1. About this page
This page describes Gitogi Srl's use of artificial intelligence. It explains when you interact with an AI system, how your data is handled and which limitations to bear in mind.
The European framework is Regulation (EU) 2024/1689, the AI Act, which includes the transparency requirements in Article 50. Italian Law No. 132 of 23 September 2025 also applies to professional services in the circumstances described in section 4.
This English version is provided for convenience. If there is any discrepancy, the Italian text prevails, subject to mandatory law.
2. Gitogi AI, the website assistant
What it is
Gitogi AI is a generative AI assistant. When you use the chat, you interact with an automated system.
The assistant provides information about Gitogi's services and lets you request contact from Gitogi. Gitogi developed the system using third-party models, puts it into service under its own name and operates it on the website. For this chat, Gitogi is both the system's provider and its deployer, the AI Act's term for the organisation using it.
Consent and identification
Before your first message, we ask for specific consent to the chat and provide the privacy notice. You can withdraw consent in the chat or on the Your consents page. Withdrawal is recorded, deletes the cookie and prevents further use without fresh consent. Conversations already stored remain for the periods described below, subject to erasure obligations under the GDPR. You can also contact Gitogi by emailing info@gitogi.com.
The chat window always displays the notice “You are talking to an AI assistant, not a person” above the messages, linked to the dialogue through aria-describedby. The assistant's instructions also require it to identify itself as AI when asked and prohibit it from impersonating people; these instructions may not prevent every incorrect response.
Responses carry the HTTP headers X-AI-Generated, X-Content-Origin and Link with the ai-transparency relation. These provide information during transmission; they do not create a marking that stays with the content.
Each response ends with a machine-readable provenance statement, digitally signed by Gitogi Srl using the same key as for blog articles. It records that the response was generated by AI without human review. It identifies the system, Gitogi AI, its provider, Gitogi Srl, the model that answered and when. It also includes a cryptographic fingerprint (SHA-256) of the response text. The statement contains no conversation text, including your question, and no conversation or personal identifiers.
The statement stays attached to the response in the chat page's code. It is also included when you download the conversation using “Export conversation”. The statements appear at the end of the file, one line per response, together with the public key URL.
The verification service linked in section 3 lets you check whether a statement is authentic and whether a text exactly matches the signed response. The watermark limitation explained in that section also applies to the chat.
Consent to data processing and information about the assistant's artificial nature serve separate purposes. Our Privacy policy explains how chat data is processed.
Models used
| Order | Model | Processing service |
|---|---|---|
| 1 | Claude Sonnet, developed by Anthropic | Amazon Bedrock, using EU inference profiles |
| 2 | Mistral Large, developed by Mistral AI | Amazon Bedrock, using EU inference profiles, as the fallback model |
Model requests go exclusively through Amazon Bedrock. If neither model responds, the chat cannot provide an answer.
The EU location refers to model processing. For other services involved, including contact request emails, see the Privacy policy and list of providers and sub-processors.
Limitations
- The assistant may produce inaccurate, incomplete or outdated information.
- Its answers are for information only. Ask Gitogi for a proposal covering a project's prices, timescales and specifications: a chat response is not a contractual offer.
- Responses are generated automatically and may be shown to you without prior human review.
- The chat does not make automated decisions about you or assign you scores.
- Do not enter credentials, health data or other special categories of personal data, information about criminal convictions or offences, other people's personal data, banking or payment details, or trade secrets. Limit any other information to what your request requires.
These points explain the service's limitations and do not exclude any liability that Gitogi cannot lawfully exclude.
Based on the information and contact request functions described here, Gitogi AI does not appear to be a high-risk system. This assessment concerns the website chat and does not extend to solutions developed for clients.
The rules for using the chat are in the AI Acceptable Use Policy.
What happens when you request contact
The contact request function collects your name and email, together with any details about your company, telephone number, needs and requested service. It creates or updates a commercial contact record and sends an email to info@gitogi.com.
The email includes your request details and up to the last 12 messages in the conversation, each limited to 700 characters, readable in the email body. The telephone number and description of your needs collected through the tool are included in the email and stored in your commercial contact record. If you type a telephone number in a message, that text remains in the stored conversation.
Bear this email transmission in mind when deciding what information to share in the conversation.
Stored data and controls
For each conversation, we store the language, starting page, your messages and the assistant's responses. We do not associate an IP address or browser details with the conversation.
A separate AI technical log records character and token counts, the model used, processing stage, response times, tools called and a cryptographic fingerprint of the system instructions. It contains no message text and is kept for 5 years.
A daily process deletes conversations after 90 days. Conversations linked to a contact request are kept for as long as the associated commercial contact record is retained; that record is subject to deletion after 24 months of inactivity. Deletion from the website's records does not amount to deletion of copies already sent by email.
The chat uses filters against attempts to manipulate its instructions and applies usage limits: 10 messages per minute and 50 per day per pseudonymised IP address, plus 3 contact requests per hour. Pseudonymisation replaces the IP address with a fingerprint calculated using a secret key.
