Legal

Privacy policy

Information on how personal data is processed on this website.

In short: this website uses no cookies and no third-party tracking or analytics tools. That is why there is no cookie banner. Only from your first interaction with an AI feature does your browser remember your path in this tab - visible and switchable off under Your focus.

1. Controller

agent-sii.com is a label of SII Germany. The controller for data processing on this website is:

SII Deutschland GmbH, Theodor-Yorck-Str. 6, 21079 Hamburg, Germany, phone +49 821 90722-0, info@sii-germany.com. The company is currently being renamed SII Germany GmbH.

Data protection officer: Kai Dömming, datenschutzkoordinator@sii-germany.com

2. Hosting and delivery

The website runs on Amazon Web Services (AWS): files are stored in Frankfurt am Main and delivered via the Amazon CloudFront content delivery network, using only locations in Europe, North America and Israel. When you visit, AWS processes technically necessary data such as IP address, time and page requested in order to deliver the site and protect it against attacks. We have not switched on CloudFront access logs, so we do not store any.

Our server functions (AI features, contact form, audience measurement) write technical logs without conversation or form content. These are stored in Frankfurt and deleted automatically after 30 days.

Processing by AWS is governed by the AWS data protection terms (Data Processing Addendum under Art. 28 GDPR), which are part of the AWS Service Terms. Where data may reach the USA, AWS is certified under the EU-U.S. Data Privacy Framework. Legal basis: Art. 6(1)(f) GDPR (secure and stable provision of the website).

A list of all sub-processors for this website is available on request.

3. Fonts

All fonts are loaded from our own server. No connection is made to Google or other font providers.

4. Contact form and email

When you write to us via the contact form, we process the details you provide: name, email address, optionally company, industry, topics, timeframe and your message. The form is sent as an email to our team via Amazon Simple Email Service (Frankfurt). The website itself does not store your request. So that you have little to type, the form shows as building blocks what you viewed, asked and answered in this browser tab. These blocks are kept only in your browser’s session storage and deleted when you close the tab. You can take out any block; only what is still in the form when you send it is included.

If you email us - for example at an address from the imprint or with a security report to informationssicherheit@sii-germany.com - we process your email address and the content of your message. None of this goes through the website itself.

We use the details solely to answer your request and delete them once they are no longer needed for that purpose and no retention obligations apply. Requests go to mailboxes at SII Germany; retention and deletion follow SII Germany’s privacy notices on sii-germany.com. Legal basis: Art. 6(1)(b) GDPR (pre-contractual steps) or Art. 6(1)(f) GDPR (answering general enquiries and handling security reports).

Applications: agent-sii.com does not hire directly, and this website does not accept applications. Please send applications by email to the People & Culture team at SII Germany (people-and-culture@sii-germany.com) or via the SII Group job board. SII Germany’s privacy notices on sii-germany.com apply to applications.

5. AI assistant

On this website you can ask questions to an AI assistant. To answer them, your input is sent via the service provider Requesty (Requesty Ltd, England; the EU has an adequacy decision for the United Kingdom; covered by a data processing agreement under Art. 28 GDPR, EU routing, without logging and with zero data retention) to language models in data centres in the EU. Requesty uses providers of language models with data centres in the EU as sub-processors; which ones are permitted is set by our organisation’s allow list. The providers do not store the data beyond processing (zero data retention), do not use it for training and do not pass it on.

So the assistant can follow the conversation, your browser stores a random session ID in session storage (sessionStorage) from your first interaction. It is deleted when you close the tab and serves only this purpose.

Protection against abuse: for the AI features your browser gets a short-lived, signed session token (valid for 30 minutes; until your first interaction only in the page’s memory, then in session storage). With it we limit the number of requests per session and minute. For this our server only counts a hashed session number or - when issuing new tokens - a hashed IP address with minute and count, and deletes these counters after 2 minutes. The token contains no information about you and is not linked to conversation content. Legal basis: Art. 6(1)(f) GDPR (legitimate interest in a service protected against abuse and uncontrolled costs). More in Public AI without a cost trap.

