About us
AI agents and open-weight models, built to measure.
CSIDIA S.r.l., Vignola (Modena, Italy). We build for companies a system that brings people, documents, case files, processes and data into a single ontology, that is a shared model, governed and connected by AI agents that work for people. For sensitive data we choose open-weight AI models, specialise them for our clients’ task and run them on our machines or theirs.
The company and the team
We sign as CSIDIA. Behind it is a team that follows every project from start to finish: the first meeting, the trial on your documents, the design of the system and its agents, the choice and specialisation of the model, the measurements, the work in service.
The AI prepares, a member of the team checks and signs what we deliver. There is one address to write to, and a person replies.
CSIDIA S.r.l. · Via Giacomo Leopardi 20, 41058 Vignola (MO), Italy · VAT IT04190980369 · engagement@csidia.com
What we believe
Four sentences that guide every job.
An open-weight model, trained on your documents and your work, does better on that work than the large general-purpose models, and runs where you decide.
The best model is the one you can use on your real data.
Your knowledge is your edge: it should not become training material for someone else’s AI.
The AI prepares, a person signs: agents too work within people’s permissions, with their confirmation.
Our machines
Where the models run, when they do not run at your site.
An NVIDIA GB300 server
The large model runs here, up to 40 requests at once.
A workstation with two NVIDIA RTX PRO 6000 Blackwell Max-Q
96 GB of memory each. This is where we train the image models and run the small models.
The large model
Qwen3.8-Flash-Next, open-weight: about 180 billion parameters in total, of which 6 billion are active for each token (a piece of a word).
Same model, same outcomes
The same model runs on the server and on the workstation: in our tests, with the same questions, the outcomes matched item by item in 96% of cases.
The rules we set ourselves
Written down, and they apply to us first.
- Clients’ technical manuals are read only by the AI running on our machines, never by a cloud model.
- For each project we keep a model register: what each model does, where it runs, its licence, the tests with their numbers and the reason for the choice.
- We write the site’s notes with the help of AI, from public sources: a second check verifies the facts and a person reads them before they go out.
- Agents work within people’s permissions: the actions that matter start only with a person’s confirmation, and every step is logged.
- Never the names of our clients, neither on the site’s pages nor in the notes.