Checkpoint (AI Session)
A checkpoint is a saved state — message log, scratchpad, tool output — that lets an agent resume or branch a long-running session without losing context.
In plain English
A checkpoint is a save file for an agent run. It captures the state needed to pick up where you left off — the message history, the scratchpad, the current plan, any important tool outputs. You can resume the run hours later, branch it to try a different path, or replay it for debugging.
For short chat sessions, checkpoints are overkill. For long agent runs they are the difference between a product that feels solid and one that feels flimsy. If the process restarts, the network blips, or the owner closes their laptop, a checkpointed session survives. Without it, every failure means starting over.
Why it matters for Black Box
Black Box checkpoints every CEO turn to persistent storage. Owners can close the tab, come back the next day, and the session resumes — including any pending approvals in the inbox. The UI also surfaces "branch from here" so owners can explore alternate paths without losing their main thread.
Examples
- A session that resumes seamlessly after a server restart.
- Branching a checkpoint to try a more aggressive tone, then picking the winner.
- Replaying a historical checkpoint to reproduce a bug in agent behavior.