Back to search

Privacy notice

hu·en·de

This privacy notice describes how the dontes.app web application (the “Service”) processes the personal data of its users. The Service is an AI-based legal search engine that, based on the question entered by the user, searches the decisions of the Hungarian courts and the Constitutional Court, the Court of Justice of the European Union (CJEU), the EFTA Court and the Hungarian Public Procurement Arbitration Board (Közbeszerzési Döntőbizottság, KDB) and the European Court of Human Rights (ECtHR; public HUDOC database). Every processing activity that takes place while the Service is used is listed below. The Service does not use tracking (analytics or advertising) cookies or any third-party web analytics service.

Data controller: Dániel G. Szabó

Account creation and sign-in are provided through the Clerk authentication system; the registration data (e.g. name, e-mail address) is primarily processed by Clerk under its own privacy policy (clerk.com/legal/privacy). The data controller accesses this data through the Clerk interface for the purpose of identifying the user.

Answers are written by OpenAI's language model

The answer to each question, and the translation of quotations into the language of the search, are produced by OpenAI's language model through OpenRouter. If that model fails to respond to a question after two attempts, the answer is written exceptionally, as a fallback, by Anthropic's Claude model. Both providers operate in the United States, and both appear in the list of data processors below.

Between 13 and 23 September 2026 the Service compared the two models: which one answered was decided at random for each question. The comparison has been concluded. A rating of 1–4 may still be given voluntarily next to the answer.

Processing activities

1. User account and sign-in
Data processedThe user identifier provided by Clerk; the registration data (name, e-mail) is processed by Clerk.
Legal basisconsent of the data subject [Art. 6(1)(a) GDPR]
Retentionfor the lifetime of the user account, or until the account is deleted
2. Processing and answering the search question (the core function of the Service)
Data processedThe question entered by the user (free text), the user identifier, the time of the question.
Legal basisconsent of the data subject [Art. 6(1)(a) GDPR]
Retentionfor the time needed to produce the answer; afterwards in the saved history (see point 3)

Questions and answers are not encrypted; the data controller and the data processors can see the questions and answers.

