Hello, is there any way to call gpt-4 over the API with a plugin enabled? If not, is there any workaround like manually inserting a prompt describing the plugin?
No. But, why would you want to?
You have completed control when you use the API. In essence, you can recreate complete functionally of any plugin and more but just providing the model with a list of commands it can use which include a terminating stop sequence.
Parse the output looking for a command, execute whatever command the model wanted to run, then resubmit the prompt and the response to that point with the result of the executed command for the model to continue from.
While the API does provide extensive control, integrating plugin functionality, such as the browsing tool, could enhance applications that require real-time web data. It would streamline the process of issuing search queries, parsing results, and quoting relevant information, enhancing efficiency in applications like a Virtual Research Assistant.
I needed the browsing plugin to be accessible over API
I have a github repository, llmsearch, that I use as a plugin in preference to the openai search plugin. I also use it with the api. As @elmstedt said, just parse api output and call your own hosted search instead.
llmsearch supports a pretty simple rest interface, so if you don’t mind hosting it yourself, it can provide pretty complete web search (including images, urls returned along with text extract, etc etc).
2 Likes