Virtual assistant
Bürokratt
Bürokratt is a network of chatbots on public sector institutions’ websites that enables people to obtain information from institutions and to use both public and information services via virtual assistants.
What is Bürokratt today?
Bürokratt is a state-created, AI-based digital assistant that helps institutions deliver modern and efficient customer service. Today, Bürokratt offers institutions the ability to use large language models (LLMs) to understand human-language input and respond to questions quickly, referentially, and around the clock.
Who is Bürokratt for?
Bürokratt is intended for people who want to communicate with the government. It is designed for public sector institutions that wish to offer user-friendly, efficient, and up-to-date customer service. It meets the needs of both local governments and national agencies — including managing information requests, automating recurring questions, and providing customer support.
Vision of Bürokratt
The vision for Bürokratt is to give people the opportunity to interact with the state as easily as with a friend — in natural language, at any time, and from any device. To achieve this, the chatbots deployed to clients will form a unified network that allows people to find answers, solve problems, and consume digital services directly from the chat window, without navigating complex portals or filling out forms.
As this vision is realized, institutions will be able to integrate virtual assistants into the network — independent AI agents that cooperate with each other, which, based on a person’s requests, find necessary information and carry out required actions (e.g. submitting an inquiry, application, or booking). In the future, these assistants will be able to offer people seamless and personalized service.
Bürokratt roadmap
What will be done by the end of 2025?
By the end of 2025, a functional foundation for broader adoption of Bürokratt as an AI-based customer service network will be established. Over the year, the deployment of a large language model (LLM / RAG) will be carried out, a general knowledge module will be developed, and a central component (global classifier) will be created, through which different Bürokratts can communicate securely with each other.
What will be done in the future?
Starting in 2026, Bürokratt will move toward a model where each institution or domain can use its own personalized AI agent as part of a unified and cooperative network of agents. The focus will be on building an LLM adapted to Estonian, enabling agents to communicate independently, and utilizing data securely and intelligently. The goal is to make public sector services more accessible and human-centered.
What does Bürokratt enable?
Bürokratt is built for public sector institutions that need to offer modern, consistent, and efficient customer service. This includes national agencies, local governments, and other public organizations that serve citizens or businesses. Bürokratt is suitable wherever people’s questions must be answered quickly, clearly, and at scale — whether for information requests, recurring questions, service routing, or supporting administrative processes.
What to do if you’re interested in Bürokratt?
The software for Bürokratt is free, but hosting the solution is done, as usual, in the State Cloud, with a monthly hosting cost of about €150, plus additional costs from using large language models (LLMs). For institutions, adopting the solution requires contributing to the collection and, if necessary, cleaning of input data. The Estonian Information System Authority (RIA) offers support from the first demo to production. See the detailed LLM guide here.
​
To get started, contact RIA at help@ria.ee
Who uses Bürokratt?
Several public sector institutions already use Bürokratt. Earlier solutions, such as Rasa-based chatbots, relied heavily on pre-defined question-answer training, which required a lot of resources from institutions. While those solutions worked reasonably for recurring questions, they demanded constant manual maintenance and content updates.
Today, many institutions are transitioning to large language models (LLMs), which can understand the semantic content of questions and use institution-defined sources to find information without requiring separate training. The first to adopt an LLM solution was the electronic identity department of RIA (via id.ee), where Bürokratt can support end users in real questions without manual training — meaning better answers and less manual upkeep.