3. Saving past questions and answers (history)
Data processedThe text of the question, the answer given to it, the decisions cited, the time, the user identifier.
Legal basisconsent of the data subject [Art. 6(1)(a) GDPR]
Retentionthe most recent 200 items per user; the user can delete items individually in the interface, and they are deleted together with the account
4. Usage statistics, cost tracking and management of the weekly question allowance
Data processedOne log entry is made for every question: the user identifier, the time, the courts searched, the settings chosen (language of the answer, the quotation–analysis balance, the values of the judge and legislation filters), the length of the question in characters, the number of decisions found per court, the device type (desktop, mobile or tablet — derived from the browser identifier, which itself is not stored here), the name of the language model and provider that wrote the answer, the amount of text processed and the cost, the length of the answer, the result of the quotation check and the processing time. For the weekly question allowance we also record its use and the fact of any “Ask for more” requests. The text of the question is not stored for this purpose.When a new account is registered, we also record where the user came from: only the domain of the referring page (e.g. google.com), the fact of arriving from the public decision pages of dontes.app, or, if the link carries one, its source tag (utm_source). The full address and any search terms are not stored, and nothing is stored on your device for this (no cookie or local storage): the tag travels in the address of the page shown after sign-in.
Legal basislegitimate interest of the data controller [Art. 6(1)(f) GDPR] — operating the Service, tracking costs, enforcing the weekly question allowance, detecting faults and learning where people discover the Service
Retentionfor the lifetime of the user account; after the account is deleted, the rows are kept without the user identifier and the filter values, in anonymised form (cost records); the registration source is deleted together with the account
5. Checking answer accuracy and improving the Service
Data processedQuestions submitted by users and the answers generated in response, together with the user identifier and timestamp.
Legal basislegitimate interest of the data controller [Art. 6(1)(f) GDPR] — improving the accuracy and quality of the Service
Retentionfor the lifetime of the user account, or until the account is deleted
6. Comparison of language models (13–23 September 2026, concluded)
Data processedThe question and the decision extracts retrieved for it. During the comparison, which model answered a given question was decided at random; the question was sent to the model without the user identifier. For the comparison we stored characteristics of the answer (length, amount of text processed, the name of the model) — not the text of the question.
Legal basislegitimate interest of the data controller [Art. 6(1)(f) GDPR] — establishing which language model produces the more accurate answer, and keeping the Service at a sustainable cost
Retentionthe comparison was concluded on 23 September 2026; results are retained in aggregated form
7. User rating of the answer (1–4)
Data processedThe rating given, the identifier of the question, the user identifier and the timestamp. Giving a rating is voluntary; the Service is fully usable without it.
Legal basisconsent of the data subject [Art. 6(1)(a) GDPR], given by submitting the rating — measuring and improving the quality of answers
Retentionfor the lifetime of the user account, or until the account is deleted
8. Measuring how answers are used (signals recorded automatically during use)
Data processedWhile the Service is being used, it records two signals without any separate action by the user:
  • Opening a decision: when the user opens the full text of a search result or cited decision — the user identifier, the time, the court and the identifier of the decision opened (e.g. its case number).
  • Interrupting the answer: when the user breaks the connection before the answer is complete (for example by closing or leaving the page) — the user identifier, the time, the time elapsed since the question was submitted, how long the user saw the answer being written, the length of the text received until then, how long no new text had arrived, the device type, the language of the answer, the courts searched and the name of the language model writing the answer.
Neither the text of the question nor any other text typed by the user is included. The Service does not record scrolling within the page, mouse movements, or how long the user reads the completed answer.
Legal basislegitimate interest of the data controller [Art. 6(1)(f) GDPR] — measuring whether the answers are actually useful (whether users open the cited decision, whether they wait for the answer) and detecting technical faults (for example a stalled answer). The signals contain only identifiers and timing data, no text; the user may object to this processing (see Your rights).
Retentionfor the lifetime of the user account, or until the account is deleted
9. Expert feedback (invited experts, dontes.app/szakerto)
Data processedThe user identifier; the question asked and the system's full answer together with the decisions found; the passages of the answer selected by the expert and the expert's free-text comments, with the time of creation and modification; the “no errors found” marker; technical data of the run (processing time, the model that wrote the answer, the amount of text processed, the browser identifier and the referring page); the time consent was given. The interface is available to invited experts only; comments are saved automatically while typing.
Legal basisconsent of the data subject [Art. 6(1)(a) GDPR], given before the first comment and revocable at any time
