IT EN
Book a meeting

What we do

One system for the company, governed by AI agents. And open-weight models, trained for your purpose.

We do two things that work together. We build a system that brings the company’s people, documents, case files, processes and data into a single ontology, and we have it governed and connected by AI agents that work for people. And we train open-weight models for your purpose: AI models you install on your own machines, instead of using them over the internet from a large vendor and paying per use, and that keep sensitive data in house.

The system

A single company ontology, and AI agents that govern it.

An ontology is a shared model of the company: every person, document, case file, process and piece of data has one name, one place and its links to the rest. Who handles which case file, which document is needed for which deadline, which process touches which data.

AI agents work on this model: they read mail and documents, keep track of case files and deadlines, prepare the work and carry out the permitted actions in the company’s programs, on behalf of people. People and agents find what they need without searching for it in many different programs.

Rules

Every agent has written rules: what it may do, what it may not, when it must ask.

Permissions

Every agent works within the boundaries and permissions of the person or organisation it works for.

Confirmations

The actions that matter start only with a person’s confirmation. The agent proposes, the person decides.

Log

Every step is written down: what the agent read, what it proposed, who decided.

Three agents at work

Three agents we built.

Three examples of the system, each for a different job. The pattern is the same: a shared model of the organisation, an agent that works for people, and their rules, permissions and confirmations.

For one person

A personal assistant that coordinates a system of agents

It works for one person on a workstation and is the centre of a system of agents: one agent per project, with the assistant coordinating them.

  • Mail, calendar and WhatsApp, handled in one place.
  • A work queue, which the assistant hands out to the agents.
  • The actions that matter start with the person’s click.
  • A log of every step.

For the clients of a firm

The agent of each client on a professional firm’s platform

On the firm’s platform, every person and every client organisation has its own agent.

  • It works within that person’s boundaries and permissions.
  • It carries out the platform’s actions, with the person’s confirmation.
  • It uses a frontier model for normal work.
  • It uses an in-house open-weight model for sensitive data.

For an organisation

An agent that runs an organisation’s day-to-day case files

It works inside the organisation’s platform.

  • Compliance duties, mail, deadlines and documents, held together.
  • It prepares the work and proposes changes.
  • It changes nothing on its own: a person decides.

Each agent uses the model best suited to the task, open-weight when the data cannot leave: in that case it runs on our infrastructure or yours. We always tell you which model reads what.

The models: our philosophy

Not the software: the model trained for your purpose.

Alongside the system, the model. Software can be rewritten and features can be copied. What grows a company is a model that knows its work, and knowledge that stays its own instead of leaving with every request to a closed model.

On that task, it does better

A general-purpose model has broad but generic knowledge. A model trained for your purpose knows your documents, your language, your rules. On that task it does better.

The knowledge stays yours

What the model learns from your data stays yours. Your knowledge is your edge over competitors: it should not become training material for someone else’s AI.

No per-use fees

Closed models are paid per use, and every request sends your documents outside. The model trained for you runs on our infrastructure or yours, and we always tell you where.

For every model, three steps

Choose, train, measure.

  1. We choose the model

    We test the open models that fit the purpose and keep a register of each one: what it does, where it runs, under which licence, how it did in the tests. The register for document reading holds 18 (as of 20/10/2026): 4 in use, the others under test, tested, discarded or retired.

  2. We train it on our infrastructure

    On our infrastructure (an NVIDIA GB300 server and a workstation with two NVIDIA RTX PRO 6000 Blackwell Max-Q): by fine-tuning an existing model, or with networks of our own trained from scratch, as for generating images of materials. Sometimes it is enough to put it to work inside a procedure built around your case.

  3. We measure it and put it to work

    Before handing it over we measure it on your cases, with numbers you can check. Then it runs on our infrastructure or yours, and a person decides.

The infrastructure

Two routes: our infrastructure or yours.

First route

On our infrastructure

We train the model and run it on our servers and workstations. To start straight away, with no hardware investment.

Second route

Infrastructure of your own, designed and managed by us

We design it around your workloads, build it, install it at your site and manage it: network, cooling, power, security, updates.

Our infrastructure, in detail

What we build on it

Four examples, for whoever has these problems.

Manufacturers, technical offices and anyone who writes manuals and technical files

Checking technical documents against standards

Every requirement of the standards that apply to you (for machinery: the Directive, Regulation (EU) 2023/1230, harmonised standards) searched for across the whole document. You receive the outcome requirement by requirement, with the citation; your expert signs.

How we do it

Technical offices, after-sales, spare parts: anyone with archives in PDF or on paper scans

Reading hard archives and documents

Manuals, drawings, tables and scans turned into searchable data, without losing the codes. You receive a structured file for each document, with the source of every piece.

How we do it

Anyone who makes or sells surfaces and materials recognised by sight

Custom-generated images of materials

One model for each of your material types, trained on your images, that generates new images at true scale and measures them against the real ones.

How we do it

Companies and public bodies with deadlines, case files and mail to follow every day

An agent on your company’s data and mail

Mail, deadlines, documents and rules in one place, with an AI agent that prepares the work and a person who approves. The agent works with the model best suited to the task, open-weight when the data cannot leave: in that case on our infrastructure or yours.

How we do it

Where we are careful

What we do not promise.

  • Not “better at everything”. On broad tasks, such as writing code or answering anything, large general-purpose models stay ahead. We work on one purpose at a time, and we measure it.
  • We tell you which model reads your data. When the data cannot leave, it is an open-weight model, on our infrastructure or yours: its name, licence and where it runs are written in the project’s model register.
  • Agents do not decide on their own. They prepare and propose; the actions that matter start only with a person’s confirmation.
  • Not every agent works with open-weight models. For everyday work they may use a frontier model; when the data cannot leave, an open-weight model on our infrastructure or yours. We tell you, agent by agent.
  • Open weights do not mean a free licence. Every model has its own licence: we check it and record it in the register before using the model.
  • We do not give legal opinions. We read the rules and build the checks; the legal assessment stays with your lawyer.