Checkout rescue with multi-agent handoff

Try it live → Run the multi-agent rescue demo with your own API key.

The shape

Your checkout flow loses users at the price-confirmation step. You want to catch them on exit-intent with a different agent — empathetic, briefer, empowered to offer a small concession.

Linda’s multi-agent + triggers stack does this in one config.

Code

import { Linda, onExitIntent, combineTriggers } from "@linda/core";

const linda = new Linda({
  transport: { mode: "proxy", url: "/api/linda" },
  config: {
    agents: [
      {
        name: "checkout",
        persona: "Helpful, brief checkout assistant. Walk the user through.",
        target: { type: "form", selector: "#checkout" },
        skills: ["payment-intake", "address-verify"],
      },
      {
        name: "rescue",
        persona: "Empathetic save-the-sale specialist. Brief. Offers up to 10% off if it'll close the deal. Never pushy.",
        triggers: combineTriggers(
          onExitIntent(),
          { type: "hesitation", waitMs: 15000 }
        ),
        tools: { coupon: { description: "Issue a discount up to 10%", maxValue: 10 } },
      },
    ],
  },
});

linda.attach();

The user experience

1. User starts the checkout. The checkout agent walks them through.
2. User pauses at the total. 15 seconds of inactivity — or they move the mouse toward the close-tab button.
3. The rescue agent picks up the chat. Different persona, different tools.
4. The handoff is invisible — the user sees one continuous chat.

What makes this work

• Shared VFS. The rescue agent reads /conversation/messages.jsonl to see what was already said. No re-asking.
• Capability-isolated tools. Only the rescue agent has the coupon tool. The checkout agent can’t accidentally offer discounts.
• Behavioral triggers. onExitIntent and hesitation are 10 lines each. Write your own.

Anti-patterns to avoid

• Don’t make the rescue agent persistent. Fire once per session, then back off.
• Don’t have the rescue agent re-explain. It has the conversation log.
• Don’t tie the coupon to LLM judgment alone. Use a hook to validate eligibility.

FAQ