Pegasus Labs

A Complete Computer Lab.No Computers Required.

Get Pegasus Labs for your Institution
labs.youruniversity.whitebird.ai
train.pymodel.py
1import torch
2from torch import nn
3from data import make_loader
4
5model = ConvNet(classes=10).to("cuda")
6opt = torch.optim.AdamW(model.parameters(), lr=3e-4)
7
8for epoch in range(40):
9 for x, y in make_loader("train"):
10 loss = nn.functional.cross_entropy(model(x), y)
11 loss.backward()
12 opt.step(); opt.zero_grad()
Python 3.11GPU · A100 attached

student@ml-lab-04:~$ pip install -r requirements.txt

12 packages from campus cache · done in 0.8s

student@ml-lab-04:~$ python train.py

epoch 11/40 ━━━━━━━━━━━━ loss 0.211 · acc 91.4% · 38s

epoch 12/40 ━━━━━━╸━━━━━ loss 0.183 · acc 92.1%

Open a link. Start coding.

No software installed on a single campus machine. Every student gets a real editor, a terminal, package installs and live preview URLs — in seconds.

Instant Workspaces

Ask for a workspace, and nine seconds later it exists — files mounted, extensions loaded, exactly as the student left it.

Start a Python workspace

ml-lab-04 · booting

Riya Malhotra · CS-405

ready in 9s

Image pulled

py-ml · cached on campus

Volume mounted

your files, exactly as left

IDE + terminal ready

extensions preloaded

ml-lab-04.preview.labs.youruniversity.ac.in

Ready-Made Templates

Five templates cover every coding course you run — from first-year C to GPU-backed deep learning, stamped out per cohort.

C / C++

Python

Java

Node

React

ready-made · yours to extend

Data Structures Lab · CS-201

64 seats

stamped from template cpp-14 · one click

Docker Isolation

One student, many containers, an app, a database, a worker — each isolated, each scheduled onto worker nodes in your own cloud.

Riya Malhotra

full-stack project · Sem 5

one student

web-app

node 20 · port 3000

postgres

data volume · private

worker

queue · cron

isolated containers · scheduled on your worker nodes

Thin-Client Ready

Any browser is a dev machine, eight-year-old lab PCs, Chromebooks, tablets — 2 Mbps is enough to code like it's local.

8-year-old lab PC

4 GB RAM · fine

Student Chromebook

nothing installed

Tablet at the hostel

keyboard optional

The same workspace

browser-native

full editor · terminal · preview

2 Mbps is enough · 5–10 recommended

Your cloud. Your data. Our platform.

Pegasus Labs deploys inside your institution's own AWS account. Workspaces, storage and code never leave your VPC — and your team holds the keys.

Data Security

Code never leaves the cloud, student work lives and dies inside your VPC — every session logged, IAM in your own account.

your VPC

student code + data

leaves the VPC · blocked

code never leaves the cloud · your team holds the keys

Access audit

every session logged · IAM in your account

Clean
Seat Management

Grant, revoke, lock — seats per course from one panel, with exam mode that cuts the internet at 14:00 sharp.

Seat management

68 / 100 seats active · 4 courses

Control panel

Data Structures · CS-201

cpp-14 · Dr. Mehra

32 seats

ML Lab · CS-405

py-ml · GPU enabled

24 seats

Web Dev · CS-310

node-20 · self-paced

12 seats

Lab exam · CS-201

starts 14:00 · internet off

locked
Runs In Your AWS

One deployment, one VPC, the control panel, every workspace and all storage inside your account — AWS bills you directly, no markup.

students · any browser

Your institution's AWS · VPC

Pegasus control panel

Authentication

Seats & users

Orchestration

Templates

Monitoring

Logs & analytics

Student workspaces · isolated

×100

Student 1

python

Student 2

java

Student 3

c++

⋮

Student 100

node

File storagePackage cacheLive previewBackupsbilled by AWS to you · no markup
Deployment

Ready in days, not months, AWS access on Monday, templates loaded midweek, the first cohort coding by Friday.

Day 0

AWS access granted

IAM role · your account

Day 1

Platform deployed

control panel + workers

Day 3

Templates loaded

C/C++ · Python · Java · Node · React

Day 5

First cohort coding

no software installed on campus

ready in days, not months

Compute that sleeps. Costs that don't creep.

Workspaces sleep when idle and wake on a click. GPUs attach for the minutes a job runs. You pay for usage — not for racks of aging lab hardware.

Pay When Used

Workspaces sleep on their own, 30 idle minutes and the compute stops billing — a lab day costs hours, not a calendar day.

ml-lab-04 · yesterday

24 hours

0006121824
awake · billed idle · winding down asleep · free

billed 6h · not 24h

sleeps after 30 idle minutes · wakes on click

GPU On Demand

GPUs attach per job, a training run borrows an A100 for 42 minutes and gives it back — billed in minutes, not semesters.

python train.py --epochs 40

needs CUDA · workspace requests a GPU

A100 · attached

on demand
epoch 12 / 40gpu util 96%
epoch 40 / 40run complete · 42 min

released · billed 42 minutes, not a semester

Built For Scale

A hundred students at once, lab-morning surges, exam days and GPU cohorts — the fleet stretches and shrinks with the timetable.

the fleet · tuesday 11:04

71 active · 21 idle · 8 asleep

holds

100 concurrent seatsexam-day surge30 GPU seats

Platform

Package Cache

pip, npm and apt hit a campus-local cache — installs in seconds, not minutes.

Platform

Live Preview URLs

Every workspace serves a shareable HTTPS preview of whatever it's running.

Platform

Snapshots & Backups

Workspace state snapshotted automatically — restore any student to yesterday.

Platform

Monitoring & Health

Fleet health, usage and cost visible live in the control panel.

Every student. Their own machine.

One deployment in your own cloud, every student coding from any browser on campus — with GPU power for the ones who need it.

Talk to us