AI Pulse · Under the Radar

Most businesses are starting to use coding agents, but few have a clear view of what those agents are actually changing across a codebase. AgentTrail is a small open-source project aimed at that gap. It watches Claude Code, Codex and Cursor sessions locally on 127.0.0.1, then shows a live dependency graph of repository-level work. There’s no cloud service and no account required.

That matters because token counts and response times don’t tell an engineering leader much about operational risk. The useful question is simpler: which files, services and dependencies did the agent touch, and what could those changes affect? A live view of the work can help a team spot when an agent’s task reaches further into a repository than expected. It also gives technical leaders a more practical way to review AI-assisted work before it becomes a production problem.

AgentTrail is still early. It has 262 GitHub stars, has been gaining around 53 stars a day, and shows active issue and pull request traffic only days after launch. That is interest, not proof that it is ready for every company. Still, local-first tools like this are worth watching, particularly for teams with strict data handling requirements or a preference not to send developer activity to another cloud platform.

This is the kind of thing we build into an AI command centre, giving operators a clearer picture of where AI is working and where it needs oversight.