City halls, government agencies, and hospitals handle large volumes of documents on a daily basis: resolutions, explanatory notes, project reports, meeting minutes, and correspondence with citizens and other institutions. A large portion of civil servants’ time is spent reading, summarizing, rephrasing, and searching for information in these documents.
An AI assistant can take on some of this work. For a public institution, however, the important question is not „what AI can do,” but „where do our documents end up when we give them to an AI?” In this article, we explain what an AI assistant can actually do in public administration and why the on-premises option—hosted on the organization’s own collaboration platform—is the right choice for the public sector.
In a nutshell
- An AI assistant can summarize long documents and discussions, draft notes and meeting minutes, and translate materials.
- Institutional documents often contain personal data. Public AI tools from outside the EU raise concerns regarding the GDPR and data control.
- An AI assistant that runs locally in Nextcloud on servers in Romania processes documents without sending them to third parties.
- The decision rests with the civil servant: The AI proposes drafts, and the person reviews and signs them.
What Is an AI Assistant for Administration?
We're not talking here about the chatbot on the institution's website that responds to citizens (I wrote about that in Guide to AI Chatbots for Public Institutions). We're talking about an assistant that is used inside, by employees, integrated into the platform where the organization already stores its files, email, calendar, and team chats.
Nextcloud is one such platform: an open-source alternative to the cloud services offered by major American companies, used by public administrations in several European countries. Nextcloud’s AI assistant can work with language models running on the institution’s or the hosting provider’s servers, without sending text to external services.
5 Practical Applications in Institutions
1. Summary of Long Documents
A 60-page report, a set of specifications, or a draft resolution can be summarized in a few paragraphs. The civil servant quickly learns what the document is about, what deadlines it contains, and what needs to be checked, and then reads only the relevant parts in detail.
2. Summary of email and chat discussions
Long email threads between departments or discussions in the team’s internal chat can be condensed into a summary: what was decided, who is responsible for what, and what remains unresolved. This is especially helpful for colleagues returning from vacation or taking over a project.
3. Drafts of notes, letters, and meeting minutes
Starting from notes or a template, the assistant can draft a first version of an internal memo, a response to a petitioner, or meeting minutes. The Context Write feature in Nextcloud allows, for example, a text to be rewritten in the style of a reference document. The civil servant simply corrects and adds to the text; they no longer have to start from a blank page.
4. Updates on the Status of Projects
With the Context Chat feature, the assistant can answer questions based on documents on the platform: „What is the deadline specified in Contract X?”, „What was decided at the last meeting regarding Project Y?”. The answers are based on files to which the user already has access.
5. Translations for Documents and Communication
European-funded projects, international partnerships, or communication with foreign citizens often require translations. Nextcloud has machine translation modules that can run locally. For official documents, machine translation should be treated as a draft that must be reviewed by a human.
Nextcloud with an AI assistant, hosted in Romania. Romleas offers Nextcloud with an on-premises AI assistant, hosted on GPU-powered servers in Oradea: files, email, calendar, video conferencing, and AI all on a single platform, with data stored domestically. See the Nextcloud solution.
Why It Matters Where AI Runs
Documents from public institutions often contain personal data: names, national identification numbers, addresses, social status, and medical information. If a public servant copies such a document into a free, publicly available AI tool, the data may end up on servers outside the European Union, and the institution loses control over it.
For this reason, more and more European institutions are opting for solutions in which the AI model runs on their own infrastructure or that of a local provider. The advantages are clear:
- Simpler GDPR Compliance: The data is not transferred outside the EU, and the processing chain is short and easy to document.
- No training on the institution's premises: The documents are not used to train third-party models.
- Access Control: The assistant can only view the documents for which the user already has permissions on the platform.
- Predictable costs: There is no charge per user or per request to an external provider.
I wrote in more detail about the digital independence of institutions in the article on Digital Sovereignty Index.
What AI Shouldn't Do
An AI assistant is a work tool, not a decision-maker. In government, a few simple rules can prevent problems:
- The man checks everything that leaves the facility. Responses to petitioners, official letters, and decisions are signed by an official after verification.
- AI does not make decisions regarding people's rights. The use of AI to determine who receives a social benefit or public service falls under the category of high-risk systems in the AI Act and is subject to strict requirements. For more details, see our article on AI Act and Chatbots.
- Employees know what they're using. A brief overview of what the assistant can and cannot do reduces errors and helps people use it effectively.
How to turn it on
- Choose a pilot compartment, a lot of paperwork: records management, procurement, human resources, or project management.
- Identify 2–3 specific uses, For example, document summaries and draft meeting minutes.
- Try it out for a few weeks and measure the time saved.
- Expand gradually in other departments, with written internal rules for use.
Frequently Asked Questions
Is the AI assistant as good as ChatGPT?
Models that run locally are generally smaller than the largest commercial models. They are sufficient for typical administrative tasks (summarization, rephrasing, translation, document search), and their main advantage is control over the data.
Do we need to replace the entire IT system?
No. Nextcloud can be rolled out gradually, starting with a single folder, and can be integrated with existing email and authentication systems.
Who has access to the documents?
Only users of the institution, in accordance with the permissions set on the platform. The AI assistant operates within the same access limits as the user who is using it.
Where is the data stored?
At Romleas, on servers in Oradea, Romania. The data is not transferred outside the country and is not used to train other models.
Conclusion
A well-chosen AI assistant can relieve civil servants of some of their repetitive document-related work: summaries, drafts, translations, and searches. For the public sector, the right choice is an assistant that runs locally, within the institution’s platform, with data stored in Romania and with a human making the final decisions.
If you'd like to see how it would work at your institution, Write to us and we'll prepare a demonstration using sample documents.
This information is current as of the date of publication. Available features depend on the Nextcloud version and the installed AI modules.
Read more
- AI Chatbots for Public Institutions: A Guide for City Halls, Hospitals, and Government Agencies
- The AI Act and Chatbots: What Romanian Companies Need to Do
- AI Assistant for Businesses: 6 Uses
