Service 03 / AI servers
Optimised AI workstations, on your premises.
We maximise the potential of your infrastructure or deploy AI directly on your premises.
-
01
Process your most confidential data
The AI runs on an isolated workstation on your premises: your data never leaves it.
-
02
Costs divided by 10
Compared with Claude or ChatGPT, for comparable use.
-
03
Total agility
To fit every business line: models, tools and interfaces tuned team by team.
Use cases
Business uses, on your own data.
Document search, business software, code: AI workstations work alongside your teams every day.
-
01
Document search
Drawings, forms and scanned archives are read, labelled and made searchable: every answer cites its source.
-
02
Business software
AI plugs into your software, such as CAD: it checks a part against the applicable code and flags deviations.
-
03
Code writing
A coding agent writes, tests and documents alongside your developers, without a single line leaving your premises.
Turnkey AI workstation
The right hardware for every use.
Rental or purchase, installation and maintenance of a dedicated, isolated and secure AI workstation on your premises. You retain control of your data.
-
AMD
GMKtec EVO-X2
64 GB of unified memory
-
AMD
GMKtec EVO-X2
128 GB of unified memory
-
AMD
Minisforum MS-S1 MAX
192 GB of unified memory
-
NVIDIA
DGX Spark
ASUS Ascent GX10
128 GB of unified memory
-
NVIDIA
RTX PRO 6000
Blackwell graphics card
96 GB of GDDR7 memory
And other configurations, sized to your needs.
Total control
Total control of your AI workstations.
Models, harness, security: every building block is chosen, tuned and maintained for your business lines.
On your premises
-
04
Business uses
-
03
Harness
-
02
Models
-
01
AI workstation
Security
An isolated and secure workstation. Tailored cybersecurity risk analysis (EBIOS RM).
Optimising your servers
Autoresearch speeds up your existing hardware.
Our MLOps pipeline (quantisation, batching…) gets the most out of the best AI models on your existing hardware.
Our autoresearch agents run dozens of experiments on it autonomously: every optimisation is measured, and only those that speed up the model without degrading its answers are kept.
Faster AI keeps pace with the business: speed is what drives adoption.
- ×1.52
- faster in 24 h
- 86.5
- tokens per second, up from 57.1
- 11 of 28
- experiments kept
- best configuration
- kept
- discarded
- failed to start
- full stack measured
Autoresearch on Qwen 3.8 Flash, two NVIDIA DGX Spark, 24 h unattended.
Case studies
Orano
Nearly ten AI workstations deployed
NVIDIA DGX Spark and AMD workstations, installed and optimised on the client’s premises.
Deployment environment
- Secure on-premises AI workstations
Public sector organisation
5 times as many cases processed
Optimisation of a quantised model for technical and legal drafting tasks. Delivered in 8 weeks.
Deployment environment
- Installation and rental of a secure on-premises server
Other services