Data protection FAQ

This page answers frequently asked questions about data protection in Ayunis Core. Open a question to view its answer. For a visual explanation, see Data flow: Where does a request go?.

Hosting and server location

Where is Ayunis Core hosted?

The platform and all data stored at Ayunis – including user data, chat content and content from knowledge collections – is held at Hetzner in Nuremberg, Germany.

Where are the language models hosted?

The models are provided by external providers and operate exclusively in Germany or the EU:

Open-source models such as GPT OSS and Qwen 3: through StackIT in Germany
GPT models: through Microsoft Azure in Europe
Claude models: through AWS in Germany/Frankfurt
Mistral Medium: through Mistral in Europe
Image generation with GPT Image: in the EU

Administrators can restrict the available models. The model selector indicates each region with a flag.

Which image generation models operate in the EU?

GPT Image 1 and its successor models operate in the EU.

How is it ensured that data remains in the EU?

Storage and processing locations are covered by the privacy terms and legal documents. These are available at ayunis.com/rechtliche-dokumente.

Data transfer and model providers

Does our data end up in the USA?

No. Ayunis stores data exclusively in Germany. Approved models operate exclusively in Germany or the EU. This also applies to models provided through Microsoft Azure and AWS.

To generate a response, Ayunis Core securely transmits the required chat history to the approved model. The provider processes this content only temporarily, does not store it and does not use it for training. In anonymous mode, personal data is replaced with placeholders on Ayunis infrastructure before transmission.

For external integrations, the additional data flow depends on the connected system. Administrators should therefore check how that system processes data before approving it.

What is the difference between ChatGPT and using a GPT model through Ayunis Core?

In Ayunis Core, GPT models are provided through Microsoft Azure in an EU region. Permanent storage takes place at Ayunis in Germany. Microsoft Azure processes transmitted content only to generate the response, does not store it and does not use it for training. Anonymous mode can additionally replace personal data before transmission.

Who decides which models are used?

Administrators approve providers and individual models, set an organization-wide default model and can restrict access by team. Users can only select from the models approved for them.

Is only the new question transmitted with each message, or are earlier messages included?

The existing chat history is transmitted to the approved model with each message so that the model can understand the context. The history remains permanently stored at Ayunis in Germany; the model provider processes it only temporarily to generate the response.

What data is transferred through external integrations?

Organizations can connect their own specialist systems through MCP integrations, such as document management or council information systems. The data required for the respective call is also transmitted to the connected system. The exact data flow depends on the integration and its purpose and should be assessed before approval.

Contracts, storage and deletion

Where can I find the data processing agreement and list of sub-processors?

The data processing agreement, including the list of sub-processors, is available at ayunis.com/avv. Each organization decides which models and providers it uses.

How long are chat histories and other data stored?

Administrators can configure deletion periods for inactive conversations. If no deletion period is set, chat histories remain until they are deleted manually. The data is held at Hetzner in Germany.

For more information, see Data retention.

What happens when a chat or account is deleted?

A deleted chat is completely removed and cannot be recovered. Customer accounts can also be fully deleted.

Where are documents from knowledge collections stored?

Uploaded original files, such as PDFs, are discarded after processing. The prepared document content and search index are stored with the other data at Hetzner in Nuremberg.

Is chat history stored, and must it be deleted manually?

Chat histories are stored at Ayunis in Germany. They can be deleted manually. Administrators can also configure automatic deletion periods for inactive conversations.

What happens to our data when the contract ends?

A complete export is available on request. The data is permanently deleted no later than 30 days after the end of the contract, and deletion is confirmed in writing.

Use for training

Is our data used to train AI models?

No. Exclusion is active by default. Neither models operated by Ayunis nor connected model providers use transmitted content for training. Providers contractually guarantee that the content is neither stored nor used for training.

Can exclusion from training use be explicitly guaranteed?

Yes. The exclusion is governed by the data protection concept.

Anonymization and personal data

Where does anonymization take place?

Anonymization runs on Ayunis infrastructure at Hetzner. No data is transmitted to an external service during this step. Personal data is replaced with placeholders before content is passed to an external model.

When is anonymous mode active, and what content is masked?

Anonymous mode is active when a user explicitly starts a new anonymous conversation or when administrators require anonymization for the selected model.

Both user input and results from connected tools are automatically masked before the model call.

Why is anonymization needed if model providers do not store the content?

Anonymous mode additionally prevents personal data from being transmitted in plain text to an external model provider. Regardless of this, permanent storage takes place exclusively at Ayunis in Germany.

