What are ChatGPT (GPT) Wrappers?

March 28th, 2024 by Sander Schulhoff

With the release of easily usable and generally applicable AI (e.g. ChatGPT, Google Bard, Anthropic Claude), more and more companies have built applications which are entirely dependent upon these tools. Since these applications generally provide a simple layer of added code/functionality (abstraction) on top of a GenAI, they "wrap" the GenAI (which is usually a GPT-based LLM), and are thus considered GPT Wrappers. ChatGPT Wrappers are some of the most popular GPT wrappers due to the prevalance of OpenAI's ChatGPT.

What is a GPT Wrapper?

Formally, a GPT Wrapper is an application which provides a thin layer of UI/UX functionality on top of API calls to a GenAI. They are often used as standalone business products.

What is an example of a GPT Wrapper?

Consider a website which allows you to put in a description of yourself then creates a story about you going on an adventure. Behind the scenes, this application is just a prompt template which combines with your user input to produce a prompt. This prompt is then sent to a GenAI to create the story. Feel free to modify this prompt using this interactive embed:

What are the Pros and Cons of GPT Wrappers?

From a business standpoint, there are a few things to consider when creating a GPT Wrapper. GPT Wrappers can often be developed very quickly, and thus be an inexpensive way to deliver a product experience. However, the fact that they can be developed quickly means they likely lack a moat, which would prevent others from building the same tool and stealing your business. Additionally, if you get signifigant traffic on your app, the costs of API calls can become prohibitive to continuing to use it. Additionally, they may be susceptible to prompt leaking--basically when malicious users steal the prompt template that runs the app. Prompt leaking is a form of prompt injection, to which GPT Wrappers are often susceptible more broadly (see MathGPT). A final consideration to keep in mind when deploying a GPT wrapper is that you are subject to the whims of the LLM provider. For example, if you build a wrapper that allows people to talk to bots with different personalities, you would soon be outcompeted my OpenAI's GPT Store.

Pros	Cons
Quick to build	Shallow Moat
Cheap to build	Expensive to Scale
	Susceptible to Prompt Leaking/Injection
	Subject to Whims of Provider

What Classifies as a GPT Wrapper?

Some complex products, such as Cognition Labs' Devin, a semi-autonomous coding tool, have been accused of being GPT Wrappers. Devin makes use of internet search and various complicated tools, so it seems reasonable to conclude that it is not a wrapper. "GPT Wrapper"ness exists on a scale. The more an app relies solely on a GenAI, the more it is a GPT wrapper. However, the more it introduces other tools and complexities, it becomes less of a GPT wrapper. We generally judge apps to be a GPT wrapper if their entire functionality can be abstracted to a few GPT API calls.

Are GPT Wrappers Dead?

We estimate that 10s of thousands of GPT Wrappers exist, a small fraction of these being monetized. As OpenAI and other providers improve the tooling around their GenAIs, adding in function calling and internet searches, the need for these wrappers will decrease. However, currently there is still signifigant functionality that GPT Wrappers can provide, so we believe people will continue to build them for the next few years.

Final Thoughts on GPT Wrappers

GPT Wrappers can be useful when combining GenAIs with certain tools or highlighting certain functionality of the GenAI (e.g. story telling). However, due to their shallow moat, and the ever accelerating pace of GenAI companies, they may not last long. When considering whether to build a GPT Wrapper, make sure to keep in mind our Pros and Cons list above. Happy Building!

Cite as:

@article{Wrapper2024Schulhoff,
  Title = {What are ChatGPT (GPT) Wrappers?},
  Author = {Sander V Schulhoff},
  Year = {2024},
  url={https://learnprompting.org/blog/2024/2/4/gpt_wrappers}
}