Have us Create your Next GPT

Have us Create your Next GPT

If you wire together a Custom GPT correctly, it changes ChatGPT from being a chatbot interface telling you what to do, to an executioner that you can tell what to do. To understand the qualititative difference here, consider these simple facts.

  • ChatGPT tells you what to do
  • You tell GPTs what to do

The above of course requires an API, which we'll come to later. Ignoring the API requirements for a while, creating a custom GPT is extremely easy.

  • Go to ChatGPT
  • Click Explore GPTs in the top/left corner
  • Click the Create button in the top/right corner

Creating a Custom GPT

Below is a demonstration of the above GPT, which we're using internally in AINIRO.IO to manage leads, clients, and our back office server infrastructure.

The structure of GPTs

GPTs have a simple structure.

  • Name
  • Description
  • Instructions
  • Conversation starters
  • Knowledge
  • Capabilities
  • Actions

The name and description are self evident, so I'll start by explaining its instruction. An instruction is basically like a "job description". With it you can tell your GPT how to behave. You can create a GPT that always answers in the style of Snoop Dogg, and you can create GPTs that behaves like a school teacher for 5th graders. Below is the instruction for our GPT.

You are an friendly assistant, and you will obey by the following rules.

  • If I ask you to create a cloudlet then ask me for customer name, customer email, cloudlet name, and what plan I want to use. Once I have provided you with all this information, show me the arguments you will use, and once I confirm they are correct, proceed to use my hub-api action to create the cloudlet.
  • If I ask you to delete a cloudlet, then inform me of the name of the cloudlet you're about to delete and ask me if I am certain that I want to delete it, and that this action is permanent.
  • Display dates as MMM. d yyyy HH:mm
  • If an action returns price_reference then do not display it.
  • If I ask you to create a payment link ask me what plan I want to use. My options are 'professional' and 'enterprise'.
  • If I ask you to create a payment link, then use the payment-link action unless I ask you explicitly to send the payment link immediately, at which point you should use the send-payment-link action.
  • If I ask you to create a setup payment link, then ask me what size to use, my options are 'small' and 'large'.
  • If I ask you to retrieve plans, then use the plans action. These are subscription based plans, and the price is per month.
  • If I ask you to scrape a website or URL, ask me what URL to scrape unless I've already told you, for then to use my scrape-url action. If I don't supply a scheme then assume I want to use https.
  • If I ask you to send an email, ask me for name, email, subject and content unless I have already told you. Then display name, email, subject and body and ask me to confirm before you send the email. Use my send-email action and pass in body as Markdown. Sign the email with Tage Hansen, CEO at AINIRO.IO unless I tell you something else.

Conversation starters are just suggestions to how to interact with it.

Knowledge

This part of your GPT allows you to upload documents, which your GPT will use RAG on to extract relevant information. You can only upload 20 documents, so it's quite limited. However, with Magic Cloud you can upload up to 30,000 documents, and connect to your GPT using an API.

We don't even use these parts when we're creating a GPT, since we've got our own RAG database mechanics, far superior to the default 20 documents OpenAI gives you. Since capabilities are fairly self evident, we'll move on the the most important parts, your GPTs' API ...

Actions

With actions you can connect your GPT to an API. This is where it gets interesting, and is a requirement for creating a GPT that actually does something. In the above video for instance we've got an action to scrape a client website, in addition to another action to send emails. By combining these two together we can tell our GPT stuff such as.

Scrape xyz.com. Then create a compelling sales argument for why they need "product B". Construct an email with these arguments and send it to john@doe.com

The above command will create a personalised sales pitch email based upon the recipient's own website and send it to the client. The process is 100% automatic once you've started it by telling your GPT what to do.

The API

The API was created with Magic Cloud, and creating the scrape website and send email actions required zero coding. In the screeenshot below you can see how we did it using Low-Code and No-Code software development.

Scrape website action

Magic has hundreds of such No-Code actions you can use when composing your API. Below are some examples.

