AI Router & Search Portal

I’m building a web app that replicates the openrouter.ai experience but on my own domain. The main view should feel like ChatGPT: a large conversational window in the center, while a fixed sidebar on the right automatically lists every free or paid LLM, SLM, or micro-service I register, each with a one-line description and link to its endpoint.

When a visitor sends a prompt the system must first hit a Perplexity-style retrieval layer that scrapes publicly available pages from the last seven days, then pass that fresh context to a free OpenAI micro or nano model to generate the answer. During typing the user receives keyword-aware hints so they can refine their request in real time (COT / reactive search).

Every prompt, the resulting answer, and the set of sources get stored in the user’s Google Drive—scoped to their own e-mail ID. On page load only the five most recent searches come back and appear in a compact history panel; nothing older is displayed.

Key deliverables
• Responsive front-end (React, Next.js, or similar) with chat window, model sidebar, and five-item history list
• Back-end microservice that orchestrates web scraping, vector storage, the chosen OpenAI micro/nano LLM, and the hint engine
• Google OAuth flow plus Drive integration for per-user caching
• Step-by-step deployment notes so I can add new models or services later

Acceptance criteria: the sidebar populates automatically, prompts return answers grounded in last-week sources, keyword hints surface during input, Drive caching works per user, and only the last five searches appear on screen.