If you don't like it, you get your money back

An AI repeat-purchase agent,for the customer who bought only once.

It writes to each customer when it's time to buy again, offers what goes with their purchase and builds an offer with the payment link. Based on your store's purchase history, not a generic list.

Book a free consultationSee what the agent does
Free demoYou own the codeBuilt around your business

What happens today at most online stores. The customer is already there; what's missing is writing to them again.

  1. 01

    Half of customers buy once and never come back.

  2. 02

    Every year it costs more in advertising to get new customers.

  3. 03

    Someone who bought a product that runs out is never told when to reorder.

  4. 04

    Nobody offers them what goes with what they already bought.

  5. 05

    Promotions go out the same to everyone, whatever each person bought.

  6. 06

    The purchase history is sitting in the store, but nobody uses it to decide who to write to.

What the agent does. Each piece answers one of those problems, with your store's customers.

01

Writes when it's time to restock. It works out when each customer will run out of what they bought and writes at that point, with the same product ready to order.

02

Offers what goes with it. Based on what each person bought, it offers the products that go with their purchase, from what's in stock.

03

Builds an offer per customer. It prepares an offer made for that customer, within the discounts the store defines, with the payment link ready.

04

Uses the store's history. It works from each customer's real orders and records who it wrote to, what it offered and what they bought.

Projects we built

See the portfolio

For your demo

A free demo, with nothing invested before you see it.

$0

Of the code is yours

Deployed on infrastructure in your name.

100%

From idea to production

Weeks, not months.

Weeks

We pick the stack the project needs

  • Claude Code
  • Codex
  • Cursor
  • Next.js
  • React
  • TypeScript
  • Node.js
  • Supabase
  • Neon
  • Vercel
  • PostgreSQL
  • React Native
  • Python
  • .NET
  • Docker
  • AWS
  • Google Cloud

How we work. From idea to production in under a month, with no months of discovery before you see something working.

  1. Step 01

    Discovery

    We work out what your company does and which part hurts most. Half an hour on a call, not a three-week audit: the scope and the price come out of it, both closed.

    • Requirements
    • Technical planning
    • Scope and timeline
    RRelevamiento
    AlcancecerradoEquipo3Entrega12 sepEstadoen curso
    ProcesosEntrevistasAlcancePlazos

    Procesos que tocamos

    Cómo entra un pedido hoyrelevado
    Quién lo carga y en qué sistemarelevado
    Dónde se traba la operaciónrelevado
    Qué se factura a manopendiente
    Qué reportes pide gerenciapendiente
    Qué se integra con el ERPpendiente
  2. Step 02

    Design and architecture

    The blueprint: screens and data model, approved by you. You see them before a line of code exists, so changing something costs a conversation instead of a week.

    • Interface design
    • Architecture
    • Data model
    DiseñoPantallas del sistema3 de 4 aprobadas por vos
    Listado
    Ficha
    Alta
    Reporteen revisión
    Pedir un cambio
    Modelo de datos
    clientes
    • id
    • razon_social
    • cuit
    pedidos
    • id
    • cliente_id
    • total
  3. Step 03

    Development

    You see progress every week and give feedback on something real, not on a status report. We build with Claude Code, Codex and Cursor, on Neon and Supabase.

    • Sprint-based work
    • Code review
    • Regular demos
  4. Step 04

    Testing

    That it works with your data and your people, not just on our machines. We load real cases and someone from your team tries it before it goes live.

    • Functional testing
    • Performance
    • Security
    Corrida #128rama entrega · hace 4 min1 para revisar
    142pasaron1para revisar3,4 sla corrida entera
    • Importar 12.000 filas del ERP1.840 ms
    • Permisos por rol120 ms
    • !Facturar sin CUIT cargadodeja pasar la factura sin CUIT—
    • Cerrar caja con pedidos abiertos310 ms
    • Remito con stock insuficiente95 ms
    • Alta de cliente duplicado64 ms
  5. Step 05

    Launch and support

    It ships documented and with your team trained. The infrastructure is in your name and the code is yours: we stay on if you want us to, not because you depend on us.

    • Deployment
    • Documentation
    • Ongoing support
    sistema.tuempresa.comen línea
    Pedidos del díaNuevo

    38

    Del día

    12

    A facturar

    4

    Demorados

    PedidoClienteTotalEstado
    PD-4417Ferretería Sur184.500confirmado
    PD-4418Corralón Andes62.300confirmado
    PD-4419Distribuidora Río910.000frenado
    PD-4420Metalúrgica Paz245.900confirmado
See the process step by step

Shall we start?

Thirty minutes, no strings. You leave with a concrete recommendation, whether we go ahead or not.

Book a call

Questions about the AI repeat-purchase agent for ecommerce. The ones we hear most before the first call.

Text message, email or both, depending on the permissions the customer has given. Anyone who opted out is respected, and the maximum frequency is set by the store.

Yes, it connects through an API to the platform to read products, stock, customers and orders, and to generate the payment link. If the store is custom-built, it connects to its database or its API.

From USD 3,000 to build it, plus maintenance from USD 300 a month. The exact price depends on the process and the integrations: we scope it on the first call and fix the price in writing. AI model usage is billed separately.

It says so and hands the conversation to a person with the full history. When it hands off, and which actions need sign-off from someone on your team, is defined during scoping.

First we build a demo with real cases from your business, so you see how it responds before it's connected to anything in production. Then it's connected with minimum permissions, every action is logged, and for the first few weeks everything it does gets reviewed.

See every question

Last step. Thirty minutes to understand how you operate and tell you what to build first.

No strings. You leave with a concrete recommendation, whether we go ahead or not.

Pick a day and time