A private AI for your company: the model runs on your side, nothing leaves.

A private AI is a language model deployed inside the company infrastructure: your servers, or a sovereign French cloud. The model is isolated: no information travels outside, not even to us.

On that base, the system indexes your documents, contracts and procedures, and answers within seconds: augmented search over your own knowledge base, served by an interface in your branding, straightforward to pick up.

The daily use is prosaic, and that is its strength: thirty seconds to find what your teams currently spend thirty minutes looking for.

The problem a private AI solves is a problem of travel: every question asked of a public assistant leaves with its context, and often with the document it concerns, towards servers the company does not control.

These uses are already there. An Odoxa survey in France measures that 12 % of employees use AI at work, and that nearly one in two has never been trained for it: in an SME, questions get asked on personal accounts, with client data, before any written rule.

When an incident happens, the employer answers for it. A private AI does not replace the usage policy, but it changes the nature of the risk: the default channel becomes one that lets nothing out.

The PropulAI dossier orders its offers by depth of data access, and figure 04, taken from the main dossier, shows it in cross-section: three trays inside an enclosure that nothing leaves, whatever the level.

Level N1 touches no data: that is team training. Level N2 reaches your documents: that is the private AI described here, our Bastion offer. Level N3 connects your ERP and prepares management actions: that is the copilot, and every action passes through human validation.

Each level includes the previous one, and the dossier maxim holds for all three: the machine reads, reconciles and alerts; decisions stay human.

The default architecture fits in three components: a European model of the Mistral kind, French infrastructure or directly your servers, and data stored in open, standard, exportable formats.

Each component is replaceable, and that is written into the contract with the exit clause: models used, location, conditions for leaving. If you leave, nothing locks shut: you take the installation over, or hand it to whoever you choose.

No client data goes to an American API, including during demonstrations.

The installation follows our five-stage engagement protocol: Diagnostic, Scoping & compliance, Pilot, Deployment, Managed operations. You decide at each step, on figures measured at your premises.

The engine is demonstrated on public data before it touches yours. Once in production, operations are contractual: models evolve, your data changes, an unmaintained infrastructure degrades, and a support pack keeps it up to date.

We are paid for the project, never for reselling licences: the choice of components follows your situation, not a catalogue.

Our agency runs ten internal vertical SaaS products in production, and has delivered more than fifty sites and platforms since 2024: our own tools rest on the components we install for our clients.

DocAgora, our e-health platform in Portugal, operates in a stricter setting than document search: health data, a supervised conversational agent, human oversight.

Detailed case studies are available on request, during the diagnostic. We would rather run a demonstration than publish unverifiable figures.