IKEA assembly, step by step: animated, manual-style 3D guides. Scan any paper manual with Together AI.
TypeScript
README: 829 products, 1,335 sizes, 1,084 clips
Youssef04010b840d48
diff --git a/README.md b/README.mdindex e92865e..601abe9 100644--- a/README.md+++ b/README.md@@ -10,18 +10,18 @@- Next.js 16 (App Router, Cache Components) with React 19 and TypeScript, plain CSS, no UI kit- three.js line-art stage (`src/engine`): ink outlines, fly-in parts, a zoomed callout per fitting, offscreen thumbnails-- A parametric plan generator (`src/lib/plan`) for 7 furniture types and 35 kinds: 778 IKEA products (1,239 sizes) drawn offline, with typo-tolerant search and article numbers+- A parametric plan generator (`src/lib/plan`) for 7 furniture types and 43 kinds: 829 IKEA products (1,335 sizes) drawn offline, with typo-tolerant search and article numbers- IKEA's public product facts (`scripts/ikea`): name, type, size and article numbers only, mapped to kinds by `classify.ts`, checked by `npm run ikea:check`; every IKEA family name is known, so MARKUS reads as an office chair and never as a typo of MARIUS- Manual scanner (`/scan`, parked for v2 and off unless `NEXT_PUBLIC_SCAN=1`): a vision model reads each page, `gpt-oss-120b` lays out the parts in millimetres, and `src/lib/plan/custom.ts` places every dowel, bolt and screw where parts touch- Together AI for unknown product names (steered by IKEA's own type, cached for good with `use cache: remote`), photo lookup and the "Stuck?" helper, with model fallback lists in `src/lib/together.ts`-- A hands-free voice on Together AI (Kokoro, `src/lib/voice/tts.ts`): every catalog line is pre-recorded with `npm run voice` (937 clips cover all 1,239 guides), the rest are recorded live, product names keep phonetic spellings where they help (`npm run voice:names`), and saying "next" turns the page+- A hands-free voice on Together AI (Kokoro, `src/lib/voice/tts.ts`): every catalog line is pre-recorded with `npm run voice` (1,084 clips cover all 1,335 guides), the rest are recorded live, product names keep phonetic spellings where they help (`npm run voice:names`), and saying "next" turns the page- Workshop (`/builds`): every finished build merged into one scene and rendered at true scale next to a person- Share cards drawn by the engine (`scripts/og/render.mjs`, `src/lib/og`), a picture of your finished build to post, an A4 print view, and a service worker so opened guides work with no signal## Cloning & running1. Clone: `git clone https://github.com/Luffixos/bygga.git && cd bygga && npm install`-2. Optional key: `cp .env.example .env.local`, then set `TOGETHER_API_KEY` (the 1,239 offline guides work without it)+2. Optional key: `cp .env.example .env.local`, then set `TOGETHER_API_KEY` (the 1,335 offline guides work without it)3. Run: `npm run dev`, open http://localhost:3000 and type `BISSA`4. Jump to a step: http://localhost:3000/build/malm-3-drawers?step=12, print one at `/build/malm-3-drawers/print`, or see finished builds at `/builds`5. Try the scanner free: run with `NEXT_PUBLIC_SCAN=1`, then on `/scan` paste `bygga-dev-mock-key-000000` as the key and add any photo (only `npm run dev` accepts it)