How can I add web browsing ability to my API assistant?

I need some hints and keywords to search how can I add web browsing ability to my API assistant please

1 Like

The first thing you’ll need to do is identify an API that can do internet searches and return results, ideally with long descriptions that can alone be satisfactory.

Then integrate that with the function-calling ability, writing a function tool that offers the ability to search, and refining the instuctions to let the AI know what it does and doesn’t know needing the search.

Also, avoid Microsoft search products, because just like they are not above hijacking people’s OSs to pop-up ads for their own paid AI product, they also will redirect you to their own services even when the AI has not written search terms instructing that:

Real “web browsing” is troublesome, because modern web pages are hard to scrape. Keyword for you: Selenium.

1 Like

“It seems that I have to define my own API to call Google search and then return the results to the assistant for processing, right? Looking for clarification.” :thinking:

@1121041505 Right, you need to define the function specs in the tools section in the assistant specs so the assistant will request it when needed and will return the parameters you need to make your own API call for web search then feed the assistant with the output of this call so he can proceed to generate the response

“I just found a plugin called myfiles_browser; it has a function ‘open_url’. Can it achieve this functionality?” :grimacing:

@1121041505 I think there is no relation between assistants and plugins. The assistants work totally outside chatgpt

The API CANNOT access the internet.

myfiles_browser is part of the undisclosed tools used by the AI for document retrieval when it is enabled in assistants. It is for searching files uploaded by the user to assistants (and GPTs).

OpenAI thought they could hide the tedious exploration methods of retrieval by loading them dynamically only out of user input context…

Additionally, the following interactive methods are provided for in-depth document exploration:

- `mclick(query: str)`: Click on items that match the query to navigate deeper into document structures.
- `back()`: Navigate back to the previous view or page.
- `mzoom(level: int)`: Zoom in or out on the document based on the provided level for a closer or wider view.
- `mdownload(query: str)`: Download a copy of documents or portions of documents matching the query.
- `mpage(page_number: int)`: Jump directly to a specified page number within the document.

Use these tools to manipulate how documents are displayed and interacted with, to enhance the accuracy and efficiency of your document searches.

edit: v2 file search is more modest in its usage of tools internally: just a single search that can accept multiple queries.

1 Like

Ok, I’ve got a clear understanding of myfiles_browser now. Thanks a lot.
:+1: