Marginalia - Chat With Your Documents - Next.js AI SaaS Template

Marginalia is the first genuinely functional AI product in the catalog - retrieval and generation actually run against real documents, not a marketing page about AI. Upload PDFs and documents, ask questions in natural language, and get streamed, cited answers where every claim links to the exact source passage and page. It ships the whole product spine: a marketing landing that demos the citation-linked chat, pricing, full auth (login, signup), a 3-step onboarding wizard, a gated app shell with sidebar and workspace switcher, a dashboard launcher that turns a question into a live cited answer, a document library with an animated ingestion state machine, a source viewer that citations deep-link into, the two-pane citation chat (the signature 'citation-as-source-preview' interaction), settings, and billing. Single-database architecture: app data and vectors both live in Neon Postgres via pgvector - one connection string, no separate vector service. Model-agnostic by construction: every AI call routes through the Vercel AI Gateway using 'provider/model' strings, so you swap models without touching code. Runs zero-config in demo mode (a seeded, pre-embedded corpus) and flips to the real R2 + embedding + pgvector pipeline behind documented seams. The 'Reading Room' design system - warm porcelain neutrals, an espresso-and-sienna palette, and Fraunces editorial serif - reskins from a few CSS variables. Flash-free light and dark mode, fully responsive, strict TypeScript, deterministic SSR (no hydration mismatches), zero backend required to run.