Retentionuntil consent is withdrawn, or until the user account is deleted; the expert can also delete individual comments in the interface
10. Machine access (MCP connection, dontes.app/en/gepi)
Data processedIf the user connects their AI assistant through machine access (no separate permission is needed; the data controller may block the access): the user identifier, the name and e-mail address (from the Clerk account, for managing the access), the status and allowance of the access, the identifier of the connecting application; for every call, the function called, the time, the result status, the processing time, the length of the question, the number of results and the cost; in addition, the parameters and outcome of the call: the text of the question or search expression, the courts selected and actually searched, the fields and values of the filters given (e.g. year, court level, legal provision, judge's name), the number of results and the language requested, the identifiers of the returned decisions and their number per court; when a decision is retrieved, its identifier and the length of the retrieved text; for the filter discovery functions, the field queried and the search term given. Point 2 applies to the processing of questions submitted by machine, and point 5 to keeping the text of the question for developing the Service; such questions are not added to the history under point 3.EUR-Lex tools (EU legislation and CJEU judgments; available to every user with machine access, the data controller may switch them off per user): the search terms, references (e.g. CELEX number, ECLI, case number), full-text search expressions and SPARQL queries given in the call by the user's AI client are sent to the Publications Office of the European Union (CELLAR, publications.europa.eu) and, for full-text search, to the EUR-Lex web service (eur-lex.europa.eu). We do not pass the user's identifier, name or e-mail address to the EU; the EU sees the server's IP address and the data controller's single, shared EUR-Lex account. The data controller's local call log holds, for these calls too, the data listed above (the user identifier, the name of the tool called, the time, the result status, the processing time, the identifier of the connecting application) and a short result note: for a successful call the length of the response in characters; for an error the name of the error class and a fixed error code. The call log contains no query content in any form: not the search terms, references, search expressions or SPARQL queries, and not the text of an error message; the length of the query is not recorded either. The daily allowance is calculated from these log rows. Public legal texts that are downloaded are cached on the server; they are not personal data.Please do not put personal data (e.g. the name of a natural person) into EUR-Lex queries, because the content of the query also reaches the EU.The user is responsible for the content of questions sent through machine access. The service asks the connected AI assistant not to include personal data in the question (e.g. the name, address, identification number or contact details of a client or other private person); incoming questions are, however, not filtered or anonymised.
Legal basisconsent of the data subject [Art. 6(1)(a) GDPR]; as regards the call log, legitimate interest of the data controller [Art. 6(1)(f) GDPR] — enforcing the allowance, tracking costs and preventing abuseThe same applies to the log and the daily allowance of EUR-Lex calls. Legal basis for sending the content of the query to the Publications Office of the European Union: legitimate interest of the data controller [Art. 6(1)(f) GDPR] — providing the EU-law search function requested by the user
Retentionuntil the machine access ends, or until the user account is deleted; the call-log rows are then kept without user and application identifiers, the text of the question and the call parameters, in anonymised form (cost records)
11. Technical logs (security, troubleshooting, visitor statistics)
Data processedThe web server makes a log entry for every request: the time, the IP address, the address of the page or function requested, the headers sent by the browser (e.g. the browser identifier and language, the referring page), the response code and the response time. The operational log of the server application may contain the user identifier and, in some entries, the first 80 characters of the question at most.From the web server log we compile anonymous, aggregated statistics on visits to the public decision pages (dontes.app/d/… and, in English, dontes.app/en/d/…): daily and hourly counts (human or robot), the page viewed, the domain of the referring page, the device type (mobile, desktop, tablet) and the browser language. No IP address goes into these; to filter out robots, the IP address is compared with public data-centre address lists during processing, but it is not stored. To count unique visitors we use an irreversible fingerprint formed with a new random key every day, which is deleted together with the key after two days.To prevent bulk downloading of the decision pages, we count requests to the decision pages per IP address and slow down serving when an address sends far more requests than usual. For this, the IP address and the times of the requests are kept for at most one week, exclusively in the server's memory: they are not written to disk or to any log, and they are erased when the throttling service restarts. The addresses of verified search engine crawlers (Google, Bing) are not counted.If the web server log shows bulk downloading from an IP address (or from several addresses of a data centre with the same browser identifier) within a short time (at least 150 decision pages per hour from one address, or at least 50 per address for coordinated downloading), we keep these requests separately in order to detect and prove the infringement: the IP address, the browser identifier, the time and number of requests and the addresses of at most 50 requested pages, and we suspend serving the decision pages to that address (for coordinated downloading from a data-centre address range, to the whole range) for up to 7 days (24 hours for a range and for addresses outside data centres) from the last such request; if this affects you by mistake, write to research.dontes.app@gmail.com and we will lift it without delay. We store no such data about other visitors. We keep it for 90 days; if we take legal action, the data concerning that case until the proceedings are closed.
