v0.2.1 · Multi-Agent · Live

Theclassroom,rebuiltfortheAIera.

Turn any topic or document into a live, multi-agent classroom — slides, whiteboard, narration, simulations, and code, generated and orchestrated by AI. In one click.

Multi-Agent NativeBuilt on Zora Platform13+ Model ProvidersSelf-hostable
Scroll
ai-classroom.live· Scene 02 of 06
Scaled dot-product attention
Each query asks every key: how relevant are you to me?
Q
K
V
T
Each query asks every key how relevant it is — that’s attention.
[01] · The thesis
Slides are static.
Videos are one-way.
Classrooms
should think back.

AI Classroom replaces passive content with a faculty of agents — a teacher that explains, a peer that asks the dumb question, an expert that goes deep, a skeptic that pushes back. You learn by being in the room.

[02] · The pipeline

One input.A living classroom out the other side.

STEP 01

Drop in a topic, a PDF, or both.

A prompt. A research paper. A 200-page textbook. The system reads it, infers the language, optionally pulls in fresh context from the web.

STEP 02

A faculty plans the lesson.

A director model breaks the topic into 5–12 scenes — lecture, discussion, simulation, quiz, project — each with its own learning objective, agents, and pacing.

STEP 03

A live classroom plays back.

Slides render. Voices narrate. Agents draw on the whiteboard, debate each other, run code, fly through 3D models — and you can cut in at any time.

new course
Prompt
"Teach me transformer attention from scratch using the original Vaswani paper."
Attached
attention-is-all-you-need.pdf
15 pages · 1706.03762
Web search on · 4 recent sources
claude-sonnet-4.6 · thinking: high
Generate
outline · 6 scenes · ~44m
Director-Graph
1
Lecture
Why attention?
5m
2
Whiteboard
Q · K · V — the geometric intuition
7m
3
Discussion
Peer Q: why scale by √dₖ?
4m
4
Simulation
Interactive attention heatmap
8m
5
Quiz
Self-check (5 questions)
5m
6
PBL
Build a tiny attention head
15m
live · scene 02 / 06
T
P
E
S
Scaled dot-product attention
Q · Kᵀ / √dₖ
Q
K
V
T
Each query asks every key: how relevant are you to me?
00:02:18 / 00:07:12
SPACE pause⌘ K ask
[03] · Inside the classroom

Eight surfaces.One continuous lesson.

Every classroom mixes lecture, demonstration, dialogue, and practice. We rebuilt each of those surfaces from scratch around generative AI — and made them work together as one room.

Cinematic slides
15+ element types — text, charts, LaTeX, video. Spotlight, laser, zoom effects scripted by the AI itself.
f(x) = …
Live whiteboard
Agents draw alongside their narration. Shapes, equations, runnable code — pan, zoom, auto-fit, full history.
Multi-voice narration
Per-agent voice, synced to the visuals. OpenAI, Azure, ElevenLabs, MiniMax — or self-hosted clones via VoxCPM2.
Roundtable discussion
A director model decides who speaks next. Raise your hand any time, the conversation makes room for you.
3D, simulations & games
Deep interactive scenes — molecular structures, physics sandboxes, puzzles. Generated, not templated.
const attn = (Q, K) =>
softmax(Q @ K.T / √d)
In-browser code
JavaScript and Python execute live in the lesson. Solve, edit, re-run — the teacher guides as you go.
?
Adaptive quizzes
Multiple-choice, short-answer, fill-in. Graded by AI, persisted per scene, summarized on completion.
PPTX
HTML
ZIP
Export anywhere
Fully-editable .pptx with embedded LaTeX. Self-contained HTML. ZIP archives for backup, sharing, or grading.
[04] · The faculty

Not one model.A room full.

A director graph orchestrates up to eight agents per course. They each have their own persona, voice, and expertise — and they actually talk to each other. The director decides who speaks next, in real time, based on the conversation.

8
agents per course
28+
action types
2
orchestration modes
conversational depth
Director
decides who speaks
T
P
E
S
H
[05] · Who it's for

One platform.Two sides of the desk.

