installation
1. Installing Python packages via PyPIpip install selvedge
2. Navigate to the target project directory and run the auto-setup wizard.cd your-project selvedge setup
3. Verify local database initialization and normal operation of Git hooks integrationselvedge init
4. Configuration and Integration Status Diagnostic Verificationselvedge doctor
FAQ
What is Selvedge?
Selvedge, released on July 10, 2026, by developer masondelan, is an open-source, local decision-tracking tool that records the underlying reasoning and evidence when an AI coding agent modifies code. This tool functions similarly to a digital black box flight recorder, intervening in real-time when the agent changes source code to capture the AI's thought process and preserve it in a local environment. Unlike traditional version control systems that only record the final modified state of the source code, Selvedge records the logical reasons and causal relationships behind why the AI modified a specific function or added a library, helping developers clearly track this information later. Existing collaboration and version control tools, such as git blame, only record the physical change history and author of the code, failing to explain the actual reasons or deeper intentions behind the code changes. In particular, when pair programming with an AI agent, a common problem is the expiration of the context window, which leads to the chronic limitation of losing the agent's long-term memory of why it designed the implementation in a certain way. To bridge this gap, Selvedge provides a local repository for recording the thoughts of the AI agent, offering a powerful solution that not only preserves the framework of the code written by the agent but also the decision-making process behind it. This tool collects metadata about the entire coding activity, as if the AI agent were generating text, and stores it discreetly and quickly in a lightweight SQLite database file in a local directory within the project. It is a completely offline structure that does not transmit any personal information or source code to external servers, ensuring strong data security and confidentiality. Developers can easily integrate it with various intelligent agents that support the command-line interface (CLI) and Model Context Protocol, allowing them to clearly visualize and maintain the entire lifecycle of research source code that requires complex design rationale, such as genomic data processing pipelines or automated protein structure prediction scripts.
When should I use Selvedge?
Developed by masondelan and released on July 10, 2026, Selvedge is an open-source, local decision-tracking tool that records the underlying reasoning process and evidence when an AI coding agent modifies code. This tool operates similarly to a digital black box flight recorder, intervening in real-time at the moment the agent changes the source code to capture the AI's thought process and preserve it in the local environment. Unlike traditional version control systems that only record the final modified state of the source code, Selvedge records why the AI modified a specific function and which libraries it used.
📝 Update Notes
- vv0.3.1710/19/2026
Selvedge v0.3.17에서는 MCP 서버를 통해 데이터 저장 및 복구 과정을 테스트할 수 있는 데모 기능이 추가되어, AI 에이전트와의 데이터 연동 실험이 더욱 용이해졌습니다. 특히 실험 결정 사항을 Markdown으로 내보낼 때 기존 기록과 일치하는지 확인하는
--check기능이 도입되어, 실험 로그의 무결성을 더욱 손쉽게 검증할 수 있습니다. 또한, 실험 거절(rejection)에 대한 라벨이 명확해지고 CLI 출력 오류가 수정되어, 복잡한 실험 이력을 기록할 때 발생하던 혼선을 줄였습니다. 이번 업데이트를 통해 연구원들은 실험 데이터의 변경 사항을 더욱 정확하고 체계적으로 관리할 수 있습니다. - vv0.3.1610/2/2026
이번 업데이트에서는 실험 자동화 및 데이터 분석 워크플로우의 안정성을 높여주는 핵심 기능들이 추가되었어요.
selvedge doctor기능을 통해 복잡한 에이전트 환경 설정을 손쉽게 점검하고 오류를 사전에 방지할 수 있습니다. 특히 새로운ledger기능은 모든 작업 내역과 변경 사항을 투명하게 기록하여, 연구 데이터의 추적성과 재현성을 확보하는 데 매우 유용해요. 또한, 실패 사례에 대한 정밀한 스코어링 도입으로 실험 파이프라인의 신뢰도를 한층 더 높일 수 있습니다. - vv0.3.159/27/2026
Selvedge v0.3.15에서는 AI의 의사결정 기억(decision-memory)을 더욱 정밀하게 평가하고 검증할 수 있는 기능이 강화되었습니다. 특히 실험의 재현성이 핵심인 생명공학 연구에서, AI의 판단 근거를 일관되게 관리하고 통제된 결과를 도출하는 데 큰 도움이 됩니다. 또한 다양한 코딩 클라이언트를 위한 어댑터가 추가되어, 연구용 AI 파이프라인을 구축하고 관리하는 과정이 한층 더 편리해졌습니다.
- vv0.3.149/13/2026
Selvedge v0.3.14에서는 MCP
prior_attempts기능의 시간 범위 설정 오류가 수정되어, 이제 최대 10,080분(7일)까지 과거의 시도 기록을 정확하게 조회할 수 있습니다. 덕분에 일주일 단위의 실험 로그나 연속적인 시뮬레이션 데이터를 분석할 때, 데이터 누락 걱정 없이 더 넓은 범위의 히스토리를 안정적으로 확인할 수 있습니다. 또한, 다양한 배포 환경의 버전 동기화가 완료되어 더욱 일관되고 안정적인 연구 워크플로우 구축이 가능해졌습니다. - vv0.3.9.17/12/2026
Selvedge v0.3.9.1 업데이트에서는 실험 프로토콜이나 데이터의 이름이 변경되더라도 이전의 시도 기록을 정확히 찾아내는 '퍼지 리콜(fuzzy recall)' 기능이 도입되었습니다. 이를 통해 실험 조건이 빈번하게 바뀌는 복잡한 연구 환경에서도 데이터의 연속성과 히스토리를 안정적으로 유지할 수 있습니다. 또한, 에이전트 실행 기록(Agent Trace)을 내보내는 기능이 추가되어 AI 기반의 실험 분석 과정을 더욱 투명하게 검증하고 기록할 수 있게 되었습니다.
🧪 Related Code of Life
No related Code of Life posts yet.