AI-powered optimization advanced settings using JSON file and/or config file.

[ovalerio]

Dear scalene developer Team,

Thanks for creating scalene. I found this software is a great tool addition in every Python software development toolbox. It also helps to create compelling visual aids to communicate with users about potential performance bottlenecks. The feature I want to request is related with the AI-powered suggestions.

** Is your feature request related to a problem? Please describe.**

• Setting the key and agent endpoint every time is error prone and could potentially leak the key.
• When the analysis is run in parallel in different hardware it could help to automate/streamline the performance profiling task.

Describe the solution you'd like I would like to have a config file that I can place in a well know location under the user home directory ($HOME/.scalene/config.json) I found json would be a good format to store the configuration, so that it can be either set/edit using the command line or directly modified by the user.

Describe alternatives you've considered I have considered also the use of environment variables that could be read by scalene. For example, it would be important that it also consider the host proxy settings. Another alternative would be that the GUI provides the prompt that needs to be passed to the LLM so that it can be copy/paste and run into an LLM web input form. This would be especially useful for developers working with free accounts that have no API access to the LLM agents or that would like to use an LLM that is not one of the listed in the default list. This is particularly important considering the fast pace in which the LLM landscape is changing and models coming and going out of fashion. Any hard-coded endpoint or API version could potentially be deprecated in few months time.

Additional context Mainly running scalene profiler through vscode in compute nodes behind an internet proxy. The code I profile has been developed by a third party and scalene helps me to learn/study the performance bottlenecks.