built run replay for AI agents.
whhen a run fails halfway search succeeded, GitHub failed, you don't start over.
you resume from the failure point. the agent picks up exactly where it stopped.
Day 11 building an AI start-up from Africa.
Worked on the tool re-run given yesterday we were storing the complete state of the tool call steps to the disk