Skip to Content

AI agents AI agents for business, connected to your data

An agent queries your ERP, support desk and email, answers by telling you where each piece of data comes from and gets ready whatever needs doing. A person decides.

  • Odoo
  • Support
  • Email
  • Accounting
  • Projects
The problem

An AI that can't see your data can only talk

A generic assistant writes well and knows nothing about your clients, your invoices or your tickets. To get a useful answer you have to copy the data into it by hand, which saves no time and isn't safe.

The other extreme doesn't work either: an agent allowed to do everything, sending emails or changing the accounts without anyone seeing.

  • Answers that sound right and that nobody can check
  • The same data looked up in the ERP, in email and in the support desk
  • Admin tasks that depend on a single person
  • Fear, with good reason, of letting an AI act on its own
What we do

Agents with a specific job and only the permissions they need

Each agent does one thing and can reach only what it needs to do it.

Questions in plain language

Which invoices fall due this week? Which tickets have gone two days without a reply? The agent queries Odoo, the support desk or email and answers with the data and where it comes from.

Accounting agent

Records supplier invoices from the PDF, reconciles the bank and flags what's missing. It doesn't pay or file taxes: it leaves everything ready for whoever signs.

Project coordination

Reviews the projects every week, flags the ones drifting from budget and brings a short list of decisions to the meeting.

Company knowledge

Answers with what the company already has in writing: procedures, resolved tickets and documentation. And it says which document each answer comes from.

Support assistance

Suggests a reply to the technician based on previous tickets and leaves it as an internal note. A person always replies to the client.

Permissions and logging

What each agent can read and write is set on the server, not in an instruction it could get round. Everything it does is logged under its name.

How we work

From the first meeting to an agent in production

Discovery

One free hour to choose a specific process where an agent can genuinely take work off your hands.

Fixed proposal

What the agent will do, what data it can reach and what it can't do, in writing and at a fixed price.

Pilot with your data

We build it on a well-defined case and test it with the real questions your team asks.

Production

Final permissions, logging and a usage cap. Your team learns how to ask it for things and how to review what it leaves.

Improvement

We review with you where it gets things right, where it doesn't and whether another task deserves its own agent.

Case studies

The agents we work with ourselves

Read on

The rest of our AI line

AI and automation

Corporate assistant

Your email, documents and ERP searchable in natural language, with answers that cite their source.

AI and automation

Process automation

We connect your software with n8n and Make so data moves from one system to the next on its own, with alerts if anything fails.

AI and automation

Safe AI and training

Rules of use, approved tools and training so your team uses AI without exposing data.

Cybersecurity and GRC

Before automating, sort out the permissions

Cybersecurity

Data loss prevention

Classifying your information and controlling email, cloud, USB and web uploads, so your data doesn't leave the company.

GRC and ESG

Responsible AI and the AI Act

An inventory of the AI your company uses, risk classification, rules of use and a management system under ISO 42001.

Frequently asked questions

What people ask us about agents

What's the difference between an agent and a chat tool like ChatGPT?

A chat tool answers with what it knows in general. An agent is connected to your systems, looks up your data and gets a task ready: a draft email, an accounting entry, the reply to a ticket.

Can an agent do things without anyone approving them?

Only the ones you decide. Usually it can read freely, and anything it writes goes through a person: the email stays in drafts and the payment is left unexecuted.

Which software does it connect to?

Odoo, Microsoft 365 or Google Workspace email, and any software that has an API. If it doesn't have one, we look into it before committing to anything.

Which AI models do you use?

Those from the main providers, always with a business account. We choose according to the task and the cost, and the agent isn't tied to one model: it can be changed.

What if our data can't leave the company?

We install the model on one of your servers or on a dedicated one of ours, and the agent works without sending anything outside. Models that fit on your own server are less capable than the big cloud ones: before we start, we tell you which tasks they handle well.

What if the agent gets it wrong?

Now and then it will, just like a new member of staff. That's why it says where each piece of data comes from, keeps a log and never carries out on its own anything that commits the company. Mistakes are caught before anything goes out.

What task would you give an agent?

Tell us what work keeps repeating in your company. In an hour we'll see whether an agent can take it off your hands.