How exactly does File Search work in ChatGPT? Is it always necessary to enable it?

[lehieu76]

Hi everyone,
I’m using ChatGPT Asistant to analyze around 30 files (PDF, DOCX, etc.) and I have some questions regarding the File Search feature:

1. When I don’t enable File Search, ChatGPT still responds very quickly and accurately, even though I uploaded many files. Why is that? Does ChatGPT pre-read and temporarily store the file content?
2. On the other hand, when I enable File Search, the response is noticeably slower. In which cases should I enable File Search, and when is it better to keep it off for optimal performance?
3. If I only need overall analysis (not quoting specific sections), is turning off File Search limiting in any way?
4. With a large number of files (30+), will ChatGPT exceed the token limit in context if File Search is off?

I’d love to understand more about the actual mechanics behind File Search to use it more effectively. Any technical insights or real use experiences would be greatly appreciated.

[_j]

ChatGPT is not the API.

ChatGPT is OpenAI’s product at chatgpt.com

There’s enough here that I think you indeed might be talking about the API, and internal tools of the “Assistants” API endpoint.

If you don’t have a tool enabled, the the AI isn’t going to spend time writing a tool call and then waiting for a search to be performed.

Your question, I answered pretty elaborately yesterday.

File_search with max num results

You have a good question, as this file_search tool and the vector store database is offered without a whole bunch of information about the applications it is useful for. However, the application you might be thinking about is a bit unclear, but it sounds like this might not be the right solution for you. The way it works: Documents are uploaded to the file storage. Then you create a vector store. One of the parameters of a vector store is how many tokens that the data resulting from document…

[lehieu76]

I’m working with OpenAI Assistants via this link: https://platform.openai.com/playground/assistants and I’m a bit confused about when I should enable or disable the File Search feature.

Here’s what I’ve experienced:

• When I ENABLE File Search, the assistant’s responses become very slow — sometimes it feels like the system is lagging or freezing.
• When I DISABLE File Search, responses are much faster — and it still seems like the assistant can access the uploaded files and use the internal data I’ve provided.

→ So here’s my key question:
Can I keep File Search OFF for performance, even when dealing with large sets of files (e.g., 20–30 documents)?
Or will I eventually need to enable it to avoid token limits or data loss during retrieval?

I’m new to the Assistants API and would appreciate any clear explanation about how File Search actually works in this context, especially regarding efficiency and best practices.

Thanks in advance!

[_j]

The AI models are pretrained on vast knowledge. They can be a non-action-taking customer service agent for Verizon without the additional quality that would come from database knowledge being searched.

They are very good at fooling people with their emulation of human languages, as spoken with knowledge backing it. You’d really need document information that an AI with a training cutoff mid-July 2024 cannot know or predict.

Off is simply off, the vector store database and the file knowledge is not used. Or else you would have found a bug in the user interface.

When the AI searches the database instead of directly responding to you, you will also see higher input token costs to go with the additional latency - the amount of documentation retrieved from file chunks.