For quality assurance, we store the conversation for 30 days, encrypted, in Frankfurt. Before storage, recognisable email addresses, phone numbers and other recognisable contact details are filtered out automatically; names or other details you write freely into a question may remain; the session is not linked to your IP address. The data is then deleted automatically. Legal basis: Art. 6(1)(f) GDPR (legitimate interest in a reliable and secure assistant).

The conversation also includes which content you open from the assistant. The suggestions (“Keep exploring”) are chosen by a model based on your questions and the content opened so far. For every suggestion in the assistant, a language model also writes a short reason; for this, your latest questions and the titles of the suggestions are sent to the model, and none of it is stored. For this we use, via Requesty, a decision model in the EU that only returns a ranking.

Before an answer from the assistant appears, the same decision model checks it: your question and the answer are sent to the model, which only returns probabilities for fixed categories (e.g. confidential details, prices or commitments, unsupported figures, tone). The answer is then shown, flagged or held back. This also applies to sections created after a freely written answer in an article. We store the result of the check (status, categories, scores) with the conversation for 30 days; for article sections only if they were flagged or held back, then together with your answer and the section, likewise cleaned of recognisable contact details. If the checking model cannot be reached, the answer still appears and is stored as “unchecked”. No further analysis takes place. There is no automated decision within the meaning of Art. 22 GDPR that produces legal effects for you or similarly significantly affects you. Legal basis as above.

In addition, we anonymously count which topics are asked about: only terms from our own content (e.g. “TISAX”) with day and number, without the wording. Once a day we also derive anonymous path patterns from the conversations (e.g. “question → reference → article → contact”) and store only the pattern and its count.

On recommendation pages and in the assistant, a model writes a short take on the page in the background and estimates whether it is more about time, quality or cost. For this, the page title and the page you came from (in the assistant: your questions and opened content) are sent to the model; none of it is stored. So you can keep asking on any page, from your first interaction your browser also keeps the conversation with the assistant (your questions, its answers including the content shown and which of it you opened) together with your page visits (at most the last 40 steps) in session storage and sends it with your next question. Your browser stores three settings permanently (localStorage), but only if you change them yourself: “No more tips”, “show instantly” (without the AI’s short pause) and “Think along” switched off. They only contain that one switch position and can be reset on the Your focus page.

Growing page, magazine editions and “Your focus”

On the home page and below articles and references, a language model writes on from your answers. For this, your answers on this path (claim, stance, chosen questions, your own words), the sections created so far, your focus (time, quality or cost) and the titles of the pages you read most recently are sent to the model; none of it is stored - unless a freely written section is flagged or held back by the check (see above). Frequent paths are kept in the server’s memory for up to 6 hours, with no link to a person. Legal basis: Art. 6(1)(f) GDPR (legitimate interest in preparing content to fit at the visitor’s request).

Magazine editions of articles and references are created once by a language model from the approved text. They are stored permanently and delivered the same to everyone; no information about visitors goes into them.

So the content fits you, from your first interaction with the AI features - typing or clicking a suggestion in the assistant, answering a claim or question, confirming a key statement - your browser remembers in this tab (session storage) which pages you read and for how long, how you got there, your answers and the takes on individual pages. From this, your browser itself calculates what matters to you more (time, quality or cost). Under Your focus you see this picture in full; “Forget everything” deletes it, and it is deleted anyway when you close the tab. When you open that page, a language model puts your visit into perspective in a few sentences: for this, the titles, reading time and order of the pages you read, your answers, your questions to the assistant and your focus are sent to the model. The result is checked like an answer of the assistant and only stored for 30 days if the check raises a flag. The picture is created only in your browser; we do not store it. Before your first interaction your browser remembers none of this; the AI features then only know the current page and the page you came from. With the “Think along” switch on the Your focus page you can turn it off completely: your browser then does not remember your path, no focus is calculated and none of it goes to a model.

Please do not enter sensitive data in the chat. The assistant collects contact details only via the contact form.

