How it works

Something’s cooking.
The chef is at work.

Behind a well-designed artificial intelligence, there is a kitchen running smoothly. Let us introduce you to the brigade, station by station, and show you, very concretely, how it all works.

The brigade, station by station

Four stations,
one kitchen.

Why talk about a kitchen? Artificial intelligence remains abstract as long as you cannot picture it. So we chose a framework everyone already knows: a restaurant, with its brigade and its organisation. Each kitchen station corresponds to a real technical component of your AI, and at every step, we give you its exact name, the one you may have already encountered without connecting it to anything concrete. And if our chef wears four stars, it is not entirely a coincidence: he needed one per pillar.

The Chef

The model

The Head Chef is the artificial intelligence model itself, the one that thinks, composes, proposes. You choose yours: we do not impose a fixed menu.

In plain terms: an open-weight LLM (Large Language Model) such as Mistral Large, Llama 3.3, DeepSeek, or equivalent, quantised (GGUF, AWQ) to run locally without wasting GPU memory. You own the weights: no dependency on a remote API, no call leaving your network.

The Stoves

The inference server

The stoves are the computing power that keeps the chef working. The more powerful they are, the faster he works, and on several dishes at once.

In plain terms: one or more GPU cards (Nvidia RTX 4090/5000 Ada, A6000, L40S), driven by an inference engine such as vLLM or Ollama, with batch processing and context management for multiple simultaneous users, measured in tokens per second.

The Dining Room and the Menu

The interface

The dining room is where you place your order: the same kind of interface used to query ChatGPT, Gemini or Claude, except here, everything stays at your premises. Each user finds their conversation history and the tool they need.

In plain terms: a web interface such as Open WebUI, connected to the model via an OpenAI-compatible API, with per-user access control, conversation history, and integration with your internal tools (Office, email, business applications).

The Filing Cabinet

The RAG

The filing cabinet is the house memory: your documents, your folders, your accumulated knowledge, which the chef consults before answering, so he never improvises on what he does not know.

In plain terms: a RAG (Retrieval-Augmented Generation) pipeline: your documents are chunked, converted into vectors (embeddings) via a model such as BGE-M3, indexed in a vector database (Qdrant), then retrieved by semantic similarity for each question asked.

The kitchen grows with demand

More diners?
We add more stoves.

Demand rarely jumps overnight: more colleagues start using it, or those who already do use it more, as the tool becomes indispensable. In both cases, the answer is the same.

We add more stoves. Nothing else changes.

An additional GPU card, a second server if volume demands it: we add computing power, full stop. No migration, no software to reconfigure, no contract renegotiation, no change of habit for your teams. The kitchen gets bigger; the recipe stays the same.

+1stove, at any time
What keeps the kitchen running, behind the scenes

Three areas of expertise,
one single project.

Running this kitchen requires three distinct skills. We have not mastered all of them for the same length of time, and we would rather tell you upfront.

30 years

Hardware

Servers, physical networks, professional hardware: a trade we have practised since F6 was founded, through our sister company FSYS Informatique.

30 years

Secure networks

Segmentation, architecture, traffic management: the historical core of our managed services for demanding clients.

10 months

Artificial intelligence itself

Here, nobody in the world has thirty years of experience, but we have been working intensively on the subject for ten months now, and we are starting to deploy our first architectures today. We took the time to build our own approach, founded on a strict logic of ownership and independence, rather than reproducing what others do. We are moving forward with a motivated in-house team and high-quality consultants, and we learn fast.

Want to see the kitchen up close?

A first conversation is all it takes to identify, in your company, what belongs to the chef, the stoves, or the filing cabinet.

Get in touch Review our position

Unfamiliar term? Check the glossary →