I want to create a ChatBot to be implemented on a website of our organization that handles customer support questions with given knowledge. I’m a beginner in working with OpenAI.
How can I make the assistant only answer the questions with given information from the uploaded files? Are there optimal ways to segment this information to make it more efficient, from a technical to an economic point of view? Can I later implement this ChatBot to a web page?
3 Likes
You can create OpenAI assistant for this. You can start with understanding how Assistant work, then move onto the Assistant API Reference.
Let us know if you have any specific question regarding Assistant that you’re unable to understand.
1 Like
I have created an assistant that answers my questions with access to given knowledge, but my problem is that it still answers questions with outside knowledge even when instructed not to. Can I implement this assistant in a web page?
Try to run the assistant in Playground and ask the assistant to give details about your knowledge base.
Yes, you can.
Thank you for your answer. I’m now trying to use the app but I keep getting errors.
The same problem persists in the Assistant, with File_scearch, it keeps answering questions outside of the files scope.
1 Like
is it working correctly in the playground or you have the same result.
If you have the same result in the playground ti means it is an OpenAI issue and we have to figure out a solution there! I had issue with this the past few days too I think
It’s not working in the Playground as well. It should be able to do as I say right? Only answers questions based on given knowledge that is.
1 Like
Yeah it should be working. It was working great before but they created Assistant V2 I don’t know if something changed in the bot behavior.
Hello Marc-Olivier
I am trying to register for the $18 plan… I guess you are French?
If you’re a programmer I recommend using LangChain in a Python app, using RAG (Retrieval Augmented Generation). Then you can make your chatbot work with ALL LLMs rather than being vendor locked-in.
I can show you my example using Gradio if you want, but I haven’t integrated RAG into it yet.
I’m not a programmer, I am a photographer trying his hand at no coding apps and assistants used as Chatbots on my various websites… This steps are easy enough, but making the chats persistent seems the most pressing issue here. It seems Threads have disappeared from the OpenAI interface?
I can see the threads on the openAI platform. It is still there, but they have moved it to a ‘tab’ at the top of the screen. It was a menu option on the side earlier, if I recall correctly.
I’m an experienced programmer, and even though I could write all the code myself, I often ask AI for help. It makes the process more efficient and usually eliminates bugs early on.
For websites, I prefer using PHP since it’s widely supported on most hosting environments. I recommend starting with GPT-4.1 via the OpenAI API — it’s powerful, and easy to work with once you get your base script in place.
For example, I had GPT-4o help me create a PHP script to connect to OpenAI models via the API. Once that was working, I used GPT-4.1 to improve the same script — adding functions and restructuring the code. It was like working with a pair programming assistant that never gets tired.
As long as you clearly describe what you’re building, the AI will help guide the process step by step. If you’re just starting out, I suggest asking it to use a modern folder structure — like having a /config folder for your API keys and a /functions folder for modular code. It keeps things clean and scalable.
And yes — once your assistant is working, you can absolutely embed it into a website.
Let me know if you want a quick starting script or folder layout — happy to help.