Can a prompt replace Perplexity?

Build a personal answer engine that wraps real search and model APIs rather than reimplementing either. For each question: query a real search API (several offer a usable free tier), fetch the top handful of results, extract the readable text from each page, and pass the question plus numbered excerpts to a model (Claude or GPT, key from the environment) with strict instructions to answer only from the given excerpts and cite each claim as [1], [2], and so on. Render citations as clickable footnotes with the full source list under every answer — if a page fails to fetch or its text can't be extracted, drop it silently rather than ever citing something that wasn't actually read. Keep follow-up questions in the same thread, with prior questions and answers carried forward as context. Store threads in a database with a sidebar listing past threads by their first question.

Do not build your own web index or crawler, model routing across providers, or mobile apps — retrieval quality here tracks the search API directly, and that's the honest trade for not running Perplexity's own infrastructure.