Please note: The algorithm descriptions in English have been automatically translated. Errors may have been introduced in this process. For the original descriptions, go to the Dutch version of the Algorithm Register.
Municipality-wide AI assistant
- Publication category
- Impactful algorithms
- Impact assessment
- AIIA
- Status
- In development
General information
Theme
Begin date
Contact information
Responsible use
Goal and impact
The Municipal AI Assistant was developed to support employees of De Fryske Marren in textual work, such as writing, rewriting and summarising internal documents and policy information. The system aims to save time through more efficient text processing, while providing insight into the uses and limits of AI within the organisation.
The impact is supportive: the AI assistant facilitates business operations without automated decision-making. Employees remain responsible for content choices and use of the generated output. The deployment is proportional, focused on non-critical tasks, and complies with the AVG via technical mitigation (filtering) and the basis of public duty. Ethical frameworks, such as transparency and privacy, are secured through an AIIA.
Considerations
There are several trade-offs to ensure responsible use. These trade-offs are recorded in the AIIA and have the following main considerations:
- A self-hosted solution within the municipal ICT architecture and a European AI model was chosen to ensure data control and AVG compliance. This is in line with the public task basis, where personal data is only processed for support tasks without automated decision-making. Technical measures, such as filtering and temporary storage of data, mitigate risks of data leakage or inadvertent reuse.
- Ethical and legal frameworks are explicitly defined. For instance, it has been established that output is always checked by employees (human-in-the-loop), and that archiving takes place according to the Archive Act through the DIV team and Team Information. In addition, processing agreements have been concluded with the supplier (technical support) and Mistral (model access).
Human intervention
Human control is central to the municipality-wide AI assistant. The employee is always ultimately responsible for the output: the AI generates suggestions, but the user judges, adapts and authorises the final text. This principle (human-in-the-loop) is enshrined in the AIIA and is in line with the AVG and the European AI Regulation, which require that automated systems remain transparent and controllable.
Risk management
The Citywide AI Assistant falls under the low-risk profile according to the European AI Regulation, as it does not include automated decision-making or high-risk applications. Nevertheless, specific measures have been taken to mitigate risks:
- Technical: Personal data is automatically filtered from prompts, and data is not used for model training. The self-hosted Ubuntu environment with Open WebUI minimises dependency on external parties and data leakage risks;
- Organisational: Use is limited to supporting tasks (e.g. text editing or summaries), with clear division of responsibilities: employees remain ultimately responsible for output;
- Legal: The solution complies with the AI Regulation and AVG, with review by the Data Protection Officer and CISO.
Risks are continuously monitored via logs and assessments, with any scaling up only after positive assessment of impact and compliance.
Legal basis
The AI regulation and AVG underpin this analysis along with municipal policies and guidelines.
Links to legal bases
- AI regulation: https://eur-lex.europa.eu/legal-content/NL/TXT/PDF/?uri=OJ:L_202401689
- AVG: https://wetten.overheid.nl/BWBR0040940/2021-07-01
Link to Processing Index
Impact assessment
Operations
Data
Technical design
The Municipal AI Assistant runs on a locally managed Ubuntu server within the municipal IT infrastructure, making data and processing completely under its own control. The interface is Open WebUI, an open-source platform in line with Common Ground principles. This provides employees with a secure, user-friendly environment to enter prompts and receive output. This uses the AI model Mistral, a European alternative to the other major models. Prompts are automatically filtered for personal data, in line with AVG requirements, and all interactions can only be accessed by authorised users via secure login.
External provider
Link to code base
Similar algorithm descriptions
- An assistance recipient can only apply for assistance from the municipality where the resident is registered. For example, as a result of moving house, a person may receive assistance from several municipalities. Stichting Inlichtingenbureau (IB)* informs the municipalities about this. The municipalities can then stop unlawfully ongoing benefits.Last change on 5th of August 2025, at 11:23 (CET) | Publication Standard 1.0
- Publication category
- Impactful algorithms
- Impact assessment
- DPIA
- Status
- In use
- To apply for life support, the municipality of Arnhem uses Centric's eDienst. Based on read-in data and answers provided by the applicant, the algorithm determines whether the applicant is eligible for social assistance (maintenance) benefits.Last change on 2nd of July 2024, at 9:52 (CET) | Publication Standard 1.0
- Publication category
- Impactful algorithms
- Impact assessment
- DPIA
- Status
- In use
- The Mantelzorg Assistent offers personalised support to informal carers. The algorithm collects and analyses data entered by informal carers, such as demographics and care severity, to provide targeted advice and information. The aim is to ease the burden of informal care and provide policy insight to the municipality.Last change on 24th of April 2025, at 7:34 (CET) | Publication Standard 1.0
- Publication category
- Impactful algorithms
- Impact assessment
- DPIA, IAMA
- Status
- In use
- The municipality of Asten uses Bastion365 as its solution for secure emailing. The software allows employees to send confidential and privacy-sensitive information via e-mail in a secure manner, and where necessary with MFA.Last change on 17th of October 2025, at 8:22 (CET) | Publication Standard 1.0
- Publication category
- Other algorithms
- Impact assessment
- Field not filled in.
- Status
- In use
- Municipality of Almere, together with TU/e and Fontys, is investigating how generative AI can support the assessment of supplier documentation in tenders. The aim is to make the process more efficient without compromising transparency, accuracy and safety. The project ties in with ongoing TU/e and Fontys PhD research on trust in AI.Last change on 14th of October 2025, at 12:57 (CET) | Publication Standard 1.0
- Publication category
- Other algorithms
- Impact assessment
- Field not filled in.
- Status
- In development