built a small open source tool to stop AI agents from regressing after changes

one of the most annoying problems when building AI agents: fix a failure, change something, same failure comes back quietly.

built replayd for this. captures failed runs as regression tests and replays them before you ship. catches the failure if it returns after a prompt, model, or tool change.

v0.1.2, pip installable, open source.

pip install replayd

star it if you want to follow progress.