Humanitext Antiqua Service Rules

1. Description of the Service

1. In response to a user's question, this Service searches a database of Western classics (original texts in Greek and Latin) for relevant passages, generates an answer with AI, and presents the original texts on which the answer is based together with a means of verifying them.
2. Users can narrow the scope of the search by specifying genres, authors, and works. Users can choose an answer mode (Q&A, Study, Research, Conversation, Custom) and a "No Context mode" that does not refer to external data.
3. The model used to generate answers and the number of original-text passages retrieved per question (number of contexts) can be specified within the range available under the user's plan. All models can be selected even on the free plan.

2. Information Handled (Service Rules under Article 3 of the Common Privacy Policy)

1. Usage statistics: the selected corpus scope, model, response mode, processing time, approximate number of tokens, success or failure, language setting, etc. These do not include the text of questions or answers, account identifiers, email addresses, or IP addresses. As for time, only the date, the hour of the day, and the day of the week are recorded.
2. Usage allowance records: for fair use and prevention of misuse, the daily number of uses and amount of processing are recorded for each account. Access frequency limits are also applied per account, and this Service does not read the IP address from which you connect (Cloudflare, which handles delivery, handles IP addresses in order to relay communications).
3. Consent records: the version consented to and the date and time, and whether research use has been permitted.
4. Dialogue records (research use): questions (including instruction text in Custom mode, text translated into English for searching, and internal values indicating a conversation thread), generated answers, and cited sources. These are recorded only if you have explicitly consented to research use under Section 3. By default, they are not recorded.
5. History in the browser: the exploration history displayed on screen is stored in your browser (local storage) and is not sent to the server.
6. To generate answers, your questions and the preceding conversation are sent to the businesses in Section 4. This transmission is temporary processing for creating answers and is separate from storage. Only the information within the scope described above is stored.

3. Research Use (Service Rules under Article 4, item 5 of the Common Privacy Policy)

1. Purpose: to analyze usage and purposes of use from dialogue records in order to improve the system, and to publish the results of the analysis as papers and conference presentations.
2. Consent: limited to cases where you have explicitly given permission on the consent screen or settings screen. The default is "Do not permit." Even if you do not give permission, you can use all functions as usual.
3. Withdrawal: you can withdraw at any time on the settings screen. After withdrawal, no further records are made. At that time, you can choose whether to keep or delete records already stored. You can also request deletion of kept records later.
4. Nature of the records: before recording, email addresses, telephone numbers, URLs, long strings of digits, and the like are automatically masked. Records are given only an internal identifier for research, and the table linking it to accounts is managed separately. However, because questions are free-form text, the possibility that an individual could be inferred from the content cannot be completely ruled out. Therefore, we do not call these records "anonymized information," and we treat them as falling under neither anonymously processed information nor pseudonymously processed information under the Act on the Protection of Personal Information.
5. Publication: individual records are never published as they are. What we publish is limited to aggregate figures, results of statistical analysis, and examples modified so that individuals cannot be inferred.
6. Retention period: 730 days from collection.

4. Service-Specific Contractors (Service Rules under Article 6 of the Common Privacy Policy)

For generating answers with AI, we use services provided under terms by which content sent is not used to train models (Google Cloud terms of service). Content sent to OpenAI for input screening is not used for training and is not stored on OpenAI's side (OpenAI's API data usage terms). Under its terms of service, Pinecone, which is used for search and reranking, may use content sent to it to improve its own services (there is no explicit statement that it is not used for model training). What is sent to Pinecone is the search text translated into English and the text of candidate original texts; account identifiers and email addresses are not sent. Please do not include personally identifiable information in your questions.

5. Service-Specific Retention Periods (Service Rules under Article 7 of the Common Privacy Policy)

6. Deletion and Export of This Service's Data

1. You can export (in JSON format) and delete your own data stored in this Service from "Account and data" on the screen.
2. Deletion erases research dialogue records and consent records, and replaces the daily number of uses and amount of processing with a form no longer linked to the account (these disappear once the retention period in Section 5 has passed). Usage statistics do not contain information that identifies individuals in the first place.
3. Deletion from the screen is limited to this Service's data, and your Humanitext ID account remains. If you delete your Humanitext ID account (Common Privacy Policy, Article 8, paragraph 3), this Service's data is likewise handled as described in the preceding paragraph.

7. Usage Allowances and Fees (Service Rules under Article 7 of the Common Terms of Use)

1. For fair use and to keep operating costs down, an upper limit on the amount of processing per day (usage allowance) is set for each user. The remaining amount is shown on screen as a percentage. The number of questions you can ask is not fixed and varies depending on the selected model and number of contexts. The usage allowance resets at 0:00 Coordinated Universal Time (9:00 a.m. Japan time).
2. This Service as a whole also has a daily processing limit; if it is reached, no new questions will be accepted until the next reset.
3. Paid plans will raise the upper limit on the number of contexts (planned to be 1–5 for free and 1–10 for paid) and the usage allowance. Information will be provided in accordance with Article 7, paragraph 3 of the Common Terms of Use when such plans become available.
4. When paid plans become available, we plan to change usage allowances to a weekly basis and to make them a common allowance for AI dialogue across Antiqua, Aozora, and GEO, for both free and paid users.
5. Even if response generation fails partway through, the amount processed by the AI up to the failure is counted against the usage allowance.
6. Dialogue with AI requires verification of your Humanitext ID email address (Common Terms of Use, Article 3, paragraph 1).

8. Announcements

The latest information is also announced on our official account (https://x.com/humanitext_x).

Established: October 6, 2026 (version 2026-10-06)