Legal basislegitimate interest of the data controller [Art. 6(1)(f) GDPR] — security of the Service, detecting abuse and faults, measuring visits to the public pages
Retentionthe anonymous, aggregated statistics without time limit; the web server log rotates by size: the full log is typically overwritten within a day, and the log of the application traffic (without the public decision pages) is kept for at most 7 days; the separately kept data on bulk downloading for 90 days (if we take legal action, until the proceedings are closed); the operational log is kept until the server application is next restarted (updated); a log extract saved for investigating a fault may be kept until the investigation is closed, and the rows of a deleted account are removed from it too
12. Decisions of the European Court of Human Rights (ECtHR)
Data processedWhat this is. The Service makes searchable the decisions of the European Court of Human Rights, which are public in any case; we take them from the Court's public database (HUDOC, hudoc.echr.coe.int).Purpose. Legal research and information: making ECtHR case-law searchable. Non-commercial.Recipients. Users of the Service; OpenAI (via OpenRouter) and, as a fallback, Anthropic (to write the answer); Voyage AI (to vectorise the texts and the questions); OpenRouter (to translate quotations); Scaleway SAS (backups); users of the machine (MCP) access. The text of the decisions can also be read on the public decision pages (dontes.app/d/…; the decisions of the CJEU and the ECtHR also in English at dontes.app/en/d/…), by anyone and via search engines, just as in the Court's own database. If the Court later anonymises a decision, the daily update follows it; you can also request anonymisation via the controller's e-mail address.
Legal basisArticle 6(1)(f) GDPR (legitimate interest).
RetentionFor as long as the Service operates; in backups for up to 3 years.
13. Receiving and recording donations (dontes.app/en/donate)
Data processedIf someone donates by bank transfer or through Revolut's payment page (revolut.me): the data sent to the controller by its bank or by Revolut — the donor's name, account number (IBAN), bank, the amount, the date and the payment reference. No account is needed to donate, and donations are not linked to use of the Service. Card and Apple Pay payments are handled by Revolut Bank UAB as an independent controller, under its own privacy notice.The controller records each donation in the income and expense ledger required by law: the donor's name, the date, amount and currency of the donation, the bank transaction ID and the payment reference. The ledger and the related bank statements and exports are kept on the controller's server, in its encrypted backup (Scaleway SAS) and in the research.dontes.app@gmail.com mailbox (Google); only the controller has access to them.About the donation page and the donation reminder on the search page, the Service records only anonymous events (the box shown, a click, closing the box, opening the menu item or the donation page): the kind of event, its time and the interface language — without user ID or IP address. When the box appears is decided by a counter stored in the browser; it is not sent to the server.
Legal basiscompliance with a legal obligation [Art. 6(1)(c) GDPR] — income records and record keeping under Hungarian Act CXVII of 1995 on personal income tax and Act CL of 2017 on the rules of taxation. The anonymous events contain no personal data.
Retentionuntil the right to assess the tax lapses: 5 years from the last day of the calendar year in which the tax return covering the donation was due
14. Decisions of the Hungarian Competition Authority (GVH) and court decisions published by the GVH
Data processedWhat this is. The Service makes searchable the decisions of the Hungarian Competition Authority and the court decisions that the Authority publishes on its website (gvh.hu), which are public in any case. Court decisions published there but missing from the Collection of Court Decisions are added to the ordinary courts; in court decisions anonymised in the Collection, we restore company names and other non-personal data from the GVH's publication (marked as such).Data of natural persons. Names of natural persons are replaced by initials; their home addresses and e-mail addresses are removed. Exceptions are persons named in the GVH's publication who act in an official capacity: the acting lawyer (with name and address), the acting members and case handlers of the GVH and the court, experts, signatories, public office holders and authors of legal literature. Masking is automated: rules, a named-entity recogniser (HuSpaCy) and, in doubtful cases, Anthropic's Claude model decides, to which we send the passage around the name.Purpose. Legal research and information: making competition-law practice searchable. Non-commercial.Recipients. Users of the Service; OpenAI (via OpenRouter) and, as a fallback, Anthropic (to write the answer); Anthropic (to decide doubtful names); Voyage AI (to vectorise the texts and the questions); Scaleway SAS (backups); users of the machine (MCP) access. The court decisions can also be read on the public decision pages (dontes.app/d/…). If a masking is missing or wrong, you can request a correction via the controller's e-mail address.
Legal basisArticle 6(1)(f) GDPR (legitimate interest).
RetentionFor as long as the Service operates; in backups for up to 3 years.

