AI Support That Escalates Instead of Guessing by Ever F.AI Support That Escalates Instead of Guessing by Ever F.

AI Support That Escalates Instead of Guessing

Ever  F.

Ever F.

TRY IT
Open the live Lumen & Leaf storefront at https://lumenleaf-support-agent.vercel.app and ask Leaf about a product or store policy. Then ask about something the store has never documented.
This is a fictional demo. Use only the sample order details shown here. Do not enter real personal, payment, or order information.
THE PROOF
Leaf does not invent a loyalty policy. It admits that the information is missing and gives the shopper a clear way to reach support.
Leaf admits the loyalty policy is missing and sends the shopper to support.
Leaf admits the loyalty policy is missing and sends the shopper to support.
Both sample order fields must match before any status appears.
Both sample order fields must match before any status appears.
A customer message cannot create a new discount or promotion.
A customer message cannot create a new discount or promotion.
THE PROBLEM
Routine support questions take time, but a fast wrong answer is worse than a slow one. A store chatbot can create a real mess if it invents a return policy, reveals an order from partial information, or promises an action the business never approved.
WHAT I BUILT
I built Leaf as a small support service for the fictional Lumen & Leaf store. It answers from the catalog, policies, FAQ, and fictional sample orders.
I kept order verification out of the language model. The complete order number and exact email must match the same sample record before Leaf can reveal a status or tracking number. That makes the lookup less conversational, which is the right trade when order privacy is involved.
Questions about missing policies, medical claims, account changes, or other sensitive actions go to a person.
HOW I TESTED IT
I reran the release checks on July 23, 2026. All 30 local checks passed. They cover twenty acceptance prompts, order privacy, unsupported medical claims, model output leakage, authentication, health checks, and the public API.
I also ran 25 checks through the public Vercel site to the private gateway. Those covered the same twenty prompts, three new questions that required grounded answers, and two extra privacy cases. All 25 passed.
The harder cases included a fake order, a partial order lookup, an undocumented gift wrapping policy, a private phone number request, a medical cure claim, and the 90 percent discount attempt shown above.
WHAT PASSED
• Documented questions returned answers from approved facts. • Partial or mismatched order lookups exposed no sample order details. • Unknown questions reached the visible support handoff. • The public widget kept no chat history and warned visitors that AI can make mistakes.
LIMITS
Leaf is not connected to a live Shopify order account. It cannot issue refunds, change an order, or move money. Its job is narrower: answer from store documents, verify fictional sample orders, and hand everything else to a person.
BUILT WITH
Node.js, deterministic privacy rules, OpenRouter compatible grounded answers, Docker Compose, Next.js, Vercel, and adversarial acceptance tests.
STARTING PRICE
AI Support Agent: $600 plus $120 per month. The final quote depends on integrations, usage, compliance needs, and the support agreement.
Fictional portfolio demonstration. The screenshots and test counts come from controlled release testing, not a client deployment or revenue result.
Like this project

Posted Jul 21, 2026

Leaf answers from store documents, verifies fictional sample orders in code, and sends unknown questions to a person.