Assistant over your documentation
An agent that answers from your manuals, procedures and knowledge base, and cites where each answer came from. Useful for the service desk, for onboarding new staff and for the questions that repeat every week.
AI agents
AI on your data,
AI on your data,
without sending it out
Agents that work on your own systems: they answer questions from your documentation, classify tickets, draft replies and flag what is about to break. With models that can run on your infrastructure, so the information never leaves the company.
What it covers
From the repetitive task to the agent
Classification and routing
Incoming tickets and emails are classified, prioritised and routed to the right team as they arrive. The agent proposes the category and a first draft; a person approves before anything goes out.
Copilot for the technical team
An assistant that cross-references monitoring alerts, logs and change history to suggest where to start looking. It always starts read-only: it recommends, it does not execute.
Models on your infrastructure
With Ollama or another local engine, the model runs on your server and data never leaves the network. For anything non-sensitive an external service can be used, decided data by data rather than by default.
Integration with what you run
The agent connects to GLPI, the ERP, the CRM, monitoring or the phone system through their own interfaces. It is not an isolated chat window: it works with the systems you already have.
Controls and traceability
Every agent action is logged, and anything with consequences requires human confirmation. Permissions are tight: an agent never has more access than its task requires.
When it makes sense
Signs it is worth a look
- Your team answers the same twenty questions every week
- Tickets are classified by hand and misrouted half the time
- You have years of documentation nobody can find when they need it
- You want to use AI but cannot send client data to an external service
- Someone spends hours a week copying data from one system to another
Frequently asked
About AI agents
Does company data end up training someone else’s model?
Not with local models: the model runs on your server and nothing leaves the network. When an external service is worth using for capacity, we define explicitly what information may leave and what may not, and filter before sending. That is a design decision, not a setting reviewed later.
What if the agent gets it wrong?
We assume it will, which is why it starts by assisting rather than deciding: it proposes the answer, the classification or the diagnosis, and a person approves. It gets autonomy only where an error is cheap and reversible, and always with a record of what it did.
What do we need to start?
One concrete, boring case: the question that repeats most, the classification that eats the most time. We implement that one, measure whether it genuinely saves time, and only then add the next. AI projects that start out trying to automate everything never reach production.
Start with one concrete task
We map which repetitive tasks consume the most hours in your team and which can be automated with a human in the loop. That produces the first agent, not a platform.