I have been building an AI power deal hunter and price tracker

Repo: github.com/oldmoldycake/Snagr I have been making a self hosted AI deal hunter and price tracker. You create categories and assign sites and items to it and the AI agent drives a playwright MCP and will hunt for deals on items that match your plain English criteria (ex. I want a copy of Pokémon emerald, no box, damage label and dry better). It also ground the price to current market rates with searXNG (Brave API coming very soon!) so it can better identify deals and has an experimental vision sidecar that uses machine vision to use a library of labeled fake/real images of an item to be able to identify fakes better from pictures combined with information present on the page and seller profile. I have been working to get a ton of polish and bug fixes out on this and I am looking for feedback as well as hoping to find someone that can benefit form this as I have already used it to source several good finds! As for token usage, use cheap models it can get token hungry if you have it check too often for new listings and I am working to find ways to get token usage as low as possible and I have a few ideas in mind that may help after I get my current bug fix round done.