Private AI · Models that stay in

Ollama, the AI that runs on our server for sensitive data

Some data must not go to an AI provider. For that data, we run open models on our own server with Ollama, and we install them on yours.

Open source Vendor’s site

Cost control

“AI at the right cost” pack: pay for AI what it is worth

€2,900 to €5,400

excl. VAT · depending on the number of systems and tasks; each building block is also sold on its own

2 to 3 weeks

See the offer

Our routing table, in short

May go to a provider

Public contentListing translations

Stays with us, on Ollama

HR dataLegal filesHealth data
What may go to an AI provider, and what stays on a model hosted on our server.

What it is

Ollama is open-source software under the MIT licence that runs open AI models on a server you control. It downloads them, serves them, and exposes them through an interface our automations call just like a commercial provider’s. The difference fits in one sentence: the question and the answer never leave the machine.

Why we chose it

Nothing leaves

HR, legal or health data is processed on your server, without passing through a provider.

A fixed cost

No usage-based bill: the server costs the same whether the model works a little or a lot.

Interchangeable

Our automations call it like a commercial provider: moving a task from one model to another does not mean rebuilding everything.

How we use it ourselves

  • Ollama is in production on our server, alongside Twenty, n8n and Matomo.
  • Our routing table sends sensitive data (HR, legal, health) to a model hosted by us, and the rest to the cheapest commercial model that does the task well.

How we install it for you

  1. Sorting

    We write down with you what may go to a provider and what must stay with you.

  2. Model choice

    Several open models compared on your own content, with quality and speed measured.

  3. Install

    Ollama installed on your server or ours, with access reserved for your automations.

  4. Connect

    The model wired into your automations or your document assistant, in the same routing table as everything else.

The offers that include it

Frequently asked questions

Is a self-hosted model as good as ChatGPT?

Not at everything. For sorting, summarising or extracting information, a well-chosen open model is often enough; for long reasoning, commercial models stay ahead. We test on your content before deciding.

Do we need to buy a server?

No: a server rented from a European host is enough. Depending on the model’s size, it needs a graphics card, and we price that beforehand.

Who can see the data?

Only the people and automations you grant access to. The model runs on your server and sends nothing outside.

Our other tools

Twenty

The open-source CRM that fits your business without development, and whose bill does not climb with every hire.

n8n

The workflow engine that connects your tools, on your server, with no per-operation bill and AI only where it helps.

Astro

The technology behind our sites: ready-to-serve pages, fast and secure, that the whole team edits without a developer.

Matomo

Web analytics hosted by you: your visit figures, without handing your visitors’ data to a third party.

Odoo

The ERP that brings sales, invoicing, projects and accounting together, set up in three fixed-price tiers, from €15,000.

Dolibarr

The light open-source ERP: quotes, invoices, inventory and projects in one place, for teams of 5 to 50 people.

Claude

Anthropic’s AI that we deliver our projects with, and that we teach your teams to use safely.

Serveurs MCP

The standard that connects AI to your software: CRM, ERP, shop, documents. Less typing, less back and forth between screens.

Shopify

The ready-to-sell e-commerce platform, which we configure, enrich and grow without losing its search rankings.

“AI at the right cost” pack: pay for AI what it is worth

We first check in 20 minutes that it fits your need, and we tell you honestly if another one would do better.

See the offer Let’s talk
WhatsApp