Developing GPTChat Plugins: Technical Overview and Requirements Guide

The emergence of GPTChat plugin development has ignited enthusiasm within the community, providing significant opportunities for the growth of GPTChat, its ecosystem, and the industry as a whole. At AXIOMA, we have conducted a technical analysis of ChatGPT plugin documentation and are preparing our prototype. Through extensive study, we have obtained key insights into how GPTChat plugins work and their technical requirements. In this article, we will explore the key aspects of GPTChat plugins, including the structure, hosting requirements, plugin submission, and review process.

By delving into the technical intricacies of developing GPTChat plugins, businesses can leverage AI-powered conversations, deliver personalized experiences, and reshape customer interactions. Join us on this journey as we explore the potential of conversational AI and drive innovative advancements in the industry.

Furthermore, according to Microsoft’s statements, GPTChat plugins are expected to be compatible with BingAI and Microsoft Copilot, which will be featured in Windows 11. This compatibility opens up exciting possibilities for further integration and synergy with Microsoft’s ecosystem.

Required structure of GPTChat Plugin

GPTChat plugins consist of two essential components: the Manifest File and the OpenAPI-format API Description.

The Manifest File

The manifest file is a structured JSON document that contains crucial information about the plugin. It includes parameters such as:

  • the plugin’s name (name_for_human and name_for_model),
  • description (description_for_human and description_for_model),
  • version (schema_version),
  • image location (logo_url),
  • plugin usage policy (legal_info_url), and
  • developer contacts (contact_email),
  • Block with different authorization schemes (auth)
  • Block with OpenAPI file (api).

Notably, there are two parameters for description and name: “description_for_human” and “description_for_model” as well as “name_for_human” and “name_for_model“. It’s important to provide both descriptions in a way that is understandable to both humans and the language model. Furthermore, it is considered best practice to write these descriptions neutrally, avoiding any attempts to persuade or encourage plugin usage.

The Manifiest file

OpenAPI-format API Description:

The OpenAPI-format API description defines the endpoints and their parameters for seamless communication between the plugin and the company’s website. It’s important to note that within this framework, the language model autonomously determines when to make requests to the plugin’s provided API.  Although developers or users can structure their queries to increase the likelihood of the model using a plugin, ultimately, neither the developer nor the user can force the model to use a plugin to process a particular query.

OpenAI recommends using a minimal number of endpoints (typically 2-3) with simplified parameters to ensure simple and easy integration. By streamlining the endpoint design and parameter configuration, the overall integration process is enhanced, facilitating smooth communication between the plugin and the underlying chat system.

Hosting the Manifest File

To meet OpenAI’s requirements, the manifest file should be hosted on the company’s website in a specific directory called “.well-known“. This directory ensures that the manifest file resides on the same domain as the user-facing information, including images and contact details, as well as the API. Hosting the manifest file in the “.well-known” directory creates a cohesive and integrated experience for users.

Plugin Submission and Review Process

The plugin submission and review process is a crucial step in making GPTChat plugins accessible to all users. Once the manifest file is ready, it is submitted for review to OpenAI’s dedicated bot, the “plugin submission bot”. This process involves meticulous evaluation, including potential manual checks, to ensure compliance with guidelines and quality standards. The review typically takes approximately one week.

If any changes need to be made to the plugin, such as updating the description or modifying the API authentication method, the review process needs to be initiated again. This means that the plugin file must be resubmitted for review by OpenAI. This ensures that any updates or modifications meet the requirements before the plugin can be approved and made accessible to all users of the GPTChat platform.

Upon approval, the plugin becomes accessible to all users of the GPTChat platform, expanding its reach and potential impact. Without approval, access is limited to a group of 15 developer users.

Limitations of GPTChat Plugins

When developing and integrating GPTChat plugins, it is important to consider certain limitations set by OpenAI. They can impact the chat window size (avalible tokens in the contextual window), the number of plugins, response time, and the model’s preprocessing of API responses.

  • Window Size: Contextual window size is an important consideration when integrating plugins into GPTChat. While newer models have increased the window size, integrating plugins can reduce the effective size, potentially impacting conversational accuracy and coherence. It is crucial to find a balance between incorporating additional functionalities and maintaining the model’s ability to understand and stay on topic.
  • 3 Plugins Limit: Currently, GPTChat supports a maximum of three active plugins simultaneously within the chat interface. This limit ensures a balanced and optimized user experience. Users need to prioritize the most relevant plugins for their specific use cases.
  • Response Time: It should take no more than 45 seconds from the moment you send a request to the plugin until you receive a response. If the response time exceeds this limit, the request will be rejected due to a timeout. It is worth noting that OpenAI plans to further tighten this requirement in the future, emphasizing the importance of prompt responses.
  • Model Preprocessing of API Responses: The GPTChat model can transform API responses according to its interpretation. For example, if a user requests a table, the model will attempt to generate a table-like structure even if the API response was in the form of a list.

By considering these limitations during plugin development, businesses and developers can ensure compatibility with the GPTChat platform and optimize the user experience within the defined constraints. It is essential to keep abreast of any updates or changes to these limitations as OpenAI continues to refine the capabilities and requirements of GPTChat plugins.

Conclusion

By understanding the structure, hosting requirements, and submission process of GPTChat plugins, businesses and developers can ensure successful integration and improve the user experience. Adhering to OpenAI guidelines and limitations and providing accurate information in the manifest file and API description is crucial for seamless integration. GPTChat plugins offer an opportunity to leverage the power of conversational AI and transform user interactions.

By embracing the knowledge shared in our article, businesses can gain a deeper understanding of GPTChat and its potential applications, empowering them to make informed decisions and maximize the benefits of conversational AI for their users and their growth.

Further reading
Back to All Articles