About

Modelystis a cross-linked catalog — a “Bloomberg terminal” — for AI research. Ten entity types (models, evals, RL envs, datasets, papers, organizations, people, tools, leaderboards, capabilities) are joined by the relations that matter and rendered as statically prerendered pages.

How it's built. Next.js App Router with React Server Components on Postgres + Prisma. Pages are statically generated with ISR (revalidate 300s); reads ride the RSC payload — there is no client data-fetching library. Two runtime APIs: /api/search (a Postgres trigram index with a pgvector semantic fallback) and /api/ask (cited, retrieval-augmented answers over the catalog).

Your data stays yours. Watchlists, notes, and highlights live only in your browser (localStorage / IndexedDB). The one thing stored server-side is the email you provide to unlock the assistant. Papers link out to arXiv; PDFs are never hosted.

Sources & credits. Model data from the Hugging Face Hub; papers from arXiv; citation counts from OpenAlex; benchmark scores from the Open LLM Leaderboard. The assistant is powered by OpenAI. Datasets, RL environments, and lift relationships are curated. All trademarks belong to their owners. See terms & privacy.