Before your AI acts on anything sensitive — refunds, emails, data deletion — mild.run intercepts and asks you first. No terminal, no SDK, no webhooks. Approve or block from your preferred channel in seconds.
Your AI reaches a critical operation — a Stripe refund, an email blast, a record deletion. Instead of firing immediately, it pauses and calls mild.run.
mild.run sends you a notification with full context: what the agent wants to do, why, and how much. The agent waits — up to 24 hours.
Tap Approve or Block from your phone. The agent acts only with your explicit sign-off. Every decision is logged and auditable.
HubSpot, Notion, custom actions, and much more — expanding fast.
Coming soonNo new app to install. Approval requests are delivered to your preferred channel — approve or block with a single tap.
More channels coming soon.
Perfect for trying mild.run on personal projects.
For developers running agents in production.
Human-in-the-loop means an AI agent pauses before a sensitive action — a refund, an email, a deletion — and waits for a person to explicitly approve or block it. mild.run adds this gate to any agent through one MCP tool call.
mild.run currently gates Stripe refunds, subscription cancellations, customer deletions and payouts, Gmail sends, and Slack messages. More integrations are added regularly.
mild.run notifies you by Telegram, email or Slack with full context on the requested action. You tap Approve or Block, and the agent receives that decision — waiting up to 24 hours before treating a non-response as blocked.
No SDK or webhook server is required. mild.run is exposed as an MCP tool, so any MCP-compatible agent can call it directly to request approval.
Yes. The Free plan includes 30 approvals per month across up to 2 integrations. The Solo plan is $12/month for unlimited approvals and all integrations.
Join developers shipping AI agents with confidence. Free tier, no credit card required.