[feature] - vLLM integration for more sophisticated / automated extraction and processing of scraped pages

[kquinsland]

Version and OS

0.49.17, Linux/k8s

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

This is a question to the dev(s?) that could result in several new features / make changedetection.io more applicable to additional uses.

As an example, the "Cannot run, more than one price detected..." issue from #3203

Describe the solution you'd like

A way to "outsource" all of the work that has to be done after a screenshot of the page can be taken.
It costs a few tenths of a penny to send an image + a basic "What is in this picture? Do you see price details, what about promotions/taxes/tarifs? Are there any other SKUs...? etc"

I can imagine implementing this in a few different ways; I'm curious if there's been any thought about doing this.

I have a super basic PoC that works reasonably well; it costs me a few tenths of a penny to ask openAI to extract pricing/availability data from a screenshot and return a basic json document.

[Kohnnn]

+1 to this, looking for a way to integrate LLMs into the app. My current setup involves changedetection to send webhook notifications that include the latest change to my N8N agents, and have it extract data from text and screenshots for me - quite a hassle.