Ombú Sofás

Overview1/6

Sample data

Interface with sample data

An AI sales agent for a furniture store, built in n8n with xAI Grok. It answers prices, stock, delivery times and installments from the catalog, computes every amount in code, and hands custom orders to a person through the CRM. Showcase demo with sample catalog data.

The problem & the solution Try live demo

Live demo · Sample catalog data for a fictional store. Prices are shown in Argentine pesos (ARS) and some product names stay in Spanish.

My role: Automation & AI development
Built with: n8n
xAI Grok
Docker
nginx
Cloudflare
Kommo API (mock)
In focus: AI sales agent with a catalog tool
Amounts computed in code
Human and CRM handoff

The problem & the solution

A sales agent that quotes from the catalog, not from memory.

The challenge

Furniture shoppers ask the same things all day, often on WhatsApp: how much is this model in this fabric, is it in stock, when can it arrive, what does it cost in installments. A generic chatbot answers fast but can invent a price, and a wrong amount in a sales chat costs trust. Custom sizes and complicated requests still need a person.

Before: 12 setups
xAI models and reasoning levels compared on the same 28 Spanish test conversations, with every amount checked against the catalog.
After: 0 invented amounts
For the chosen setup across 190 test conversations, with a median of 1.7 to 1.9 seconds per turn in the benchmark.

Our own benchmark and live check, run October 17, 2026 on sample catalog data. Not an independent audit.

Start with the questions customers really ask.

Suggested prompts show what the agent handles: a price in a given fabric, a 12 installment follow-up that remembers the sofa, an out-of-stock model, the cheapest option, a fabric comparison and a custom size.

Ombú Sofás · Suggested prompts · Sample data — Suggested prompts · Sample data

Limits run before any paid AI call.

Message length, per-chat and global limits are checked first. Rules route custom sizes and seller requests to the handoff; everything else goes to the agent, which can only read the catalog.

Ombú Sofás · n8n workflow · Sample data — n8n workflow · Sample data

One table holds the truth.

The catalog tool reads the data table on every call and applies the pricing rules in code. Change a price or a stock row and the next answer changes, with no redeploy.

Ombú Sofás · Catalog tool · Sample data — Catalog tool · Sample data

A person takes over with context.

Handoffs build the exact Kommo lead payload and store it in a mock table: name, pipeline stage, tags, requested size, fabric and the original message, so a seller can follow up with a firm quote.

Ombú Sofás · CRM leads (mock) · Sample data — CRM leads (mock) · Sample data

The difference

In our benchmark of 12 xAI model setups, the chosen setup invented no amounts in 190 test conversations. In a live check of 8 prompts on the Spanish chat, replies arrived in 0.6 to 4.4 seconds. The demo is public in English and Spanish: try the chat, then open the behind-the-scenes page to see the workflow.

Showcase demo built and hosted by Fernando Marcano, based in the Dominican Republic, on his own VPS behind Cloudflare. Ombú Sofás is a fictional store: products, prices and stock are sample data. The English chat shows sample prices in Argentine pesos (ARS) and some product names in Spanish; leads go to a mock CRM table, not a real Kommo account.

Something worth building?

Tell me what you have in mind.

Let's talk