v1.2.2

Added

• Own assistants #155

  • It is now possible to create custom assistants. This feature allows users to develop specialized assistants for recurring tasks, equipped with a system prompt.
  • Examples of assistants include:
    • English Translator: Translates all inputs into English.
    • Test Generator: Creates helpful test cases based on the entered program code.
    • Proofreader: Corrects entered texts and suggests alternative phrasings.
  • To create an assistant, the user describes the desired functionality in a text field. MUCGPT then generates an appropriate title, description, and system prompt, which can be further customized.

• French and ukrainian localization in #154

Fixed

• Backend statistics can be now queried again: #152
• Various frontend bugs #151

Changed

• The design of the MUCGPT user interface has been updated to be more clean

Full Changelog: v1.2.1...v1.2.2

v1.2.1

Added

• redirection option for unauthorized users by
• reproducable dependency managment with uv insead of plain requirements
• translation into plain and easy language based on the awesome work of https://github.com/machinelearningZH/simply-simplify-language

Fixed

• performance problems in long chats

Changed

Full Changelog: v1.2.0...v1.2.1

v1.2.0

Added

Fixed

• brackets are no longer swallowed when rendering code

Changed

• switched from Quart backend du Fastapi

v1.1.4

Added

• Maximum tokens from the config divided into input and output tokens. This means that models with smaller context windows (e.g. Mistral) no longer run into an error.

Fixed

• Version number is saved correctly again and displayed in the settings.
• Max_tokens and other parameters are now correctly set for the non-default LLMs

Changed

v1.1.3

Added

• Support for multiple models via configuration on front and backend
• token counting for Mistral and newer OpenAI Models by using the correct Mistral/Tiktoken Tokenizer

Fixed

• #41
• improvements for summarisation logic (#40)
  • using structured output/Json mode -> no need for complex cleanup after the LLM call
• Removed depreciated LLMChain logic

Changed

Docker instead of Azure dependencies

• Docker based build based on a json-configuration
  • contains a list of available OpenAI/Azure-OpenAI compatible endpoints
• Removed all IAC files for azure, as well as all azure dependencies

v1.1.2

Added

chat history

There is now a history of all conversations that have taken place for the chat function, that is stored inside the browsers IndexedDB

• All chat histories in the "Chat" tab are saved automatically
• Chats can be deleted, renamed or favoured in the "History" window
• Favourite chats are always displayed at the top
• Chats are sorted according to the last time they were edited and are
  grouped into "Today", "Yesterday", "Last 7 days" and "Older"

Fixed

Changed