Glassroom

Shape the AI stack your class learns with.

Glassroom gives your course its own AI stack — you pick the models, write the instructions, and set the limits, so it works the way your course does. The record of every exchange stays on your hardware and becomes insight the whole class can learn from — instructors and students both. Nothing leaves the room.

Read the setup guide → ~20 min from a bare machine
m.okafor · session 12today 14:32
Student

Why does my double-slit simulation still show fringes when I cover one slit?

Model

Check the aperture mask in your code — if the covered slit's amplitude isn't actually zeroed, the model still has two coherent sources, so interference fringes persist. Print the mask array first…

optics-101 · class ledger24 students
StudentSessionsExchangesLast active
a.gupta1119811:05
m.okafor1221414:32
j.tran916709:41
s.whitfield7122yesterday

One record · two lensesRead one student's session end to end, or step back to the ledger of the whole class — the same record, seen through two panes.

Your course's AI stack

Tuned to how you teach.

A writing seminar and a robotics lab shouldn't get the same assistant. Pick the models, write the instructions, set the budgets — each class gets an AI stack shaped for its work, as conducive to learning as you can make it, and you can revise it as the course unfolds.

One gateway

One stack, one place for the class.

Students chat in a clean web interface, or point their own coding tools at the same endpoint. You manage the roster; every request runs through a single gateway you control — no personal accounts, no scattered logins, no API keys to hand out.

The record

Insight the whole room can use.

Each prompt and response is tied to a student and a conversation, and it stays on your machine. See where the class leans on the model, where it pushes back, where it gets stuck — then bring what you find back into the room, where students can learn from it too.

Your hardware

Runs on your hardware, on your terms.

Host it on a GPU in your lab and student data never reaches an outside vendor. No GPU on hand? Point the same stack at a hosted model instead. Open source end to end — stand it up in an afternoon, take it apart just as easily.

An afternoon, start to finish.

STEP 1

Stand it up

One compose stack on a GPU box in your lab — or point it at a hosted model. About 20 minutes from a bare machine.

STEP 2

Create a class

Each class gets its own address, chat site, and private record — optics-101.glassroomai.com, sealed off from every other class.

STEP 3

Add your roster

Students sign in with their school email. No accounts to buy, no keys to distribute, nothing for them to configure.

STEP 4

Read the record

Per-student sessions, whole-class patterns, usage counts — in a portal made for instructors, not infrastructure teams.

Students know. That's the point.

Glassroom works because everyone can see the glass. Students are told what's kept and who reads it — the record is a shared lab notebook, not a camera in the corner. Used well, it changes the conversation: instead of policing AI use you can finally teach it, with the actual transcripts on the table.

NOT A PROCTORING TOOL · NO DETECTORS · NO KEYSTROKE LOGGING · NO DATA SOLD, EVER — THERE'S NO ONE TO SELL IT TO

Built on vLLM, LiteLLM, Open WebUI, and Langfuse — proven open parts, assembled into one stack an instructor can run alone. This page loads zero external resources; the product keeps the same promise.

Bring your class inside.

A GPU box in your lab, the setup guide, and about twenty minutes. Everything is on GitHub.

Read the setup guideopen source · self-hosted