Cloud IDE

The cloud IDE
that schedules itself.

CP-PPO scheduling, eBPF telemetry, adaptive sandboxing, LSTM prewarming, multi-cluster federation, and conflict-free collaboration, in one open research platform.

0ms
Firecracker microVM boot
0%
Resource utilization
0
Regions, 3 continents
astra-ide@cloud
user@iiitm:~$

Live topology

Workspaces around the world

The CP-PPO scheduler places each workspace across the clusters this deployment runs on, using live node telemetry and grid carbon intensity. Connecting to the control plane…

Clusters
direct
Placement
Greenest
Traffic originsCluster upCluster down

Live demo

Adaptive sandboxing, in real time

When a user submits code, the risk scorer routes it to the right isolation tier: runc (low overhead), gVisor (user-space kernel), or Firecracker (hardware microVM).

astra-ide@scheduler
user@iiitm:~$

Seven breakthroughs

Built for research, designed for production

Click any card to see what it does in plain language and how to use it.

Scheduling

Eight schedulers, one honest benchmark

Placement is a choice, not a black box. Compare deep-RL against the classical heuristics on real workloads and pick what fits.

Pick your scheduler

The control plane ships eight placement strategies. Choose per deployment, or compare them live on any workload.

Multi-objective heuristicCP-PPO (deep RL)HEFTMin-MinLeast loadedCarbon-awareRound robinRandom

Honest result: at 15 tasks CP-PPO leads both with and without rollout search, beating HEFT by 5.8% and by 2.5% when HEFT is given the same search budget. At 40 tasks that margin disappears and an equal-budget HEFT edges it out (169.6 vs 170.5). We report both scales rather than only the flattering one.

Makespan by algorithm

lower is better
CP-PPO (best-of-32)68.5
HEFT + best-of-3270.2
CP-PPO (greedy)71.3
HEFT (upward rank)72.7
Min-Min + best-of-3276.8
Max-Min86.0
Min-Min88.2
Random131.1

benchmarks/b1_scheduler - 15 tasks, 4 VMs, 100 identical DAGs at a fixed seed. The priority-order Greedy rule (243.2) is omitted from the chart: it serialises the DAG onto one VM, which also gives it the lowest energy.

How it compares

A research control plane the others don't have

Mainstream cloud IDEs fix their scheduling, isolation and placement. ASTRA-IDE makes each an adaptive, measured decision.

CapabilityASTRA-IDEGitHub CodespacesGitpodReplitCoder
Self-scheduling control plane (RL)
learned placement policy
Adaptive sandbox tiers
runc / gVisor / Firecracker by risk
eBPF syscall telemetry
Tetragon, per-workspace
Predictive pre-warming
forecast sessions, cut cold starts
Multi-cluster federation
Karmada, cross-region failover
Carbon-aware placement
live grid intensity
Real-time CRDT collaboration
multi-cursor, conflict-free
Intrusion detection on syscalls
anomaly IDS
Open source
Self-hostable
ASTRA-IDEASTRA-IDE

Team

Prasanna Mishra
Prasanna Mishra
2023IMT-059
Udit Srivastava
Udit Srivastava
2023IMT-084
Yash Wani
Yash Wani
2023IMT-087

Ready to try the future of cloud IDEs?

Spin up a workspace in seconds. Get a private Monaco editor with collaborative editing, real-time risk-tier assignment, and one-click code execution.

Create your free account