Agentic AI

Agentic AI understands not only individual questions but also the underlying intent and context. It autonomously selects suitable data sources, search methods and tools, performs multi-step tasks and presents results in a format tailored to the specific use case.

This makes it possible to research information, analyse documents, provide access to administrative knowledge and deliver interactive information services for citizens and employees.

Retresco provides a ready-to-use technology platform for agentic AI applications in the public sector. The AI system analyses search intent, orchestrates internal and external information sources and delivers precise, real-time streaming responses.

Results can be presented as text, lists, tables or structured datasets – including source references, timeframes, filters and relevant metadata.

Retresco serves federal, state and local government authorities, as well as public enterprises, social insurance institutions, chambers, associations and organisations serving the public interest.

Its AI solutions are particularly suitable for organisations seeking to make extensive or distributed information repositories accessible in a secure, transparent and user-friendly way.

Potential AI applications include:

• Internal knowledge assistants for employees
• Citizen and service portals
• Interactive administrative information services
• Research across documents, guidelines and regulations
• Assistance with applications and administrative procedures
• Information services covering elections, traffic, events and local government services
• Analysis of extensive document collections and datasets

Applications can be made publicly accessible or restricted to specific user groups.

No. Retresco supports the retrieval, preparation and provision of information but does not automatically make sovereign or legally binding administrative decisions. Government agencies and organisations define which tasks AI is permitted to perform, where human review is required and when enquiries must be passed to responsible employees or specialist systems.

This enables agentic AI to accelerate routine tasks and reduce employees’ workloads while responsibility, decision-making authority and specialist oversight remain with the respective organisation.

Conventional RAG systems often search a single dataset and provide retrieved content to a language model. Retresco goes further.

Agentic AI analyses search intent and conversational context, dynamically selects suitable data sources, search methods and tools, and can perform multi-step tasks. This enables more precise, up-to-date and better-structured answers – even when dealing with complex and distributed information repositories.

Each information source is individually connected, structured and made accessible through suitable search and retrieval methods. These can include conventional full-text search, semantic search, vector search, filters and structured queries.

For each query, the AI system determines which sources and methods are required. The results are then brought together into a coherent and traceable answer. If a query is unclear, the AI application can ask targeted follow-up questions.

Retresco can process structured and unstructured information from different systems, including:

• Document management systems
• Intranets and knowledge databases
• Specialist systems and databases
• Websites and content management systems
• Guidelines, regulations and administrative provisions
• PDF, XML, JSON and HTML documents
• Internal and external interfaces
• Up-to-date data sources covering traffic, elections, events and similar topics

Yes. Retresco can connect internal documents, specialist data and knowledge repositories with third-party systems and verified external information sources.

Permissions, user groups and predefined rules determine which sources may be used for a particular query. The system then dynamically selects the most relevant information sources.

Depending on the source, new and updated content can be processed in real time, near real time or at predefined intervals.

Timestamps, source references and other metadata provide transparency about how current the information underlying an answer is. This makes the applications suitable for topics where up-to-date information is essential.

Government agencies and organisations retain full control over their information resources. They determine which sources are connected and approved for individual applications, how content is prioritised or weighted, which user groups are granted access and which AI models may access specific data. Output formats, source citations and response rules can also be configured precisely.

Retresco provides dedicated content pipelines, indexes, access control mechanisms and endpoints for this purpose.

Retresco can be integrated into existing websites, portals, specialist applications and AI platforms through standardised interfaces. Integration via the Model Context Protocol (MCP) is also possible.

Core functions such as intent recognition, data preparation, retrieval, source orchestration and structured output are available as a service. This eliminates the need to develop a dedicated search, data and agent infrastructure from scratch.

No. Retresco can be integrated as a technical AI service into existing applications or used with a ready-to-deploy user interface.

The responsive widget can be embedded in websites, intranets, citizen portals, apps or restricted-access areas. Its design, suggested starter questions, notices and functionality can be tailored to the respective service.

Retresco is fundamentally model-agnostic. The prepared information can be provided as context to different language models, including those from OpenAI, Anthropic, Mistral, EuroLLM, Teuken and other providers.

The choice of model can be based on requirements relating to data protection, hosting, performance, costs and the intended application.

Answers are delivered through real-time streaming and formatted to suit the specific use case – for example, as continuous text, summaries, lists, step-by-step instructions or tables. Source references, timeframes, filters and structured datasets for downstream systems can also be included. This allows results to be displayed directly or processed automatically.

Retresco enables organisations to systematically analyse usage, answer quality and information needs. Integrated analytics reveal which questions are asked most frequently, which topics are relevant, which sources are used and where information is missing or satisfactory answers cannot be provided.

These insights can be made available through a dashboard or via an interface for downstream analytics and reporting systems. This enables datasets, response logic, suggested starter questions and information services to be continuously refined according to user needs.

Internal information repositories are processed through dedicated content pipelines, indexes and endpoints. Data and applications belonging to different customers remain technically and logically separated.

Access rights can be managed according to user groups, roles, information sources and use cases. Organisations retain control over which data an AI model receives and which content may be used in its answers.

The specific technical architecture and data protection arrangements are tailored to the requirements of each organisation.

The infrastructure is designed to handle extensive information repositories and high query volumes. Even large datasets and substantial document collections can be made searchable and accessible through interactive services.

Data sources, indexes and search processes can be scaled independently. Retresco can manage the setup, operation, updating and ongoing development of the technical infrastructure.

Agentic information services powered by Retresco can be tailored to existing administrative structures, specialist systems, datasets and security requirements. Federal administrative structures can also be accommodated – for example, through separate data environments, specific areas of responsibility, tiered access permissions and different sources and response logic for federal, state and local government bodies or individual agencies.

Organisations can start with a clearly defined use case and gradually expand the service to include additional administrative departments, user groups, data sources and features. Response formats, source prioritisation and connected AI models can also be configured to meet specific professional, organisational and regulatory requirements.