Under the Amazon Bedrock service terms, AWS does not retain content submitted to the models or use it to train them. This concerns AWS: Gitogi retains conversations and logs for the periods described above.
You can use Your data to access data linked to your email. Conversations without a contact request cannot be linked to your identity through this feature.
3. Website content
Blog articles are prepared with AI assistance in Gitogi's editorial console. A person must approve each article and choose its publication date and time before it is published. Every article prepared with AI assistance displays “This article was generated with AI assistance and may contain inaccuracies” below its title, linking to this page. Gitogi retains editorial responsibility for what it publishes.
Gitogi developed the system used to prepare these articles and marks its output as required by Article 50(2). Alongside the visible label, each article page includes a provenance statement in JSON, a machine-readable format. Gitogi Srl digitally signs the statement using the Ed25519 algorithm. The page also includes the ai-generated and ai-provenance meta tags.
The statement records the article's URL, language, title, publication date and update date. It states that AI helped prepare the article and that a person approved it before publication. It names Gitogi Srl as editorially responsible and identifies Gitogi's editorial workshop as the system used. It specifies Anthropic's Claude for writing and OpenAI's Codex for text review. A cryptographic fingerprint (SHA-256) covers the title and article text.
You can check the signature using Gitogi's public key. The same address provides the rules for reconstructing the signed data. The signature proves who issued the statement and that it has not been altered. It does not establish that the content is accurate.
The verification service checks whether a provenance statement is authentic. Changing even one field causes the statement to fail verification. You can also submit a passage of at least 25 words to check whether it comes from an article published on the site. The service recognises copied text despite changes in capitalisation, punctuation or spacing. It does not recognise paraphrases.
One limitation applies to both articles and chat responses. A watermark is a statistical mark embedded in the words themselves. It stays in the text when you copy it without the provenance statement. Only the provider of the model generating the words can apply it. Gitogi does not add its own watermark. Where the model provider applies one, it remains in the text.
The website's photographs (people at work, vehicles, yards, loading bays, warehouses and offices), their line-drawn versions and the cover photos for blog articles are generated using AI models from descriptions written by Gitogi. They do not depict real people, Gitogi's clients or collaborators, or Gitogi locations: the people shown do not exist and have no names. On the “About us” page, the photograph of a group at work also carries the label “AI-generated image” on the image itself, because the group could be mistaken for Gitogi's team. For images on other pages, this paragraph provides that disclosure.
4. AI in our work
Tools used within Gitogi's activities include:
| Tool or service | Scope and relevant terms |
|---|---|
| ChatGPT Business and Codex | Work support tools under business terms that exclude the use of data for training |
| Perplexity | Research using public sources, without entering personal data |
| Claude Max | Plan governed by Anthropic's consumer terms, under which Anthropic acts as a controller, without a DPA; it must not receive clients' personal data and is not a sub-processor |
| Amazon Bedrock | Model processing for Gitogi AI, as described in section 2 |
A tool's inclusion in this table does not mean that it receives every client's data. Terms excluding training do not, on their own, authorise the disclosure of personal or confidential information.
Where Gitogi processes personal data on a client's behalf, the use of AI services must follow the client's documented instructions, data processing agreements and any required sub-processor authorisations. Client personal data must not be entered into tools governed by individual-use terms that lack the safeguards required for that processing.
Where a service falls within the scope of intellectual professional services, Article 13 of Italian Law No. 132/2025 requires AI to serve a supporting role, with intellectual work remaining predominant, and the client to receive clear, simple and comprehensive information about the systems used. For those services, this general description must be supplemented with information specific to the engagement.
If you have restrictions on AI use or particular data, tell us at info@gitogi.com so they can be addressed in the contract. Using AI does not remove Gitogi's responsibility for its services.
5. AI solutions for clients
Gitogi integrates AI into solutions for businesses and professionals. Depending on the project, software may run on the client's accounts or Gitogi's accounts, and external collaborators or partner companies may be involved.
AI Act roles depend on the activities actually performed, the system's intended purpose and how it is put into service. Account ownership alone does not determine those roles.
Each project needs to establish:
- its purpose, users and intended uses;
- the system's classification and applicable obligations;
- the roles of Gitogi, the client and other parties;
- suppliers, data processed, processing regions and any international transfers;
- human oversight, user information and required documentation;
- responsibility for technical and organisational tasks.
Practices prohibited by the AI Act remain prohibited even if a client requests them. For a high-risk system, an agreement allocating tasks does not replace legal requirements.
The relevant contractual documents are the Terms and conditions and, where Gitogi processes personal data on the client's behalf, the Data processing agreement.
6. Reporting concerns
If a Gitogi AI response seems incorrect, inappropriate or risky, email info@gitogi.com. Describe the issue and, if helpful, include the date and a short extract from the response, avoiding unnecessary personal data.
You can use the same address for questions about Gitogi's use of AI.