# Checkpoint

> The pager agents use when they need a person.

Checkpoint is a human-in-the-loop pager for AI agents. When an agent is mid-task and needs a fact, an approval, or a judgment that only a specific person has, it pages that person through Checkpoint. The human answers in the context of the task. That answer goes back to the agent so it can continue.

Checkpoint routes the page by role, ownership, or on-call. It is not a generic chat handoff, and it does not ask the agent to guess an email address.

## What a page contains

- Why the agent stopped
- The question or decision it needs
- Who should answer: a role or owner, not a guessed inbox
- The task context the human needs in order to reply

## What comes back

The human's answer, attached to the same task, so the agent continues with that context instead of inventing it.

## Status

Checkpoint is in early access. There is no public API yet.

## Pages

- [Home](https://checkpointdemo.ca/): what Checkpoint is
- [How it works](https://checkpointdemo.ca/how-it-works): the loop from stall to reply
- [For agents](https://checkpointdemo.ca/for-agents): what an agent would send and what it would get back
- [Console](https://checkpointdemo.ca/console): a signed-out preview of a page inbox
- [Early access](https://checkpointdemo.ca/early-access): request access