AI NewsOpen sourceAnnouncement
Bygga, an open source 3D IKEA assembly guide, has 412 stars
Bygga, an open source IKEA assembly guide, draws every step in 3D and reads it out loud. It has 412 stars.
Image: GitHub
Why it mattersBygga is a worked example of what an AI assistant can do for a non-developer task. The tech stack is open, so a builder can copy it.
Anyone who has sat on the floor with a sheet of IKEA pictograms and the wrong bolt knows the problem. Bygga is an open source web app that draws every assembly step in 3D. It reads each step out loud. Hassan ("Nutlope") El Mghari published it 10 days ago. The repository has 412 stars at the time of this item.
The user types a product name or the article number on an IKEA box. Bygga pulls the matching plan from a local catalog. The catalog holds 1,171 IKEA products at 2,180 sizes across the UK and US ranges. The app then draws the assembly in line-art 3D with Three.js, matching the look of the paper manual. The voice reads each step from one of 1,476 pre-recorded clips. A Whisper mic listens for the word "next" through an echo-cancelled channel.
What makes it practical
The catalog is built offline from IKEA's public listings. The 3D drawing runs with no API key for the common case. A user who wants to go off-catalog adds a Together AI key. Together then drafts a plan from product photos and runs a help chat. The help chat answers with a drawing of the exact panel, fitting or step the user is stuck on. The repository carries a mock key for local testing.
Bygga also prints a paper version at an A4 size. It shows every build the user has finished side by side at true scale. A service worker keeps opened guides working after the signal drops. The share cards for each product are drawn by the same engine that draws the steps.
What a builder can copy
The README names the parts of the stack. Next.js 16 with App Router and Cache Components, React 19, plain CSS with Motion and no UI kit. The line-art stage sits in src/engine and the plan generator in src/lib/plan. The listening pipeline is /api/listen with intent handling in src/lib/voice/commands.ts. Tests cover 283 recorded voice runs and every row in the IKEA catalog.
The project runs on a pattern that is now cheap. The hard data (shapes, sizes, fittings) is deterministic and stored locally. A model is called only when the user leaves the catalog or asks a help question. The common path is a static plan with no model latency. The AI key cost belongs to each visitor rather than the operator.
Why this works as a demo
Bygga answers the "what can you build with an AI tool this weekend" question. The product it builds is one a non-developer would use. The hook is boring in the right way: find the IKEA product, follow the steps, finish the shelf. The AI is in two specific slots. It generates a plan when the catalog does not have one. It also answers the help question when the user names what is stuck.
The design is also a working example of a pattern teams building AI products ask about: how to keep the deterministic path fast and cheap while letting a model handle the long tail. The deterministic path here is 1,171 IKEA products drawn offline. The model path is a Together AI call when a user falls outside the 1,171.
Bygga is MIT licensed. The project is not affiliated with IKEA. The repository notes that the MIT licence grants no rights in the IKEA trademark.
Source
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
