Frequently asked questions

The Agentic Discovery Platform (ADP) is an artificial intelligence platform developed jointly by the Hungarian Research Network (HUN-REN) and Óbuda University as a consortium. A separate instance built for HUN-REN is operated by the HUN-REN AI Service Centre.

The ADP gives you access, through a single interface, to local models running on HUN-REN's own infrastructure and to language models and AI tools from external providers.

Besides models, the ADP offers further tools and agentic features. Depending on the tool, these can process files, search for information, run code, analyse data or carry out multi-step workflows.

For research, research and development, administrative, operational and other research support tasks. Typical examples include summarising texts, translation, document processing, literature research, code generation, data cleaning, statistical analysis, data visualisation and hypothesis generation.

Staff of HUN-REN institutes that have signed both the framework service agreement and the individual service agreement for ADP access.

Staff of contracted institutes automatically get basic access. For extended access, you need to complete the ‘Agentic Discovery Platform training – Researcher access level’ (Hungarian title: ‘Agentic Discovery Platform képzés – Kutatói hozzáférési szint’) course and pass the final exam. Your institute controls the use of tools that incur costs through users' token budgets.

You sign in through your institute's login service (KININ SSO). You must not pass your personal access on to anyone else.

The ADP is available at: https://adp.hun-ren.hu/

Open the ADP (https://adp.hun-ren.hu/), select ‘Bejelentkezés HUN-REN SSO’ (Sign in with HUN-REN SSO), choose your institute and sign in with your own institutional credentials (KININ SSO). Use only your own account.

If your institute isn't listed or sign-in fails, contact your institute's Platform Administrator. Contact them too if the tools available with extended access don't appear after you have passed the exam, or if a tool appears but you can't use it because you have no token budget. You can report technical faults to adp.support@hun-ren.hu.

The framework agreement sets out the general legal and operational terms of cooperation between the HUN-REN Central Office and the institute. Users don't sign contracts individually: the institute must have the agreements in place, and it manages its staff's access and permissions. You can find the current list of contracted institutes at ai4science.hun-ren.hu/aszf.

You can find the current list of contracted institutes at ai4science.hun-ren.hu/aszf. You can also ask your institute's Platform Administrator, AI officer or institutional AI ambassador, or its legal or management contact. If you don't know who holds these roles, your institute's director or IT security officer can point you in the right direction.

Staff of contracted institutes get basic access automatically, so you don't need to request it. You get extended access by completing the course and passing the final exam.

For questions about token budgets, other permissions or institutional conditions, contact your institute's Platform Administrator or AI officer.

No. Everyone may use the ADP only with their own login credentials. Passing on your account or login details counts as misuse and may lead to your access being restricted or withdrawn.

You don't need any training for basic access. After signing in, staff of contracted institutes can automatically use the locally run tools available with basic access, including LocalGPT.

To get extended access, you need to complete the Moodle course ‘Agentic Discovery Platform training – Researcher access level’ (Hungarian title: ‘Agentic Discovery Platform képzés – Kutatói hozzáférési szint’) and pass the final exam. This is required to access tools in security categories 2, 3 and 4. To use custom solutions in security category 4, you may also need a separate request from your institute, a risk assessment and approval.

The ‘AI Governance basics’ course (Hungarian title: ‘AI Governance alapok’) is not required to use the ADP, but we strongly recommend it for the responsible and secure use of AI.

The ADP course, ‘Agentic Discovery Platform training – Researcher access level’ (Hungarian title: ‘Agentic Discovery Platform képzés – Kutatói hozzáférési szint’), is available in HUN-REN's Moodle system: https://elearning.hun-ren.hu/ The courses are currently only available in Hungarian.

If it's your first time in Moodle, sign in with your institutional username and password. The system then automatically creates your user account, which takes a few minutes; after that, you'll see the available courses. If you still can't see them, sign out and back in, or read the login guide for further steps: Login guide (PDF)

If you have used HUN-REN's e-learning system before, the courses assigned to you will appear automatically under your courses when you sign in. If the ADP course doesn't appear, or you don't know whether your institute has joined the ADP, contact your institute's Platform Administrator.

Once you pass the final exam, all the tools available with extended access automatically appear for you in the ADP. Which of them you can actually use depends on the token budget set for you.

The version of ToolUniverse that runs on LocalGPT is free to use without a token budget. Tools that incur costs (such as frontier models, including ChatGPT) can only be used if your institute provides a budget for them.

Your institute's Platform Administrator can help with questions about your token budget.

