Custom AI agents, connected to the systems your company already runs

An AI agent is not a chatbot that answers questions. It is a system with access to your data and your tools, which carries out tasks end to end and knows when it has to ask a person for approval.

What an AI agent actually is

A generic assistant answers from what it learned on the internet. A custom AI agent works with three things the generic assistant does not have: access to your company data, access to your tools, and rules about what it may do on its own.

In practice that means that instead of explaining how to fill in a proposal, the agent pulls the client history from the CRM, applies your current price list, generates the document and leaves it ready for review. The difference is not the language model. It is the connection to the business.

The three components that make an agent work

Context: controlled access to company documents, databases and history, so answers are grounded in your reality rather than an internet average.
Tools: the ability to act in your systems, such as creating a CRM record, sending an email, updating a spreadsheet, checking stock or opening a ticket.
Rules and limits: an explicit definition of what the agent decides alone and what needs human approval, with a log of everything it did and why.

Why most AI projects never get past the experiment

Most companies have already tried AI. Few have managed to put it to work in the processes that make money. The problem is almost never the model. The problem is that the tool does not know the company data, cannot act in the systems where the work happens, and nobody defined what it may do unsupervised.

Signs you are at this point

The team uses public assistants for daily tasks, with no control over what information leaves the company.
Someone tested an AI automation, but it stalled because there was no safe way to connect it to the ERP or the CRM.
People still copy data between systems by hand, every day, because no integration covers the real cases.
Customer replies take hours or days, and the information needed is spread across email, files and people's heads.

Where an agent really pays off

We deliberately pick back office processes with an obvious metric, because that is where the return shows up first and is easiest to prove.

Sales

Lead qualification and follow-up

The agent reads incoming requests, cross-checks the CRM history, scores them for fit and drafts the first reply. Contacts with potential reach the sales person with context; the rest still get an answer.

Customer support

Answers grounded in your real documentation

The agent answers from your manuals, contracts and case history, and shows the source of every answer. Anything it cannot resolve is routed to the right person, already summarised.

Admin and finance

Data extraction from invoices and contracts

Instead of someone copying figures from PDFs into the system, the agent extracts them, validates against business rules and flags anything unusual for human review.

Operations

Reports and alerts on current data

The agent queries the systems, combines the information and produces the report on whatever schedule you set. When an indicator moves outside the expected range, it tells whoever has to decide.

To the systems you already use

Privacy and control, defined upfront

We do not ask you to change systems or migrate data to a new platform. We connect to what already exists, through an API where there is one and through a custom integration where there is not, and the agent writes back into the same records your team looks at every day. When the system the AI needs to work in does not exist or does not fit, we build that too.

This is the question that stops most purchases, so we deal with it at the start rather than at the end. The systems we build run on infrastructure in the European Union, in accounts your company controls, and your data is not used to train public models.

Portuguese business software: Primavera, PHC, Sage, Moloni, InvoiceXpress and legacy systems with no API
CRM: HubSpot, Salesforce, Pipedrive, or the custom CRM you already run
Microsoft 365 and Google Workspace, including email, calendars and shared files
Internal databases, spreadsheets and document folders
WhatsApp Business, website chat and other conversational channels

Each agent only accesses the information it needs, with permissions by user profile.
Answers cite their source, so anyone can verify where the information came from.
Actions with real impact, such as sending a proposal or changing a record, only run unattended if you explicitly allow it.
There is a log of what the agent did, when, and with what data, for internal audit.
We sign a confidentiality agreement by default, before any technical conversation.

How it is done, phase by phase

We work in short phases with a formal decision point at the end of the first, so you can stop if it does not make sense to continue.

01
Diagnosis
30 to 45 minute session, free
We understand the operation and pick candidate processes with you. You get a written summary of which ones have the best return-to-effort ratio, and an estimate of hours saved.
02
Design and validation
2 to 3 weeks
Process map, proposed architecture, list of required integrations and a definition of what the agent decides alone. You end up with a plan with scope, timeline and investment fixed.
03
Building the first agent
4 to 6 weeks
One agent, in production, connected to a real system, with demos every 2 weeks. We test against real cases from your business before anyone depends on it.
04
Rollout
Depends on scope
With the first agent working and measured, we extend to more processes or departments. This is where it makes sense to talk about an internal platform.

Investment

We give reference bands upfront to save both sides time. The exact figure is fixed in the proposal, after the diagnosis, and does not change midway.

First agent

€7,500 to €15,000

One agent in production, connected to a real system, delivered in 4 to 6 weeks. The figure varies mostly with the number of integrations.

Internal AI platform

€25,000 to €45,000

Private knowledge base, several department agents and ERP or CRM integration, with access control by profile.

Extended platform

From €45,000

Multiple departments, agents that take action in your systems, and ongoing supported evolution.

Frequently asked questions about AI agents

A chatbot talks. An agent acts. The chatbot answers from a script or a set of documents; the agent has access to company tools and can complete the task end to end, such as creating the record, sending the email or updating the system. An agent may have a chat interface, but the value is not in the conversation, it is in the action.

Typically 4 to 6 weeks once the design is validated, and the design takes 2 to 3 weeks. Counting from the first conversation, it is reasonable to expect around two months until something is in production. Processes with many integrations, or legacy systems without an API, take longer, and we say so in the proposal.

No. If everything were organised you probably would not need us. Part of the job is precisely dealing with information scattered across files, emails and systems that do not talk to each other. What we need to know is where the information lives and who is allowed to see it.

We agree with you which actions the agent may run alone and which require human approval. For anything that affects a customer or money, the default recommendation is human approval. The agent prepares, the person confirms. On top of that, answers cite their source, so a mistake is detectable rather than invisible.

No. The systems run on infrastructure in the European Union, in accounts your company controls, and the data is not used to train public models. We define in writing what information each agent may read, and usage is logged.

The code and the infrastructure are yours. We hand over technical documentation, access and team training. If you want to move maintenance to another team or bring it in-house, you can do it without renegotiating anything with us.

Next step

Want to know what this would do in your operation?

Free diagnosis, no commitment. We look at your process and come back with a plan covering scope, timeline and investment. Reply within 48 business hours, NDA included by default.

100% free diagnosis
No commitment to proceed
Reply within 48 business hours
NDA included by default

Request free diagnosis

Two minutes to tell us the essentials. We reply within 48 business hours with a first read on your case and, if it makes sense, a proposed session.

Explore other solutions

Every vertical has its own challenges. See how we approach other sectors where we have proven experience.

Digital Brain