Hypertool: Making Sense of the Cloud-to-Edge Sprawl

Modern applications no longer live in a single, tidy data center. They stretch across powerful cloud servers, on-premises machines, and small devices at the network's edge, from factory sensors to computers inside connected cars. For applications driven by AI, the Internet of Things, and time-sensitive tasks, getting work done in the right place at the right moment is everything.

But the tools that coordinate all these machines, most notably Kubernetes, the industry standard, were built with a simple picture of what a computer is: processor cores, memory, and storage. That worked well when infrastructure was stable and predictable. It falls short today, when the resources powering an application can be wildly different and constantly changing.

As part of the HYPER-AI project, researchers from the National and Kapodistrian University of Athens, the Cyprus University of Technology, and eBOS Technologies have developed Hypertool, an open-source framework that closes this gap.

Two problems, one tool

Hypertool tackles two long-standing headaches. The first is how machines are described. Knowing a server has eight cores and sixteen gigabytes of memory says little about whether it's right for a given job. Is it energy-efficient? Does it have a specialized chip for AI? Is it secure? Where is it located? Hypertool enriches each machine's profile with these attributes, giving the system a realistic understanding of what every resource can actually do, including its performance, energy efficiency, cost tier, reliability, security posture, network quality, location, and any specialized hardware such as GPUs.

The second is lifecycle management, the process of adding, tracking, and removing machines. Traditionally this is a manual, error-prone affair that often needs a specialist. Hypertool automates it, so machines can join and leave a cluster smoothly, with far less risk and effort, complete with safety features like a "dry run" mode and automatic rollback if something goes wrong.

Crucially, it builds on Kubernetes' own native mechanisms rather than bolting on heavy new machinery, so it stays lightweight and slots neatly into existing setups. The team validated it across three settings, from a single machine to a diverse public-cloud cluster with GPU-equipped hardware, confirming it is stable and ready for production use.

Why it matters

Hypertool moves the computing continuum a step closer to managing itself. By giving orchestration systems a richer, continuously updated picture of their resources, and by automating the tedious work of keeping a fluid cluster in order, it makes demanding modern applications far easier to run across today's messy, heterogeneous infrastructure. The team plans to extend it further with deeper network insight and predictive features that could anticipate and heal problems before they bite.

The full paper, presented at CLOUD COMPUTING 2026, is available here.