The range of tools may change. You can find the current core list in the ADP tool list (ai4science.hun-ren.hu/aszf). Institutes decide for themselves which tools they fund and who can use them, so the tools actually available may differ from one institute to another.

For more sensitive data that can still be handled lawfully, a solution running on HUN-REN's internal infrastructure or a separately approved local solution may be suitable (e.g. LocalGPT). If a tool uses web search or an external service, bear in mind that part of your query may be sent to an external provider.

With a category 4 agent that can take actions, always grant only the minimum permissions strictly needed for the task, and any significant or irreversible action must be approved in advance by a person.

The security category shows where a tool runs, whether it connects to external services, and whether it can take actions on its own.

Category 0 tools are local solutions that run on the user's workstation or on isolated institutional infrastructure, with no internal or external network connection. They can be provided as a local installation on request; they are not part of the service automatically available at adp.hun-ren.hu. Their installation, operation, updates and security monitoring must be arranged separately.

Category 1 tools run on HUN-REN's internal infrastructure – currently the Science Cloud environment provided by HUN-REN SZTAKI and HUN-REN Wigner – with no connection to external tools.

Category 2 tools also run in an internal environment but may use web search or other external tools. As a result, part of your query may be sent to an external provider.

Category 3 tools run partly or entirely in an external provider's environment.

Category 4 tools are autonomous agents that run in an external provider's environment and can take actions. Depending on the permissions they are given, they can create or modify files, send messages, change databases or other IT systems, or start external processes.

You can find the current category of each tool in the tool list on the ADP terms page (ai4science.hun-ren.hu/aszf).

To use tools in categories 2, 3 and 4, you need to complete the required training and pass the final exam.

LocalGPT runs on HUN-REN Science Cloud infrastructure, in an environment provided by HUN-REN SZTAKI and HUN-REN Wigner, so processing stays within HUN-REN's environment. (This is not the same as a category 0 solution, which can be installed on request on the user's workstation or on isolated institutional infrastructure and is not part of the service automatically available at adp.hun-ren.hu.)

With a solution that uses the web or external tools, part of your query may be sent to an external provider, so you also need to take that provider's data processing terms into account.

Running a model locally does not make it lawful to enter prohibited, unauthorised or confidential data.

Under the contract, basic access and the use of open-source tools running on HUN-REN infrastructure are free of charge.

Some tools from external providers may require a fee, a token budget, an API key or a subscription. For security category 4 tools, integration, logging and security monitoring may also involve extra costs.

Your institute decides how these are funded and what usage limits apply.

A token budget is an allowance for using paid external models or tools. It does not come automatically with ADP access.

The budget is allocated by the Platform Administrator appointed by your institute, based on the institute's internal decisions, policies and available funds.

Yes, the ADP service framework allows custom tools and agents to be developed. The costs of tokens, subscriptions, integrations and operation are borne by the institute.

First discuss your request with your institute's Platform Administrator or AI officer and have it approved under your internal rules. Then send it to ADP support (adp.support@hun-ren.hu), briefly describing the task, the data, the result you want and who is expected to use it. We carry out a technical, security, data protection, legal, cost and capacity review for every request.

The tools can handle many kinds of documents, text, code, image and data files. What matters, though, is not the format but the content: only upload data that you are authorised to handle and that is consistent with the tool's security category, your institute's rules and the principle of purpose limitation.

Wherever possible, keep personal data out of your prompts and uploaded files. If this can't be avoided, you need to clarify in advance the legal basis, how data subjects will be informed or give consent, which institutional permissions are needed, and which tool may be used.

Special categories of personal data, health or biometric data, data relating to children and employee performance data must not be uploaded without separate prior permission. If in doubt, don't upload the data – ask your institute's data protection contact or Data Protection Officer (DPO) first.

Only if using the data is lawful, your institute's and project's terms allow it, there are no confidentiality or partner restrictions, and you have chosen a tool suited to the sensitivity of the data. A local tool can reduce the risk of data being transferred externally, but it is no substitute for checking your authorisation and any confidentiality obligations.

If the research data is subject to a partner agreement, NDA, grant condition, publication embargo or other restriction, don't upload it without prior approval.

Don't upload it until this has been clarified. Write a short description of the type of data, its source, the purpose of the planned use, the tool, and whether any data would be sent to an external provider.

Take this to your institute's data protection contact, Data Protection Officer (DPO), IT security officer or AI officer. If the case may involve high-risk use, prior written notification and approval may also be required.

