Umi is a fully open-source LLM user interface, designed to unleash the power of prompt engineering with existing tools integration.

Main project goals

  • Integration of tools eg. terminal, web browser
  • Implementation of the newest prompt engineering solutions
  • Specialized user interface for specific tasks
  • Easy to use, deploy and configure

State of the development

Application is in prototyping state and has no installer. The concepts of how to exactly combine LLMs with external tools, use memory or command the prompts is in question. The prototype version offers couple of prompt plugins and simple chatbot usage of LLMs using ChatGPT servers.

See the source-code

More about the project

Large language models get better and better over time. With the progress in model accuracy, rises more and more practical usages, but at the same time, the complexity of implementing such systems prevents fast growth in AI. To combat this, we need a new framework that makes standard procedures we do with AI like generating files, decision making etc. much easier. Like with the prompt engineering methods rising, the number of statistical methods that work on low-level information also rises, but there's no platform that would implement those algorithms in some higher abstraction. We're still on the beginning of the process of understanding the new AI, the one that can generate meaningful responses. There are no beaten paths when it comes to solving many problems that have an root in connecting the classical programming with LLM. One of the keypoints of the project is to find more solutions to those problems, by creating a programming environment for exploring those issues.

Main concept

Application uses python scripts to implement tool usage and more sophisticated techniques of controlling the model. By solving more and more practical problems author believes that it will be possible to create an self-defined API for working with models.