Agents that read, decide and act in your processes, and call a person when they should.
Not a chatbot. Software that receives an email, a document, a review or a message, understands what it is, decides what to do with your company's rules and executes in your tools. When the job is customer contact, it talks too.
Who it's for
Companies with repetitive decisions that today only a person can make: classify, extract, route, reply, approve. Teams that spend the day reading email, attachments and portals to feed other systems. Businesses with repeated questions from customers or staff, missed calls and bookings left undone.
Two kinds of agent, one foundation
Internal agents (workflows)
They work with nobody talking to them: email and document triage, data extraction into the ERP or CRM, review replies, daily reports from your data, content quality control, proposal preparation. They run on n8n, with an AI model chosen per task.
Contact agents
They answer customers or staff on WhatsApp, email, phone or browser, from your documentation (RAG), with access to the calendar, CRM and orders, and hand over to a person when they should.
What's included
- A map of the decisions the agent makes and the ones that stay with a person (the escalation rule, written before any code).
- Access to your tools: email, calendar, CRM, ERP, invoicing, portals, via API or automation.
- Knowledge base built from your documentation (RAG), with the source shown in every answer.
- Tests with real cases (emails, documents, conversations) before going live.
- A dashboard with tasks completed without intervention, escalations, accuracy and cost per run.
What you get
An agent in production, internal or customer-facing, with metrics for tasks resolved without intervention, escalations, accuracy and cost.
What it isn't
A generic chatbot that says whatever it likes. A replacement for your team. An agent making decisions the company wouldn't trust an intern with on day one.
Real examples
Field Aerospace (aerospace, about 250 people, US) uses AI workflows on n8n to read solicitations and extract requirements: hours of manual work became 15 to 20 minutes, and a proposal draft reaches 80% in about 25 minutes (case published by n8n).
XIBIX Solutions (IT services, about 120 people, Germany) put an internal agent answering HR questions from company documentation; repeated questions to the HR team fell by more than 50% (case published by n8n).
Well-configured text agents in e-commerce resolve roughly a quarter to half of support requests on their own (cases published by Gorgias, 26% to 56%). The rest go to the team with context.
Before you get in touch
Isn't this just automation?
An automation follows fixed rules: if X, do Y. An agent reads, interprets and chooses between options, with the rules we give it and a clear limit on what it can't decide alone. When the rules are fixed, we build an automation; it's cheaper and more predictable.
Does the agent make things up?
It's instructed to answer and decide only from the information it has, to say when it doesn't know, and to escalate. We test that with real cases before going live and measure it after.
Does it work in Portuguese?
Yes, and in English or other languages if needed.
What about the data?
Processed on GDPR-compliant EU servers (or your infrastructure), consent where the law requires it, models used via API with no training on your data.