Data processors

The following data processors take part in the processing (their privacy policies are linked in parentheses):

For the processing under points 2–3, the question entered by the user is transmitted to OpenAI (through OpenRouter) and Voyage AI — exceptionally, as a fallback, to Anthropic — in order to produce the answer, and is processed and stored on the servers of the hosting provider netcup. These providers process the data under their own privacy policies, to the extent necessary to perform the task.

The data stored on the Service's servers (points 1, 3–5, 7–10 and 12) is backed up at regular intervals, and the backups are stored in Scaleway's Paris data centre. Backups are encrypted before they are uploaded; the encryption key is held solely by the data controller, so Scaleway cannot access the contents of the backups.

If a user deletes an item of data (for example a history entry or an expert comment), it is removed from the live system immediately. If a user deletes their account, the data belonging to it is deleted within 24 hours at the latest from the live system and from the earlier copies kept on the server; the cost rows under points 4 and 10 are kept without any identifier, in anonymised form. The fact of the deletion — an irreversible digest of the account identifier, the time and the number of rows deleted — is kept until the backups rotate out, for at most 3 years, to prove the deletion and so that it can be repeated after a restore from backup. Deleted data disappears from earlier backups as those rotate out, within three years at the latest.

Recipients of personal data, data transfers

Personal data is not transmitted to anyone other than the data controller and the data processors named above, except as described in the paragraph below. Some of the data processors operate outside the European Union, in the United States (Clerk, Anthropic, Voyage AI, OpenRouter, OpenAI); no data is transferred to any third country other than the United States.

Exception — EUR-Lex tools (point 10). When a user uses the EUR-Lex tools, the content of the query (search terms, references, search expressions, SPARQL queries) is sent to the Publications Office of the European Union (Luxembourg; publications.europa.eu, eur-lex.europa.eu). The Publications Office does not act on behalf of the data controller but processes the requests it receives as an independent controller, under its own notice; it is not a data processor. The request does not contain the user's identifier: the Publications Office sees the server's IP address and the data controller's shared EUR-Lex account. The Publications Office operates within the European Union, so this does not involve a transfer to a third country.

Consequence of not providing data

Without a user account and a question, the Service cannot be used, since processing the question is a precondition of providing an answer.

Your rights

You have the right to

Right to object

You may object at any time, on grounds relating to your particular situation, to processing based on legitimate interest (points 4–6, 8, 11 and 12, and the call log under point 10) (Art. 21 GDPR). The data controller will then no longer process the data unless it demonstrates compelling legitimate grounds which override your interests, rights and freedoms, or the processing is needed for the establishment, exercise or defence of legal claims.

The data controller will respond to your request within one month at the latest. The Service does not make decisions about users based solely on automated processing (Art. 22 GDPR).

You can delete your account yourself in the user menu (click your profile picture: “Manage account” → “Delete account”); your data is then deleted as described in the paragraph on backups, without a separate request.

Legal remedy

Regarding the lawfulness of the processing, you may initiate a procedure before the Hungarian National Authority for Data Protection and Freedom of Information (NAIH) (1055 Budapest, Falk Miksa utca 9-11.; postal address: 1374 Budapest, Pf.: 603; website: www.naih.hu; phone: +36 (1) 391-1400; e-mail: ugyfelszolgalat@naih.hu), or you may turn to the court having jurisdiction at the data controller's seat or at your place of residence.

Data subject requests

Questions, comments and requests relating to the processing can be sent to research.dontes.app@gmail.com.

Effective date of this privacy notice: 2 June 2026. | Last updated: 18 November 2026.