selvedge

Selvedge provides long-term memory for codebases written or modified by AI agents, addressing the challenge of understanding the intent behind code changes after the fact. Unlike traditional tools that only show what changed and when, Selvedge captures the reasoning—the why—directly from the AI agent at the moment a change is made. This information is stored locally, enabling users to retrieve the motivations for specific changes even after the original session, prompt, or model version are no longer accessible.

The tool operates as a local MCP server that AI agents such as Claude Code, Cursor, and Copilot can call while working. Through commands like log_change, Selvedge records both the change and its rationale in the same context window that produced the modification. email or api/v1/checkout—rather than by line numbers, making it easier to search for all changes related to a particular part of the codebase. Users can query the stored history using commands like selvedge blame, selvedge history, or selvedge search, retrieving explanations for changes and summaries of previous attempts on specific entities, including outcomes and reasons for rejected paths.

Selvedge is delivered as a Python package installable via pip, with a setup process that detects supported AI tools and integrates with their configurations. selvedge directory and operating entirely locally—no accounts, external servers, or telemetry are involved. The tool can also be wired into editors such as VS Code, Cursor, and others, either automatically or through provided configuration steps. bak) to preserve the original state.

The platform is open source under the MIT license. It exports its history in the Agent Trace format, enabling integration with existing observability or audit tools that support this format. Selvedge is intended for teams managing long-term AI-coded projects as well as solo developers who want to retain the context behind their own or their agent's code changes over time.