To provide access and run the service, the platform may process your institutional email address and authentication data, log data on your sign-ins and model use, the names, types and content of the files you upload, and your prompt and response history. For access to category 2, 3 and 4 tools, we also keep records of your completed training and exam.

Only staff whose job requires it can see this data. When you use an external model, that provider's data processing terms also apply.

According to the privacy notice, personal data related to your use of the platform is generally kept for 24 months. Technical and security log data may be archived in anonymised form for longer.

For external tools, the provider's retention rules also apply. This means that deleting a conversation or upload in the interface does not necessarily delete all related logs and data held by external providers straight away.

Yes. Under the privacy notice, you can request access, rectification, restriction or – where the legal conditions are met – erasure, and you can object to the processing. Send your request first to HUN-REN's data controller contact or to the Data Protection Officer.

Your rights in detail, the exceptions and the current contact details are set out in the current ADP Privacy Notice (ADP Adatkezelési Tájékoztató) at ai4science.hun-ren.hu/aszf.

On the ADP, you are interacting with artificial intelligence. The answers the tools display are generated or compiled by AI; they are not verified expert opinions from a human.

The tools can be a useful starting point, but they can also give incorrect, incomplete, outdated or fabricated information. Check every result to a degree that matches the risk of the task. Don't base important professional, research, legal, financial or medical decisions solely on AI-generated answers.

Ask for sources and a description of the method, but check these too. For research results, validate the data, calculations, code, references and reasoning.

You, the user. AI does not take over professional, research, data protection, copyright or decision-making responsibility. A person must check the result before it is used and, where necessary, validate it in line with your institute's approval procedures.

Depending on the situation, yes. When publishing AI-generated or manipulated content, you must follow the labelling rules, especially for deepfake-type image, audio or video content.

For scientific publications, also follow the AI use and authorship rules of the journal, publisher, funder and your institute; these may be stricter than the general rules.

Intellectual works created in the course of employment or another institutional relationship are governed by your institute's intellectual property rules, the relevant contracts and applicable law. Using the ADP does not override these.

Third-party content used as input or appearing in the output may be subject to copyright or licence terms.

You must not use the ADP for unlawful, harmful, deceptive or discriminatory purposes. You must not circumvent access controls, overload the system, share login details, upload infected or unlawful content, or engage in practices prohibited under Article 5 of the AI Act. For example, the ADP must not be used for social scoring, prohibited biometric categorisation or emotion recognition, or to create harmful or inciting content, forged documents or other unlawful content.

The EU AI Act classifies as high-risk those uses of AI that can significantly affect people's rights, opportunities or circumstances. Examples include using AI in admissions, examinations, educational assessment, employment decisions or employee performance evaluation.

You must not start an activity like this on the ADP on your own. If a planned use could be classed as high-risk, first obtain your institute's internal approval, then notify AISC in writing in advance at adp.support@hun-ren.hu.

First check your internet connection, refresh the page, sign out and back in, then try another supported browser. Note the time of the error, the tool affected and the error message, and take a screenshot (without any sensitive data).

Report technical faults to adp.support@hun-ren.hu. Don't send confidential prompts, personal data or entire sensitive files by email; give only the minimum information needed to reproduce the error.

An incident is any event, or well-founded suspicion, involving personal or confidential data, unauthorised access, prohibited data entry or AI use, high-risk use, information security, or a harmful result used without being checked.

You don't need to wait until you're certain: a suspected incident must also be treated as an incident and reported promptly.

Follow the current incident management protocol, which you can find on the ADP terms page: ai4science.hun-ren.hu/aszf

General ADP questions, technical support and notification of high-risk use: adp.support@hun-ren.hu

Reporting an AI incident: adp.incidens@hun-ren.hu

For data protection questions, first contact your institute's data protection contact or Data Protection Officer (DPO). The current contact details of HUN-REN's Data Protection Officer are given in the ADP Privacy Notice on the ai4science.hun-ren.hu/aszf page.

Continuous, round-the-clock availability is not guaranteed. Some features may be restricted or suspended for planned maintenance, for security reasons, because of faults at external providers or in cases of force majeure.

If an external model is unavailable, the ADP operator may not be able to restore it directly. In that case, check whether another tool suitable for the same purpose is available.

Access may be restricted, suspended or withdrawn because of a breach of the terms of use, a system security risk or maintenance. Reasons may include sharing your account, entering prohibited data, circumventing access controls or putting the system at risk.

Documents relating to the ADP are available at ai4science.hun-ren.hu/aszf.