  • Read data from database
  • Send email
  • Scrape website
  • Create log entry
  • Authorize
  • Validate reCAPTCHA
  • Etc, etc, etc ...

When ever you've got an action in Magic, you can use it without coding. Below is an example of how it looks like to ask Magic to scrape your website.

Scrape website action

The above is all the "programming" you need to do to scrape a URL. The send email action is a little bit more complex, because we wanted it to use a template based email, something you can see from the design of the email in the above video. However, everything was done with a simple "drag'n'drop" interface. Below is the code required to send an email.

/*
 * Sends an email with the specified [subject] and [body]
 * to the specified [name] and [email].
 * 
 * Will use the default HTML template transforming [body] from Markdown to HTML
 */
.arguments
   name:string
   email:string
   subject:string
   body:string

/*
 * Ensures the specified [arg] was given.
 *
 * This action will throw an exception with a descriptive tex
 *  explaining if the argument is missing.
 */
execute:magic.workflows.actions.execute
   name:validators-mandatory
   filename:/misc/workflows/actions/security/validators-mandatory.hl
   arguments
      arg:x:@.arguments/*/name

/*
 * ... many more lines of code, 117 in total ...
 */

The above is 117 lines of code and was created 100% automatically by Magic without manually creating a single line of code. If you understand basic logic you know everything required to create your own custom GPT with Magic.

I created the above code in some 5 minutes and deployed it to production in 30 seconds. A traditional software developer would probably require a couple of hours to create it, and some 4 to 5 hours to deploy it to some server. This allows us to deliver something like the above for $5, while even an Indian outsourcing company would charge you $200 or maybe $500 for the same.

Too complex? Head hurts?

Have us create your next GPT

One of our services is creating APIs for GPTs and connecting these together. For a lot of things, we can deliver a custom GPT 1,000 times faster than other developers. In addition we can deliver it straight to production, managing your server infrastructure for your GPT. And we can typically deliver it 100 times less expensive. The reasons of course is Magic and our Low-Code framework allowing us to simply deliver 1,000 times faster than everybody else out there.

We're also a SaaS company, and our primary revenue comes from subscriptions. This allows us to deliver your GPT with a net loss, believing in the long term we'll make up for it. We charge $198 per month in subscription fees. This allows us to deliver a GPT for some few hundred bucks, where most others, even developers in 3rd world countries, would purely logically have to charge you thousands of dollars to even make a tiny profit.

The reasons are simple, we've got an AI-based Low-Code and No-Code software development platform, allowing us to create software 1,000 times less expensive than "everybody else out there". Hence, even though we live in Norway, which is an expensive country, we're still able to spend 0.1% of the time others needs to spend to deliver the same.

This isn't true for all tasks, but for the tasks where we've got existing actions we can combine together, we can underbid everybody on the planet, while delivering higher quality than everybody else.

Increased Quality

Yet again, we're a SaaS company, and we're willing to lose money on a single delivery. The reasons are because we believe we're able to catch up for it in the long run. This implies we're in a position where we need to deliver kick ass service, month after month, in the long run, to ensure you've got amazing quality and that you're happy with paying our hosting fees. Because if you're not happy, you can simply cancel the subscription, walk away, and leave us holding your bag.

This becomes your quality guarantee, not having to pay the full price for the project, before you've used it for a year

For a GPT such as the one demonstrated in the above YouTube video we wouldn't even charge you anything for creating the GPT itself. As long as you're paying the first month subscription, we'd throw in the "send email" thing for free. For other more complex GPTs we would charge you a minor setup fee up front, but much less than whatever everybody else would charge you.

In addition you'd get an AI Chatbot and AI Search for "free" if you chose us - It's simply included in the cloudlet we'd use to serve your custom GPT 😊

If you're interested in hearing more about our services, you can contact us below.

Thomas Hansen

Thomas Hansen I am the CEO and Founder of AINIRO.IO, Ltd. I am a software developer with more than 25 years of experience. I write about Machine Learning, AI, and how to help organizations adopt said technologies. You can follow me on LinkedIn if you want to read more of what I write.

Published 24. Jan 2024