/query endpoint not returning conversational text

[Katlego-Chagane]

The APP is deployed with Digital Ocean with Pinecone as a vector database. When the following is performed:

{
   "queries":[
      {
         "query": "What is banana",
         "top_k": "1"
       
      }
   ]
}

The results returned are not conversational from the OPENAI perspective when a POST request is sent to the endpoint; the /query only shows results that are similar to the search query.

What could be the issue?

[akrabulislam]

same to me

[sbc-max]

@Katlego-Chagane That's exactly the intended behavior of this plug-in/API. The endpoint is intended to be called by ChatGPT when it needs information about a given topic as defined by your plug-in. ChatGPT will then use the response from the query api to create a conversational response to the user within the chat window.

[JanHorcicka]

@Katlego-Chagane That's exactly the intended behavior of this plug-in/API. The endpoint is intended to be called by ChatGPT when it needs information about a given topic as defined by your plug-in. ChatGPT will then use the response from the query api to create a conversational response to the user within the chat window.

Sorry, but can you clarify this? What's the use case of the query API if it only returns the top 3 embeddings and doesn't call ChatGPT? Is this package supposed to be part of some larger project using for example langchain?

[pb30]

This isn't a chatbot. This is an example plugin for ChatGPT. When a plugin like this is enabled, ChatGPT will determine if/when to send the end-users query to the /query endpoint. This plugin will return relevant document data to ChatGPT, and ChatGPT will then formulate a response.

If you want to make your own custom chatbot using similar methodology to this (embeddings and vector DBs), look at LangChain to get started.

[JanHorcicka]

This isn't a chatbot. This is an example plugin for ChatGPT. When a plugin like this is enabled, ChatGPT will determine if/when to send the end-users query to the /query endpoint. This plugin will return relevant document data to ChatGPT, and ChatGPT will then formulate a response.

If you want to make your own custom chatbot using similar methodology to this (embeddings and vector DBs), look at LangChain to get started.

I thought this is an API for the ChatGPT retrieval plugin, but looks like I was mistaken. So what this package does is that it enables very easy document upload. But how do I combine this with ChatGPT? Let's say I want to create a ChatGPT powered app that allows retrieval. My understanding is:

• Have this package running on a server.
• Create another app that is calling ChatGPT
• Decide when to call the query API from the other app??? Thanks for any help.

[pb30]

This package includes APIs for you to add documents to your vector database, as well as the ChatGPT API endpoint for the plugin. No other custom apps are needed for this.

Are you looking to use ChatGPT (chat.openai.com) or just GPT API on your own site/app? They are separate, this is for ChatGPT Plugins. What's your end goal?

[JanHorcicka]

This package includes APIs for you to add documents to your vector database, as well as the ChatGPT API endpoint for the plugin. No other custom apps are needed for this.

Are you looking to use ChatGPT (chat.openai.com) or just GPT API on your own site/app? They are separate, this is for ChatGPT Plugins. What's your end goal?

Oh yeah, then I got it all wrong. I am planning to build an app using the ChatGTP API. I don't care about the ChatGPT UI at chat.openai.com. I thought this package is a supplement to langchain, but looks like I was mistaken. Thanks for clarification.