But wait! Don’t you guys already have the OpenAI plugin to help with development and all that?

Yes, but this stuff will take you to a whole new level. Imagine no-coding your project without even trying to piece logic cubes together and just having the AI sort out what you want in Directual—based on what you said to it in plain text. 

That’s what we’re talking about.

What is Directual-GPT (D-GPT)?

This is a ChatGPT plugin similar to a ReAct agent, capable of processing and calling openapi methods with predetermined parameters in mind. It can also validate data and request extra parameters based on openapi documentation. 

Too hard/don’t care? That’s okay. Basically, if you tried ChatGPT, you know that it has a chat interface that performs tasks you want it to do. It’s the same thing, but in Direcutal, for Directual projects. 

How does it work?

Like magic. However, there are (of course, the catch!) some limitations.

Since ChatGPT sometimes cannot generate complex requests with method descriptions over 320 characters, certain endpoints send the original command to create a scenario within our own API. 

Then, we launch our own directual-ai module trained on non-identifying user data and a text description. As a result, we get domain-specific language instructions, which are then transformed within our backend into Directual logic instructions. 

Why?

Oh, it’s nothing really…other than the potential to export your Directual project in any programming language.

Here’s a handy scheme of how data is processed within this miracle plugin:

USER REQUEST > CHAT_GPT > Directual API BACK > Directual AI > CHAT_GPT > USER

We’ve tested, trained and included into D-GPT the following models to determine the ideal environment:

• LLAMA, as a first test subject for logic cube generation and local launch. This version, however, isn’t great for commercial use, although LLAMA2 is. It’s too slow but performs the tasks well and does not require a GPU to perform.
• Databricks’ Dolly, a model that is easy to train but it’s way too basic in its base form. It’s well-documented and has a lot of tools to make your life easier, though.
• Alpaca, a mirror copy of Dolly, with all the pros and cons of it.
• Phi-1, a seemingly ideal model trained in Python and specialized in code generation. This is the model currently in use in our R&D department. 

What is D-GPT capable of?

A whole bunch of things, namely:

• Create API endpoints that you can fine-tune to your liking
• Create Web pages to display said API endpoints
• Create Directual scenarios using logic cubes 
• Create web-pages with components linked to an API endpoint.

Here’s what it looks like in action:

Afterword

Want to learn more how this wonder-plugin works and how it can help you build no-code apps even faster? Shoot us a message at hello@directual.com, or better yet, head to our communities—links in the footer below.