KURE
한국어Sign in

Terms of Service

This is a translation; if it differs from the Korean version, the Korean version prevails.

Effective date: October 21, 2026

1. Purpose

These terms govern the use of the Korean retrieval model APIs offered at kure.kullm.ai (the "KURE API"; currently embeddings, sparse embeddings and reranking) and the website (together, the "Service") provided by the NLP & AI Lab and HIAI Institute, Korea University (the "Operator"), and set out the rights and obligations of the Operator and users.

2. Definitions

  1. User: anyone who agrees to these terms and uses the Service. A user who signs up with a GitHub or Google account is a member.
  2. API key: an authentication string issued to a member for API requests.
  3. Input: text a user sends to the KURE API or the playground (such as search queries and documents).
  4. Output: what the Service returns from inputs (such as embedding vectors, sparse vectors, similarity and relevance scores, and rankings).

3. Posting and changes

  1. The Operator posts these terms on the Service.
  2. The Operator may change these terms within the limits of applicable law and will announce changes on the Service at least 7 days before they take effect, or at least 30 days before for changes unfavorable to users, also emailing members where an email address is available.
  3. Members who do not agree to the changes may delete their account. Continuing to use the Service after the effective date means you accept the changes.

4. Sign-up

  1. The agreement is formed when a person signed in with GitHub or Google agrees to these terms and the Privacy Policy and completes sign-up.
  2. You must be at least 14 years old to sign up.
  3. The Operator may refuse or later terminate an account that uses someone else's account, was previously restricted for violating these terms, or is created for a prohibited purpose under Section 9.

5. The Service and changes

  1. The Operator provides the KURE API, a playground, documentation, and a member dashboard. The available APIs and models are listed in the Models documentation, and the Operator may add APIs and models.
  2. The Service is a free research preview. The Operator may change or discontinue all or part of it, and will give 30 days' notice before shutting down the whole Service.
  3. Models are retired according to the model lifecycle, at least 30 days after notice.
  4. The Service may be suspended without notice for maintenance, outages, or other unavoidable reasons.

6. Limits

  1. The Operator applies per-account limits on requests and tokens. Current limits are shown in Limits & errors and the dashboard.
  2. Limits may change with operating conditions. Lowering the default limits will be announced 7 days in advance.
  3. Members may request higher limits; the Operator may adjust limits per account after review.

7. Accounts and API keys

  1. Members must keep their GitHub or Google account and API keys secure, and must not place API keys where others can see them, such as public repositories or client apps.
  2. Use made with a member's API key is treated as the member's use. If a key may have leaked, revoke it in the dashboard right away and let the Operator know.

8. Non-commercial use

  1. The Service and outputs may be used only for research, education, personal learning, and non-profit purposes.
  2. The following count as commercial use and are not allowed without a separate agreement with the Operator:
    • including the Service or outputs in, or using them to operate, a product or service offered for profit
    • providing the Service or outputs for a fee, or reselling them
  3. For commercial use, please contact us about enterprise use.

9. Prohibited conduct

Users must not:

  1. use the Service for purposes that violate laws or public order
  2. send inputs that infringe others' rights, including personal information, trade secrets, or copyrights; in particular, sensitive personal data such as national ID numbers, bank or card numbers, or health information without a lawful basis
  3. circumvent limits, for example by creating multiple accounts or using others' keys
  4. overload the Service, probe or breach its security, or disrupt its operation
  5. collect outputs at scale to replicate the model or extract its weights or training data
  6. transfer, lend, or sell API keys
  7. use the Service commercially in breach of Section 8

10. Rights in inputs and outputs

  1. Users keep the rights to their inputs and warrant that they are entitled to send them.
  2. Users may use outputs freely within the scope of Section 8.
  3. By default the Operator does not store input text. To run the Service it keeps only usage records such as request time, number of inputs, and token counts (see the Privacy Policy).
  4. Rights in the Service, models, documentation, and website belong to the Operator or the rightful owners.
  1. Members may allow input storage at sign-up or in dashboard settings. This is optional, and declining has no disadvantage other than the default limits; members who allow it get higher limits.
  2. Inputs a consenting member sends after giving consent (via the KURE API and playground) are used only for research to improve Korean retrieval models, including building training and evaluation data.
  3. Stored inputs are never published or provided to third parties. Models trained with them may be released, and the Operator takes care that they do not reproduce stored text.
  4. Before storage, the Operator masks national ID numbers, card numbers, phone numbers, and email addresses. Free-form information such as names and addresses may not all be masked.
  5. Stored inputs are deleted 180 days after collection.
  6. Members may withdraw consent or delete stored inputs at any time. Withdrawing and deleting are separate. After withdrawal, new inputs are not stored; inputs stored before withdrawal are kept until their retention ends unless deleted. Deleted inputs are removed without delay and not used in later research data or models. However, the influence of specific inputs cannot be removed from a model that has already been trained.
  7. Consenting members must still not send the information described in Section 9(2).
  8. The Operator will ask members for consent again before storing non-text inputs or changing the purpose in paragraph 2.

12. Restriction and termination

  1. If a user violates these terms, the Operator may revoke API keys or restrict the account without prior notice. Except in urgent cases, the Operator will tell the member why by email (where available) and give a chance to respond.
  2. Members may delete their account at any time from the dashboard. API keys and sessions stop working immediately, and personal data is handled under the Privacy Policy.

13. Disclaimer

  1. The Service is provided as is. The Operator does not guarantee uninterrupted service, the accuracy or suitability of outputs, or fitness for a particular purpose.
  2. Embeddings, sparse vectors, and similarity and relevance scores are outputs of a statistical model; users are responsible for decisions based on them (search results, recommendations, classification, and so on).

14. Limitation of liability

  1. The Operator is not liable for damages arising from use of this free Service unless caused by the Operator's intent or gross negligence.
  2. The Operator is not liable for damages caused by events beyond its control, such as natural disasters, power or network failures, or outages of third-party services (such as GitHub or Google).
  3. Users who cause damage to the Operator or third parties by violating these terms must compensate for that damage.

15. Notices and contact

  1. The Operator gives notice on the Service or by email to members (where available).
  2. Questions: Contact us or kure.nlpai@gmail.com.

16. Governing law and jurisdiction

  1. These terms are governed by the laws of the Republic of Korea.
  2. Disputes are resolved by the court with jurisdiction under the Korean Civil Procedure Act.

These terms take effect on October 21, 2026. (The October 17, 2026 version with sign-up via Google added. The October 17 version was the October 16, 2026 version with the Service widened to the whole KURE API at kure.kullm.ai. The October 16 version added the sparse embedding and reranking APIs to the October 15 version.)