ProToPortal.jl

Welcome to ProToPortal, the portal to the magic of PromptingTools.jl and Julia-first LLM coding! ProToPortal is a personal project designed to enhance productivity, potentially yours too!

Given it's a UI-rich application, it will contain many bugs I'm unaware of—let me know!

Scroll to the bottom to see some screenshots.

Warning

This application is still in development. Use at your own risk.

Getting Started

This guide assumes that you have already set up PromptingTools.jl! At the very least, you need to have OPENAI_API_KEY set in your environment.

1. Clone the repository
2. Run the following command to install the necessary dependencies:

using Pkg; Pkg.activate("."); Pkg.instantiate(".")

3. Launch the GUI

# as a quick hack if you don't have your environment variables set up, run the below line with your OpenAI key
# ENV["OPENAI_API_KEY"] = "<your_openai_api_key>"
ENV["DATADEPS_ALWAYS_ACCEPT"] = "true"  # required for caching
using ProToPortal
launch(; cached = true)

Then head to your browser and go to http://127.0.0.1:8000 to see the app.

It will now cache similar LLM requests by default (disable with cached=false in launch() function).

For the purists: simply run julia --project -t auto main.jl in your terminal (once installed)!

How to start? Type Say hi! in the question box on the Chat tab and click Submit (or press CTRL+ENTER).

Preview:

Motivation

Imagine you're walking your dog and suddenly come up with a brilliant idea for a code snippet you need urgently.

Using ChatGPT on your phone is feasible, but slow and cumbersome—typing out or dictating a precise prompt can be a real hassle, not to mention you can't close the app or run the code to see if it works.

Enter ProToPortal: Select the JuliaExpertAsk template, use speech-to-text on your phone to dictate your needs, enable auto-fixing with Monte-Carlo Tree Search (airetry!), and watch as your phone processes 6 iterations of the code before ChatGPT generates its first response!

For a preview, see the video: docs/src/videos/screen-capture-code-fixing.mp4

Key Highlights

• Accessible Anywhere: Use it on any device—laptop, phone, etc.—with numerous prompt templates designed to save your keystrokes.
• Code Evaluation and Fixing: The only GUI that lets you evaluate and fix your Julia code directly, saving you from the hassle of copy-pasting.
• Automatic Replies: Automate standard replies to refine responses, either through fixed messages or critic templates mimicking a separate agent.

Features by Tab

Login Page

• Comes with a simple login if you want to deploy it (Fly.io works like a charm!)
• Ignore the password when deploying locally (enter an empty password).

Chat Tab

• Advanced Settings:

  • Rewind conversations by deleting messages you dislike or to restart the conversation.
  • Control creativity with model temperature settings.
  • Enable code evaluation (AICode) and auto-fixing using advanced algorithms like airetry! and Monte-Carlo Tree search.
  • Automate replies with pre-configured templates (~critic agents) or custom messages.

• Templates:

  • Begin conversations from a template, opening new input fields interpolated into text variables.
  • Add your templates by placing them in the templates/ folder for automatic loading.

• Chat:

  • Submit your question by clicking "Submit" or via CTRL+ENTER.
  • Edit any message in the conversation (simply click on it and make a change).
  • Delete unwanted messages with ease (accidental deletions made difficult by hiding the button in the "Advanced Settings").
  • Copy the text of any message (see top right corner icon).
  • Start "new chat" with conversations automatically saved both on disk and in history.

History Tab

• Browse and load past conversations with a simple click.
• Fork past conversations for continued exploration without altering the original history.
• View the current session or reload to fetch all saved conversations.

Templates Tab

• Explore all available templates with previews and metadata to select the most suitable one.
• Search functionality for quick filtering by partial names or keywords.

Configuration Tab

• Change the default model or add new ones from PromptingTools.
• Modify the default system prompt used when not employing a template.

Meta-Prompting Tab

• An experimental meta-prompting experience based on arxiv.
• The model calls different "experts" to solve the provided tasks.

Prompt Builder Tab

• Generate prompt templates (for use in Chat) from a brief description of a task.
• Generate multiple templates at once to choose from.
• Iterate all of them by providing more inputs in the text field.
• Once you're done, click "Apply in Chat" to jump to the normal chat (use as any other template, eg, fill in variables at the top).

And rich logging in the REPL to see what the GUI is doing under the hood!

Screenshots

Very simple chat interface, easy to use on the phone!

Many templates are available, allowing you to save keystrokes.

Advanced settings - evaluate code, automatically critique or fix your Julia code.

Alternatives

ProToPortal is a simple personal project and it cannot compete with established LLM GUIs!

If you're looking for more robust software, consider the following tools:

• Open Webui: https://github.com/open-webui/open-webui
• Oobabooga's text generation webui: https://github.com/oobabooga/text-generation-webui
• Simple server in llama.cpp (for the purists)

Contributing

Contributions are what make the open-source community such an amazing place to learn, inspire, and create. Any contributions you make are greatly appreciated. Please open an issue first, because this is first and foremost a simple tool to interact with LLMs on-the-go.

License

Distributed under the MIT License. See LICENSE for more information.

Acknowledgments

This project would not be possible without the amazing Stipple.jl from the Genie.jl family! It's just a Stipple.jl wrapper around PromptingTools.jl.