MUCGPT - A versatile AI tool for public administration

MUCGPT has four main functions:

• Chat to interact with the underlying LLM
• Summarize to condense long texts and files
• Brainstorming to create mind maps on a topic
• and Simple Language to translate texts into easily understandable language.

Additionally, MUCGPT offers the possibility to create custom assistants with specific functions to adapt the interaction with the LLM to individual needs. This allows users to customize the tool to their needs and achieve an even higher integration into their workflows.

Chat

With the chat function, users can write with the LLM on almost any topic.

MUCGPT uses a large language model and provides understandable, context-related answers. A system instruction can be prepended to any message, and the temperature can be adjusted to make the output more factual or creative.

The chat history is stored locally in the browser (IndexedDB). Conversations can be continued later, and important chats can be marked as favorites.

In addition, various output formats such as Markdown, HTML, embedded Mermaid diagrams, and mathematical LaTeX formulas ($$ ... $$) are supported.

Summarize

The summarize function allows long texts or documents to be condensed and the most relevant information extracted to create a new summarized text.

The user can adjust the level of detail of the generated response in three stages (short, medium, long).

Brainstorming

The brainstorming function allows users to create mind maps to find inspiration on a specific topic.

The created mind maps can be downloaded and further processed with Freeplane.

Simple Language

Translates complex texts into understandable language. This promotes inclusion and accessibility, as it helps people with reading difficulties or non-native speakers.

The logic is based on a project by the Canton of Zurich administration (Github).

Custom Assistants

Users can create tailor-made assistants for specific tasks. System instructions to control behavior as well as model parameters (such as temperature) can be defined. These assistants are stored locally in the browser.

Administrators can also provide global "Community Assistants" via configuration files. These are read-only, ensure standardized responses, and can be provided with example prompts.