About GPUPilot

An agentic AI product for the teams that actually run GPU clusters.

Built for the platform engineers, MLOps leads and infrastructure owners who lose sleep over a stalled training run, a silently degrading NVLink lane, or a row-remap counter that is one number away from an RMA.

What we do

GPUPilot monitors NVIDIA GPU clusters on Kubernetes and closes the loop between detection, diagnosis, and remediation. One read-only agent, deployed in 30 seconds. Real DCGM signals — XID errors, ECC (SBE/DBE) counters, row remaps, PCIe replays, NVLink bandwidth, thermal headroom. Real correlation against pods, events, and node pressure. Real suggested fixes, expressed as the kubectl command you would type yourself, that a human on your team reviews and approves. AI-assisted, not autonomous.

We ship two deployment models on the same agent: Connected for clusters with outbound HTTPS, and Air-gapped for sovereign and classified environments where telemetry cannot leave the perimeter. Both are in production today.

Who delivers GPUPilot

GPUPilot is delivered by Bynet Data Communications, one of Israel's most established infrastructure integrators. Bynet has been building, deploying and operating enterprise-grade networks, data-centres and cloud platforms for over fifty years, and today employs more than nine hundred engineers across Israel. That pedigree matters for a product like this: GPUPilot is not a code-only side project you install and hope for the best. When something fires on your cluster at 3 a.m., there is a real support organisation on the other end — local procurement, local invoicing, local escalation, and the same engineers that already run large-scale infrastructure for Israeli enterprises and the public sector.

Behind the product engineering sits Altostratus, the development team that builds GPUPilot in partnership with Bynet.

Our approach

Where we work

Israel-based. The Connected service is hosted in reputable cloud infrastructure with per-customer database isolation. The Air-Gap deployment runs entirely inside customer perimeters and is used today on multiple production GPU clusters.

Contact

Sales, procurement, support, security: computingIT@bynet.co.il. Careers: bynet.co.il/en/career.

← Back to GPUPilot

Start monitoring in 30 seconds

One kubectl. Read-only. Works on any NVIDIA GPU cluster on Kubernetes.

Book a demo