Is Context Isolation in the API Possible?

API
1

Hi,

I’m trying to build an AI chatbot for an internal use web application, something similar in functionality to Intercom’s AI Agent (Fin). At the same time, this will be a good practice for some of my skills.

I want to validate whether my approach is correct and also get the community’s feedback on parts I haven’t thought through.

Tenant (User) Creation

  1. User signs up
  2. I assign them a unique tenant_id

Providing Knowledge to their AI Agent

  1. User uploads PDF or provides raw text
  2. If PDF, I extract the text from the PDF
  3. Generate embeddings using OpenAI’s Embeddings API
  4. Store the embeddings for each tenant in a vector DB such as Supabase Vector DB

User Interaction

  1. User types in their question to the chatbot from their own web app
  2. The widget sends the query and the tenant ID to my backend
  3. The backend queries the Vector DB to get the embeddings
  4. Then, using the embeddings as a foundation I call the actual ChatGPT API
  5. I forward the response to the user

I’m not looking for something overly complex, I am trying to get an MVP up and running, I just want to understand whether I’ve thought this through.