This CLI takes a query, searches the web, extracts text from a result, and asks Gemini to turn that material into a structured Markdown answer. It uses Trafilatura to pull readable page content out of HTML, so the model gets source text rather than just a list of links.
I also added a small memory feature for facts the assistant marks as worth keeping. The project started as a way to make search, page extraction, and synthesis work in one terminal loop.
Code: Perplexity-Clone-CLI.