Local, sovereign AI
Your contracts and customer files go into public AI tools.
We install AI at your premises, on your machine. At acceptance we cut the internet: the assistant still answers.
The problem
- 40 % of French small businesses use AI, 19 % pay for it (France Num 2026).
- Per‑user prices that change without notice.
- Nobody knows what the tool does with the data.
What you check at acceptance
- Server cut from the internet: the assistant answers.
- One access per employee, with a request log.
- Dashboard: usage, GPU memory, response time.
- Your tools connect to the OpenAI‑compatible API.
Each point is checked with you on delivery day.
Included
- Drivers, Ollama or llama.cpp engine, model choice and tuning.
- OpenAI‑compatible API, accounts and log.
- Team training.
- Monthly care: driver, engine and model updates.
What sets us apart
- Fixed cost: no per‑user price.
- You own the machine.
- Setup priced below the low end of published market prices.
Proof
Our models run on GPU behind an OpenAI‑compatible proxy that tracks tokens, GPU memory and model switching.
We write CUDA code: our HEVC video decoder runs on GPU cores.
Questions
Is a local model as good as an online service?
To search, summarise, sort and draft on your documents, a recent local model is enough in most cases. We check it on your examples before any purchase.
Do we need a big server?
Not always. We size the machine by number of users and tasks.
Ready to start?
Describe your context. You get a price and a date.