How Itria uses artificial intelligence.
This page explains how we integrate AI into the systems we build, which technologies we use, how we protect clients' data, and which documents are available. Last updated: September 2026.
AI doesn't replace human judgement. The systems we build automate what is repeatable and make suggestions; the decisions that matter about a customer, an offer or a communication that commits the business stay with a person. It is a design constraint, not a promise: no Itria system sends a real customer a message that hasn't been approved, either beforehand on the content or at that moment on the individual message. Which of the two is decided by the owner, channel by channel, and the setting can be changed whenever they like.
The data stays our clients'. We don't use a client's data to train models, nor do we share it with third parties who aren't under contract. Where we use a third-party model (Anthropic Claude), the relevant data processing agreement excludes the use of the data for training.
Secrets never in the code. Credentials and API keys live in environment variables, never in versioned files, prompts or logs.
| Technology | Role | How it is governed |
|---|---|---|
| Claude API Anthropic |
Reasoning on content: reading a website, multilingual drafts, qualifying requests. | Data Processing Agreement with Anthropic; data excluded from training. |
| n8n | Orchestration of the automated processes: the layer that connects requests, reminders, reporting. | Self-hosted on Itria infrastructure, not a third-party SaaS service. |
| Google Sheets | The state of the operational data (contacts, appointments, logs). The client can read it at any time. | Standard Google Workspace DPA. |
| Next.js · Vercel the Plancia |
The control panel the client sees, and the APIs that feed it. | Standard Vercel DPA, selectable hosting region. |
| Hetzner EU server |
The infrastructure that hosts n8n. | Servers in the European Union, Hetzner DPA. |
| WhatsApp Business API | The channel on which the systems that use it send reminders, confirmations and replies to end customers, always behind the human approval gate described above. | Meta platform, Business API terms, data minimised to what the conversation needs. |
| Google Drive | Destination of the nightly backup copy of the client's state store (contacts, bookings, logs), separate from the backup of the server infrastructure. | Standard Google Workspace DPA, same scope as Google Sheets. |
| Cheerio · Puppeteer | The engine that reads a real website line by line for the Quadro: structure, content, technical signals. | Runs on Itria infrastructure, reads only public pages indicated by the user. |
| Resend | Sending periodic reports and transactional communications (confirmations, notifications) to the client. | Provider's standard DPA, tracked sends. |
What happens if a system stops at night, or if some data is lost?
An internal system that watches over every automated cycle: if a cycle doesn't report its completion within the expected window, or completes with an anomalous outcome, the alert reaches the operator straight away. We don't wait for the client to notice.
External, independent monitoring of the availability of the Plancia and the public APIs, active 24 hours a day.
A dated nightly copy of the operational data, with rotating retention, independent of the backup of the server infrastructure.
A nightly copy of the server that hosts the orchestration.
Anti-bot verification on public forms (forms, self-service tools), with a limit on requests per address.
Every processing activity has a legal basis defined, in writing, in the agreement with the client. We don't process a piece of data without knowing exactly on what grounds we do so: performance of the contract, legitimate interest or consent. It is documented before the system goes live, not after.
Sensitive data doesn't enter the AI systems. Where a process handles information about end customers (names, contacts, order history), the data stays minimised and pseudonymised to the level the model needs to do its work. No health, financial or special-category information (Art. 9 GDPR) is ever passed to a third-party model.
A local architecture is always an option. For clients with stricter compliance constraints, the steps that today use the Claude API can be replaced with a model running on local or dedicated infrastructure, at the cost of less reasoning capability on language. An explicit technical choice, discussed case by case, never a silent default configuration.
The technical measures include: dedicated authentication, separation of data by client, secrets kept out of the code, encryption in transit, a log of executions. In the event of a data breach, the client is informed without undue delay.
Every provider that comes into contact with a piece of data, what it does with it, and on what grounds. The same providers as in section 2, reorganised for compliance.
| Provider | What it processes | Contractual basis |
|---|---|---|
| Anthropic (Claude API) | Content passed for reading in order to generate drafts, classifications, suggested replies. Never used to train models. | Data Processing Agreement, data excluded from training. |
| Google (Workspace) | Current operational data (Sheets: contacts, appointments, logs) and the nightly backup copy (Drive). | Standard Google Workspace DPA. |
| Vercel | The data shown in the control panel (the Plancia) and the requests to the APIs that feed it. | Standard Vercel DPA, selectable hosting region. |
| Hetzner | No content data: it only hosts the server infrastructure that runs n8n. | Servers in the European Union, Hetzner DPA. |
| Meta (WhatsApp Business API) | Messages to end customers (reminders, confirmations), always behind the human approval gate. | Business API terms, data minimised to the conversation. |
| Resend | Email addresses for sending periodic reports and transactional communications. | Provider's standard DPA, tracked sends. |
How long we keep a piece of data, and what happens afterwards. The same rules for every client, written before a system goes live.
| Data | Retention | What happens afterwards |
|---|---|---|
| Data from a concluded client relationship | 30 days | Deleted, or returned at the client's request. |
| Logs of the Motori executions | 6 months | Deleted automatically on rotation. |
| Leads collected from the website (Diagnostico, Quadro, contact) | 24 months from the last contact | Deleted if no commercial relationship arises, unless withdrawn earlier. |
| Backups (client data and infrastructure) | 30 days on rotation | The oldest copy is overwritten by the next one. |
Our role. Itria is at the same time a deployer of third-party models and the provider of the systems it builds and sells under its own name. The transparency obligations of the European regulation on artificial intelligence therefore apply to Itria, not only to those who produce the models.
The risk level. The systems we build fall into the minimal-risk category and the transparency obligations. They don't deal with access to employment, education, credit, essential services, biometric data or the administration of justice: they are therefore not high-risk systems under Annex III.
What it means in practice. Anyone who interacts with an automated Itria system is informed that they are interacting with an automated system. No generated content bearing a client's name is sent without explicit human approval. No data belonging to the special categories enters the systems.
The transparency obligations under Art. 50 apply: those that directly concern the systems we build.
The AI literacy obligation (Art. 4) applies: whoever operates the systems must have adequate competence to understand and supervise them.
The obligations on high-risk systems were postponed to these dates (Annex III and Annex I respectively) by the Digital Omnibus, in force since 27 July 2026.
These dates concern us, as we build the systems and put them into service. What changes instead for a business that simply uses artificial intelligence lies in a distinction of role that reduces the obligations to four things: the AI Act for a small business.
The complete map of the AI systems we use, with purpose, model employed and level of human oversight, is available on request and is attached to the contract for clients.
To request a document: write to info@itria.io stating which one you need. We reply within 5 working days.