Radar / Ideas / ClassroomBox: an interactive AI…

ClassroomBox: an interactive AI classroom for schools with no internet

daily ideamoderateJEV confidence 0.59JEV monetization 0.522026-11-05

OutcomeAn interactive AI classroom running in a school with no internet connection at all.

The problem

Most edtech assumes an internet connection, which puts it out of reach for the roughly 2.6 billion people without reliable internet — including rural and refugee schools where the teacher is often the only resource in the room.

The idea

A classroom-in-a-box: a low-cost PC image that runs an OpenMAIC-style multi-agent classroom, a locally served open model via colibri, and Palette-style deck generation, all over the school's local Wi-Fi with zero internet. A teacher picks a topic, the box generates an interactive lesson with AI teachers and classmates, plus printable decks, and students join from whatever devices exist. The MVP covers one subject area and one language, and deliberately skips student accounts, grading analytics and cloud sync.

Why now

OpenMAIC (November 3) shows multi-agent classrooms work as software, colibri (v2.1.0, October 31) proves frontier-scale models run on consumer hardware, and Palette (November 3) adds local lesson-deck generation — the three pieces for a fully offline classroom all landed within days of each other.

What it combines

OpenMAIC's multi-agent classroom plus colibri's offline frontier inference plus Palette's local deck-building: together they become a classroom that needs no internet at all — AI teachers, classmates and lesson decks running on one school PC — which none of the three delivers alone.

MVP

Weekend MVP: one PC running an interactive multi-agent lesson on a single topic over local Wi-Fi, with one generated deck, tested with two devices and no internet cable plugged in. Skip: multiple subjects, languages, student tracking, and any teacher dashboard.

How it makes money

Budget owner: an education NGO programme manager or a ministry school-equipment budget — the same budget that buys offline learning hardware today. Pricing anchor: We Who Look Up's offline learning station at about 4,000 SEK per unit (wewholookup.org), an adjacent paid product proving schools pay hardware prices for offline AI classrooms. Willingness-to-pay evidence: Learning Equality's Kolibri is installed in over 220 countries and territories and used by UNHCR, UNICEF and World Vision in refugee and rural schools — large buyers already procure and deploy offline learning at scale.

The moat

The largest edtech incumbents are cloud-first: their products assume connectivity, accounts and subscription revenue, and their content and analytics pipelines live in the cloud. An offline-first box earns nothing from subscriptions and needs a content model they would have to rebuild from scratch — it cannibalises the cloud seat licence, so they will not ship it.

Distribution

B2B2C through education NGOs and ministries: sell the box to organisations that already deploy offline learning (UNHCR, UNICEF and national school programmes) and let them place it in schools, rather than selling school by school.

Why it wins

Unlike cloud classrooms that need connectivity and accounts, the box runs entirely offline on one PC; unlike Kolibri's static library, it generates interactive multi-agent lessons and decks on demand.

Risks

Biggest risk: local model quality on a single PC disappoints compared with cloud classrooms, and teachers abandon it. The MVP de-risks it by testing one subject with real teachers first and tuning the lesson format to what the local model does well (structured lessons, quizzes, decks) rather than open-ended chat.

Build it with

OpenMAIC: turn any topic into a multi-agent interactive classroomTurns any topic into a multi-agent interactive classroom with AI teachers and classmates.
colibri: run 744B-parameter MoE models on consumer hardwareRuns frontier-scale models on consumer hardware with no internet, the offline engine for the box.
IBM Palette: a 20B open-weight model that builds decks in your design languageBuilds on-brand teaching decks locally, giving teachers ready-made lesson materials.

Repo to start from

classroom-box: a PC image for an offline multi-agent classroom with local lesson and deck generation.

Evidence

Get the week's best AI launches, plus 3 ideas worth building

One email every Saturday. Ranked by traction, not hype. Free.