Access for AI agents (MCP and A2A)

Besides the website we offer two interfaces for software: an MCP server (/mcp) through which AI assistants retrieve our published content, and an A2A endpoint (/a2a) through which other AI agents ask our assistant questions. Such requests are usually made by a program on behalf of a person, so they may contain information that person gave the program.

  • What we process: the request (for MCP the tool and its arguments, such as search terms; for A2A the question), our answer, technical data such as time, duration and error code, and the name the calling program reports for itself. We use the IP address only briefly in memory to limit the number of requests (hashed, with a key that changes daily); it is not stored.
  • Language model: A2A questions are answered by a language model under the same conditions as the AI assistant (section 5: Requesty with EU routing, no logging, zero data retention, data processing agreement, EU models only). Before sending, the decision model checks the answer. The MCP server uses no language model.
  • 30 days: we store requests, answers and check results encrypted in Frankfurt, cleaned of recognisable email addresses, phone numbers and other contact details and without IP address - for quality assurance and protection against misuse. They are then deleted automatically. Technical logs without content are also deleted after 30 days.
  • Permanently, anonymous: counters per day - number of requests, type of calling program as a category (for example “AI assistant” or “script”), topics as terms from our own content, tools used, outcome of the request and tokens used. No wording, no names and no link to individual requests.
  • No contact requests: these interfaces never create a request to us; the interested person contacts us themselves via the contact form. Answers from the A2A endpoint are marked as AI-generated.

Legal basis: Art. 6(1)(f) GDPR (legitimate interest in reliable, secure interfaces protected against misuse and in developing our offering).

6. Audience measurement without cookies

When you open a page, your browser only sends its path to our server - for content you open in the AI assistant, with the note “via the assistant”. It also sends which page of this website you came from; if you came from elsewhere, the server only evaluates the type of origin (direct, search engine, AI chat, social media, other website) and does not store the address. From this we store nothing but anonymous counts (views per content item and day, transitions from page to page per day and - unlinked to each other - which claim on the home page was shown and clicked and how often “Your focus” is opened and cleared), without IP address, device identifier or cookies. No link to individual visitors is possible. Legal basis for the brief processing during transmission: Art. 6(1)(f) GDPR.

7. Cookies and local storage

This website uses no cookies and no third-party analytics, tracking or marketing tools. Only what a feature you use yourself needs is stored on your device:

  • From your first interaction with an AI feature (typing or clicking a suggestion in the assistant, answering a claim or question, confirming a key statement), in this tab’s session storage: a marker that you use the features, session ID and session token, the conversation, your path through the website, your answers, the takes on individual pages and which tips have already been shown (see section 5). Before that, none of this is stored.
  • When you fill in the contact form: your draft in session storage, so it is not lost when you change pages.
  • Only if you change them yourself: three settings in localStorage (section 5).

This storage is strictly necessary for us to provide the feature you explicitly use and is therefore permitted without consent under Section 25(2) no. 2 TDDDG (German Telecommunications Digital Services Data Protection Act). Everything in session storage is deleted when you close the tab. “Forget everything” under Your focus deletes the conversation, path and answers immediately; thinking along stays on. You can switch it off completely with the “Think along” switch on the same page. Audience measurement (section 6) stores nothing on your device.

Links to LinkedIn, GitHub or to external sources in our articles are plain links. A connection to the respective provider is only established when you click them; their privacy notices apply there.

9. Your rights

You have the right to access, rectification, erasure, restriction of processing and data portability. You can also lodge a complaint with a data protection supervisory authority, for example the authority responsible for us, the Hamburg Commissioner for Data Protection and Freedom of Information (HmbBfDI), Ludwig-Erhard-Straße 22, 20459 Hamburg, Germany.

Right to object

Where we process your data on the basis of Art. 6(1)(f) GDPR (legitimate interest), you may object to this processing at any time on grounds relating to your particular situation (Art. 21 GDPR). We will then no longer process your data unless we can demonstrate compelling legitimate grounds that override your interests, or the processing serves the establishment, exercise or defence of legal claims. An informal objection is enough, ideally by email to datenschutzkoordinator@sii-germany.com. For storage in your browser you can use the “Think along” switch under Your focus.

Last updated: 10 November 2026.