WRUD

wrud is an open-source command-line interface designed to meticulously record the activities of your agents during each session. This includes logging tools utilized, files accessed, specific models engaged, and token consumption, all stored in a local SQLite database. Key features: • Local activity logging for agents • Records tools, files, models, and tokens • Identifies trivial work by larger models • Includes built-in support for agents like Claude Code & Cursor • Supports any agent via its SDK • Operates entirely locally with no cloud dependencies wrud helps users optimize their operations by highlighting instances where high-capacity models are used for simple tasks, facilitating model right-sizing and reducing unnecessary context costs. It also captures recurring errors, turning them into actionable lessons to improve agent performance over time. The system provides a plain-language recap of each session, summarizing actual activities instead of just raw statistics. All data remains on your machine, ensuring privacy and data sovereignty. This tool is ideal for individual developers, engineering teams, and anyone managing agent-driven workflows who needs detailed visibility into agent behavior, cost optimization, and performance improvement without relying on external cloud services or proprietary analytics.