Skip to content

Open-source project for running LLM agents in a user-friendly desktop app

License

Notifications You must be signed in to change notification settings

JohannesVC/SuperModels

Repository files navigation

__SuperModels

SuperModels is an open-source project for running LLM agents in a user-friendly desktop app.

Not a llama

Give it a try! Download the executable here!

purpose:

A self-reflecting machine:

  • Think Wittgenstein not Frankenstein
  • Built-in reflection mechanism to steer thinking
  • Fully customisable

__in short A lightweight, simple, and easy to adapt Agent.

Stack: Python, Electronjs, and TailwindCSS.


walk-through:

1. SuperModel

  • I implemented a series of powerful LLM APIs for you to play with.
  • Most notably, Groq which is faster than any other LLM.

2. ToolUser

The tools are basic but performant:

  • The python_wizzard generates and executes Python on your system. Note that running generated code is not without risk. - Be prepared for SuperModels to take control over your OS and start operating your home automation.
  • A second tool, online_llm uses Perplexity AI under the hood for natural language web results.

3. Agent

The real magic happens here (click for source code):

  1. This is where the agent breaks down complex problems into steps:
break_down = BreakdownReflection(prompt)
  1. Every step is then passed to the tooluser:
for step in break_down.steps:
    answer = tooluser.call(step)
  1. The agent reflects on the tooluser's answer, and whether it should break it down further:
is_satisfying = YesNoReflection(prompt, answer)

In short: The combination of a super fast tooluser and nominally typed answers allows for quite impressive reflections and reasoning. Although I still haven't got it to turn the lights off.


instructions

Give it a try! Download the executable here!

Then:

  • Join the FREE beta at groq.com to get a free API KEY
  • Input the API key when the app prompts you