Pause any AI agent at a critical step. Wait for human approval. Resume with one API call. No message brokers. No infrastructure. Just one HTTP request.
Your agent calls POST /api/checkpoint with the task description and current state. It gets a checkpoint ID back.
A human gets notified (email, Slack, dashboard). They review what the agent wants to do and approve or reject it.
On approval, the agent's webhook fires with the original state. It picks up exactly where it left off. Rejection includes a reason.
// In your agent's critical step:
const checkpoint = await fetch("https://www.agent-utils.com/api/checkpoint", {
method: "POST",
headers: { "x-api-key": "au_...", "Content-Type": "application/json" },
body: JSON.stringify({
agentName: "deploy-bot",
taskDescription: "Deploy v2.3.1 to production",
state: { version: "2.3.1", target: "prod" },
webhookUrl: "https://your-agent.com/resume"
})
});
// Agent sleeps. Human approves via dashboard.
// Webhook fires → agent resumes with original state.The EU AI Act mandates human oversight for high-risk AI systems. Checkpoint gives you auditable approval logs out of the box.
Agents making production deployments, financial transactions, or data deletions? Add a human gate before the irreversible action.
No Redis. No message queues. No state machines. One API call pauses your agent, one webhook resumes it. We handle the rest.
Every approval/rejection is logged with timestamps, reasons, and the full agent state. Export for compliance audits.
Free tier: 100 checkpoints/month. No credit card required.
Get your API key →