Layout (desktop / lg+ only):
- 40px top padding above the page title.
- Deck grid uses minmax(320px, 1fr), so two cards fill the content width
and share a right edge with the header (same max-width container).
- Group headings only render once real folders exist; with none, a single
quiet "All decks" label replaces "Unorganized".
- Import JSON becomes a plain text button (mobile keeps the pill).
Deck cards (desktop):
- The whole card is a link; Study is a compact secondary button instead
of a full-width solid pill.
- The ... control is a 32px ghost button with a real ellipsis icon,
aligned to the title's centre.
- Meta line always shows the same fields (activities · runs / Not played
yet); the score is a separate quiet "Best 53%" stat.
- 12px rhythm between title/meta/tags/summary, 16px before the actions.
- Hover: lighter surface, stronger border, soft shadow and a 2px lift in
light mode (pointer devices only, so touch has no sticky hover).
Sidebar (desktop rail):
- All decks + Unorganized are views above the FOLDERS label; only
user-created folders sit under it.
- Count only on All decks. The mobile drawer is unchanged.
Theme:
- Dark: sidebar is a step darker (#181818 to #141414 vs #212121 page).
- Light: cards get a slightly stronger border (#e2e2e2) and a soft
shadow; the sidebar divider and edge shadow are stronger too.
- Medium difficulty darkened to #7a5300 (6.2:1 on the light chip).
Mobile is untouched: layout changes are behind lg: (or the desktop prop);
theme tweaks are shared tokens.
Verified: key-gate 49/49, install sheet 64/64, retry 14/14, zoom 6/6,
deck chrome 15/15, deck fixes 15/15, final ux 12/12, contrast 20/20,
security 8/8, pwa 16/16.
Next.js 15 app that turns photos, PDFs, DOCX/PPTX and text into a swipeable
feed of AI-generated activities (flashcards, multiple choice, true/false,
fill-in-the-blank, matching, ordering, short answer) using the user's own
Ollama models.
Highlights:
- All AI calls run in server route handlers; the browser never talks to Ollama.
- Bring-your-own-key mode (default): the key lives only in the browser and is
sent per-request; server, local-daemon and shared-key modes also supported.
- Client-side document extraction (pdf.js, mammoth, jszip) with scanned-PDF
page rendering for the vision model; notes stay in their original language.
- ChatGPT-style shell: collapsible sidebar / mobile drawer, neutral token-based
light + dark themes, folder organisation, PWA (manifest, SW, offline page).
- Strict JSON validation with zod + auto-repair, rate limiting, and request
guards (cross-origin, body-size, API-key sanitising).
- setup.sh bootstraps deps, .env.local and a production build.
Gitignore excludes .env*, the generated pdf.js worker and local scratch files.
