MakerLAB Tools · Cornell Tech
An AI that helps you operate, debug and create in the MakerLAB. It knows every machine in the lab, reads their manuals, and tells you which page it read.
Opens the Bambu Lab X1-Carbon with MakerLAB AI ready
Every tool in the lab · 12 languages · no account needed to ask
The X1-Carbon's page, a question to the assistant, its answer from the quick start guide, then the lab status screen. Recorded on the live site.
Three kinds of question, one AI. Each answer below is real, from the X1-Carbon's page on the live site.
“What does the quick start guide say about the first print?”
How to use a machine, step by step, from its own manual, with the page it came from.
“My first layer isn't sticking to the build plate. What should I check?”
What to check when something goes wrong, what not to force, and when to stop and get staff.
“I want to print a two-colour phone stand. Which printer, and what settings?”
From an idea to the right machine in this lab, the material, and the settings to start from.
Every tool in the lab has a page: what it does, the training and PPE it needs, the emergency stop, specs and materials, the SOP and manuals, and every physical unit with its status and location. Scan the QR label on a machine to land on its page. Staff print the labels from the inventory, at the sticker size they use, and anyone can save a tool’s code from its page or ask MakerLAB AI for it.
The assistant searches each machine's manual, page by page, and links the exact pages it used. Tap a page reference to open the manual there. When the manual doesn't cover a question, it says so instead of guessing.
Staff photograph a machine or paste an inventory list. The assistant identifies each item, then researches it in the background: the manufacturer's page, specs, the manual and a clean product photo. Nothing is published until a staff member reviews and approves it.
Tell the assistant a machine is broken. It helps you try the obvious fix first, then files a ticket in the maintenance queue with the machine, the unit and a priority. Staff see it at once, and the lab screen counts it until it's resolved.
Open /kiosk on the TV or an iPad at the door: machines that are down, open tickets, today's hours and a featured tool. It refreshes itself, and its QR code opens the assistant on a visitor's phone.
Students submit what they made, with photos and the tools they used. Staff approve each project before it's public, and every tool page shows the projects made with it.
Which tools people ask about, which they never do, the busiest hours, and the questions the assistant couldn't answer, so staff can add the missing manual. The value report turns this into staff hours saved and after-hours coverage, on one printable page for whoever funds the lab.
Staff can ask the assistant to resolve a ticket, publish a tool, add a unit or approve intake. It prepares the change on a card that shows exactly what will change, and nothing happens until you press Confirm. It acts with your permissions, never more.
It never touches the super admin role, removes people, blocks addresses, handles secrets, deployments or backups, shows anyone's full email address, or sends email.
Claude, ChatGPT and other assistants can use the same catalog over MCP. Browsing needs no account. Sign in with Google from the assistant and it can file reports as you, with your role's permissions.
The site and the assistant work in English, 中文, Español, हिन्दी, 한국어, العربية, Français, Português, Русский, Türkçe, 日本語 and עברית. Pick one from the language menu; tickets still reach staff in English.
Signed-in members see the lab's floor plan with every tool placed in its zone. Search for a tool and its station lights up.
Model costs per unit of work, measured on MakerLAB Tools (a model call is the only cost that grows with use). Monthly figures are estimates for the MakerLAB as of 2026-09-28.
| A chat answermodel cost, 0.03¢ | ≈ $0.0003 |
|---|---|
| A chat answer that searches a manualadds the reranking step that picks the best pages | ≈ $0.002 |
| Researching a new toolthe agent reads the web, the manual and finds a photo; runs in the background | $0.02–0.05 |
| Indexing a manual20 manuals, 1,136 pages, cost $0.013 on 2026-09-28 | < $0.001 |
| Reading a scanned manual (OCR)capped at 150 pages; most cost cents | ≤ $1 |
| AIchat, research and indexing | ≈ $10–15 |
|---|---|
| Hostingthe site, the database and file storage | ≈ $20–40 |
An agent looking things up costs more than a chat because it reads pages: about 5¢ to research a tool against a fraction of a cent to answer a question. That is why research runs only when staff ask for it.
Measured with the Vercel AI Gateway's reported costs. Estimates, not quotes; prices change.
Built by Isaac Steinberg (Tech Lead, Johnson Cornell Tech MBA '26) for the MakerLAB, run by Niti Parikh (Director) and Luis Rodrigo Navarro (Assistant Director).
Open the X1-Carbon's page and ask the assistant anything about it.