For teachers

Bottlenecks, removed. Judgment, intact.

planner — course/attention.maic
Brief
Senior undergrad. 60 min. Emphasize geometry over algebra.
01
Lecture
Why attention?
02
Whiteboard
Q · K · V intuition
03
Discussion
Peer Q: scale by √d?
04
Simulation
Interactive heatmap
05
PBL
Build a tiny head
.pptx · .html · .zip
Publish
  • Build a full course
    in the time it takes to brief a TA.
  • Personalize per cohort
    student bios shape examples, depth, and pacing.
  • Approve, then publish
    review every outline before students touch it.
  • Export to .pptx
    fully editable, with LaTeX preserved as native equations.
  • Project-based learning
    milestones, deliverables, AI peer grading — built in.
For students

The classroom catches up to how you actually learn.

classroom — attention · scene 02
scaled dot-product
Q
K
V
T
P
E
S
ask
02:18
Each query asks every key — how relevant are you to me?
P
Peer · Wait — why √d and not d?
  • Learn by doing
    simulations, code editors, debates — not autoplay videos.
  • Cut in any time
    pause, ask, rewind, change the depth without losing the thread.
  • A peer who asks the dumb question
    so you don't have to.
  • Self-check with quizzes
    graded live, with feedback that explains, not just scores.
  • Take it anywhere
    export to PPTX, HTML, ZIP — own your learning artifact.
[06] · Capabilities

Provider-neutral.By design.

Plug in the model you trust, the voice you license, the GPU you rent. Swap any layer without rewriting the lesson. Self-host when it matters — voice clones and PDF parsers run on-device for institutional control.

OpenAI
Anthropic
Gemini
DeepSeek
Grok
Qwen
ElevenLabs
MiniMax
VoxCPM2
Azure
Ollama
OpenRouter
AI Classroom
Provider layer
Language models
OpenAI
GPT-5
Anthropic
Claude 4.7
Google
Gemini 3.1
DeepSeek
v4
Qwen
Kimi
GLM
Grok
xAI
Hunyuan
MiniMax
OpenRouter
Ollama
local
Voice — speech in & out
OpenAI
Azure
ElevenLabs
MiniMax
VoxCPM2
self-host
Whisper
ASR
Image & video
DALL·E
GPT-Image-2
SeeDream
VEO
Sora
Kling
Hailuo
Web & PDF
Tavily
Brave
Bocha
Baidu
MinerU
UnPDF
KaTeX equations ·PPTX export ·Self-hosted voice clones ·Director-graph orchestration ·Multi-agent debate ·Live whiteboard ·On-device PDF parsing ·3D & physics sims ·Browser code execution ·Streaming generation ·6-language UI ·MCP tools for PBL ·KaTeX equations ·PPTX export ·Self-hosted voice clones ·Director-graph orchestration ·Multi-agent debate ·Live whiteboard ·On-device PDF parsing ·3D & physics sims ·Browser code execution ·Streaming generation ·6-language UI ·MCP tools for PBL ·
[07] · The bigger picture

AI Classroom is one room.Zorais the institution it sits inside.

Zora Platform is the AI-native operating system for higher education — where professors retain approval gates, students get both structured lessons and personal tutoring, and institutions keep control of their data and outcomes.

Zora Campus
Institutional
Timetables, attendance, grading, role-based access. The system of record for schools and universities.
YOU ARE HERE
AI Classroom
Structured delivery
Faculty-approved lessons, delivered live by AI. Multi-agent, interactive, exportable. You are here.
NeuralMentor
Personal tutor
A study companion before and after class. Q&A, problem-solving, RAG over uploads, OCR, podcasts.
The closed loop

Before, during, and after class — same student record.

Most platforms ship one half of the loop. Zora ships both — under the same roof, under institutional control.

BeforeNeuralMentorDuringAI ClassroomAfterNeuralMentor

Step into the classroom.

Sign in to generate your first course. A few sentences in, a faculty out.

A faculty in a few sentences · slides, voice, code, debate