Munich Open Source

English

Deutsch

English

Deutsch

Appearance

Logo MUCGPT
Code
Application
Inhouse
Start 2024
AI
MIT
Logo MUCGPT
Code
Application
Inhouse
Start 2024
AI
MIT

MUCGPT is an assistant-first AI toolkit. It enables secure chat, text based task automation, and lets users create and share reusable assistants for recurring workflows.

MUCGPT focuses on assistants: reusable, workflow-specific chat configurations with their own instructions, behavior, optional tools and sharing scope. Chat is where work happens; assistants provide structure so recurring tasks become faster and more reliable. MUCGPT connects to OpenAI-compatible LLM endpoints and keeps chat history locally in the browser.

Start Page ​

Start Page A focused entry point to discover assistants, jump back into recent work and access tutorials.

Chat ​

Chat
Work with the LLM on almost any topic. Configure a system instruction to guide responses and adjust temperature to balance factual vs. creative output. Multi-turn conversations are supported; history is stored locally in the browser and can be revisited.

Create Assistants ​

Assistant Creator Design reusable assistants that encode recurring instructions, tone, constraints and model settings. Optionally add starter prompts, follow-up actions and tools. Private assistants remain yours; you decide when to share.

Share Assistants ​

Share Assistants Publish assistants to selected teams or broader audiences. Ownership, scope and configuration are transparent so colleagues understand what an assistant does before relying on it.

Additional Features ​

  • Extensible tools: examples include summarization, brainstorming and easy language; the platform is designed to grow with organizational needs.
  • Model selection: choose from available models and adjust creativity to fit the task; model choices are transparent.
  • Tutorials: guided examples and tips help users get value quickly.
  • Dark mode and i18n: accessible UI with dark mode and internationalization.
  • Output formats: correctly renders Markdown, plain HTML, Mermaid in Markdown code blocks, and LaTeX formulas ($$ ... $$).

Technical details ​

MUCGPT follows a lean microservices architecture and can be deployed as containers together with a PostgreSQL database.

  • Frontend: React-based UI (from a Microsoft Azure template), implemented with Typescript and Javascript.
  • Core service: FastAPI in Python for LLM orchestration and tools, using LangGraph to manage agent workflows.
  • Assistant service: FastAPI in Python with PostgreSQL for assistant configuration and sharing.
  • API Gateway and SSO: entry point with authentication (e.g. Keycloak). Optional observability with Langfuse; optional document parsing via Kreuzberg.

More information: https://ki.muenchen.de/ki-systeme/mucgpt