How reliably does anonymous mode detect personal data?

The model is designed to detect too much rather than too little when in doubt, but complete detection cannot be guaranteed. The feature therefore does not replace organizational rules or careful review.

Are sensitive user inputs blocked?

No. Ayunis does not generally block user input. Instead, technical and organizational measures help protect sensitive data:

Administrators can configure models so that they can only be used in anonymous conversations.
Administrators can define organization-wide exceptions so that certain data types or content are not anonymized, such as the organization’s own name.

How does anonymization handle indirectly identifying information?

Indirect information such as “the data protection officer of the mayor’s office” is generally not anonymized. Complete detection cannot be guaranteed. For more control, administrators can define exceptions based on data categories and optional patterns.

Are placeholders replaced with the original values in the response?

Yes. After the model responds, Ayunis Core restores the original values and highlights them in color. A document is only anonymized in the context of an anonymous chat, not by default.

What can I do if anonymous mode masks too much?

Anonymous mode tends to mask too much rather than too little. Administrators can define organization-wide exceptions based on data categories and optional patterns. One hundred percent accuracy is not technically possible.

Does anonymization apply to directly uploaded images or screenshots?

No. Anonymization applies to text and documents, but not to directly uploaded images. If an image is first inserted into a supported text document, text read from it can be processed.

Can anonymization be configured per team?

No. A single anonymization configuration currently applies to the entire organization.

Can a specific abbreviation be excluded from anonymization?

Yes. Administrators can define organization-wide exceptions. These rules are based on data categories and optional patterns, such as regular expressions.

Can personal data be entered?

Technically, yes. For additional protection, a conversation can be started in anonymous mode. The organization itself decides which personal data may be processed, for example together with its data protection officer.

Logging and audit

Which access and activities are logged?

Only metadata is logged, such as the number of messages sent and which tools the AI used. Message content is not logged. Only Ayunis has access to these logs; no external logging service is used. Logs are deleted after 90 days.

How is GDPR compliance ensured?

Data is stored exclusively at Ayunis in Germany. During a model call, only the required content is transmitted and the provider does not store it. Accounts and chats can be deleted, and storage duration can be configured. Further details are governed by the legal documents.

What precautions are taken regarding the EU AI Act?

Ayunis Core is not a high-risk system. There are therefore no special requirements beyond the usual data protection and security requirements. Processes are presented transparently, data access only takes place as directed by users, and administrators can restrict the available models.

Does Ayunis have access to data processed in the application?

The data is held at Ayunis in Germany. Chats and accounts can be deleted, and administrators can configure deletion periods. Data subject rights are supported in accordance with contractual and legal requirements.

Can access be restricted to an administrative or Citrix network?

Administrators can restrict access by IP address under Security. This can limit access to specific networks or Citrix connections. It cannot completely prevent private use in every situation, for example when working from home.

Does the German server location make uploading personal data risk-free?

Permanent storage takes place in Germany. To generate a response, however, the required content is transmitted to the approved model endpoint in Germany or the EU and processed there temporarily. Anonymous mode is recommended for particularly sensitive data. The organization’s own rules remain decisive.

Can a skill be restricted to one specific model?

An individual skill cannot currently be restricted to exactly one model. Administrators can, however, control which models are available to the organization or individual teams.

Is there a ready-made data protection skill?

No. Organizations can create their own skill and provide their privacy terms through instructions or a knowledge collection.

How is information security ensured?

Ayunis has an established information security management system. ISO certification is currently in progress. A data processing agreement is also concluded with each organization, and security measures are based on the German Federal Office for Information Security’s IT-Grundschutz framework.

Legal documents and operation

Where can I find a record of processing activities under Article 30 GDPR?

The record is provided on specific request and is not published by default.

Where are liability provisions defined?

Liability provisions are set out in the terms and conditions, not in the data processing agreement.

Where are recurring data protection topics collected?

Data-protection-related topics are collected in the data protection concept, which is available on request. Recurring questions and answers are collected in this FAQ.

Can Ayunis Core be self-hosted?

Yes. Ayunis Core is published under the AGPLv3 license and can be self-hosted. In that case, the organization’s own IT team is responsible for operation, updates and infrastructure.

Web search

What data is transmitted to Brave during a web search?

Only the search query is transmitted to Brave, comparable to entering it into a search engine. The rest of the chat history remains at Ayunis and is not transmitted to Brave. Web search can be disabled for an organization.