VibOps Request access
GPU · VM · Agents · Sovereign

From code to GPU in one conversation.

Deploy, monitor, govern and bill your applications and your agents on any GPU, any VM, any cloud — driven by the model you choose, including one that never leaves your network. One layer above your toolchain, not another tool in it.

Bring your own LLM On-prem Air-gap Multi-tenant SOC 2 audit trail MCP server & SDK: MIT
The VibOps console: GPU cluster health, cost per tenant, and the conversation thread that runs the actions.
NVIDIA Inception Program member
33
connectors — accelerators, clusters, hypervisors, clouds
11
operations identical on NVIDIA, AMD, Intel and Groq
0
inbound ports opened on your network
The problem

Capacity scales. Operations don't.

Buying accelerators is a purchase order. Running them is a team. Capacity now arrives faster than the ability to operate it — so the estate is held back by the operator, not by the hardware. And an AI estate is never only accelerators: services on VMs, inference on GPUs, on-prem and in cloud. Every addition doubles a line below.

A second vendor A second toolchain, a second runbook, a second on-call rotation. The team that knows one stack does not know the other.
A new site Another inventory, another metrics stack, another set of credentials — and no single view of what is actually running.
More tenants Isolation to prove, quotas to enforce, and spend to attribute. Usually reconciled in a spreadsheet, after the invoice.
More GPUs Hours nobody has. Capacity sits idle not because it is unavailable, but because nobody has time to place work on it.

One control plane. The same operations, whatever is underneath.

How

One request. Four vendors. No translation layer in your team's head.

Eleven operations carry the same name, the same payload, the same policy check and the same audit record on NVIDIA, AMD, Intel and Groq. The vendor difference lives below the line, where it belongs.

accelerator_list_devices
NVIDIA H200 · 141 GB nvidia.com/gpu
AMD MI300X · 192 GB amd.com/gpu
Intel Gaudi 3 · 96 GB gaudi.intel.com/accelerator
Groq LPU inference endpoints api.groq.com
Who it's for

Teams that own their infrastructure

GPU operators

cloud · datacenter · HPC

Turn raw compute into a managed platform. White-label console, per-tenant chargeback with configurable margin, enforceable budgets, multi-accelerator. Your customers deploy self-service; the billing follows.

Integrators and MSPs

multi-site managed services

Operate your customers' VMs — Proxmox, vSphere, XCP-ng — and their GPU clusters from a single console. Proactive anomaly detection, backup compliance, SLA tracking. One operator holds ten sites instead of three.

Enterprises

on-prem · cloud · hybrid

Operate GPU and VM without demanding Kubernetes expertise at every level. SOC 2 audit trail, multi-vendor fleet, the same gestures everywhere.

The proof

Watch it work

No mockup: the real console, on real infrastructure.

deploymentGPU monitoringcost per tenantaudit trail
Capabilities

What the console holds

GPU Ops
Inventory, health, saturation, anomaly detection and failure prediction across NVIDIA, AMD and Intel — with no per-machine agent.
VM Ops
Proxmox, vSphere, Vates XCP-ng, Nutanix. Lifecycle, GPU passthrough, backup compliance, SLA tracking.
FinOps
One cost model for GPU and VM alike. Per-tenant chargeback, configurable margin, budgets that block before the invoice does.
Agent FinOps
An inference proxy: cost per agent, per model and per organisation, with ceilings that actually stop the spend.
Your own LLM
The Infrastructure Agent runs on the model you choose — Claude, Ollama, NVIDIA NIM or any OpenAI-compatible endpoint, including one served from your own GPUs. The agent holds no credential: it emits an intent that VibOps authenticates and policy-checks, so changing model does not change what is exposed.
Governance
Deny by default on any unknown action, mandatory preview before destruction, hash-chained and anchored audit trail.
Sovereignty
On-prem, air-gap compatible, tenant isolation enforced by the database rather than by code discipline.
Silicon

It speaks to what you already have

NVIDIA H100 · H200 · Blackwell AMD MI300X Intel Gaudi 3 AWS Trainium Google TPU Groq LPU Cerebras WSE
KubernetesSlurmProxmoxvSphere Vates XCP-ngNutanixHPE VMERedfish
vLLMOllamaNVIDIA NIM Anthropicany OpenAI-compatible endpoint
Getting started

Three steps

VibOps sits beside your infrastructure, not in front of it. No agent on the compute machines.

01

Deploy the control plane

A self-hosted stack, on your own ground.

$ curl -fsSL https://vibops.ai/install.sh | bash
02

Connect your sites

One lightweight gateway per site. It discovers the clusters, the GPUs and the hypervisors on its own — and proposes nothing you have not confirmed.

03

Operate

From the console, or from your editor over the MCP server. The MCP server and the Python SDK are public and MIT-licensed: vibops-mcp and vibops-sdk.

$ pip install vibops-mcp
Access

You operate GPUs.
Let's talk.

We open access in waves, so that every install gets attention. Tell us what you run.

No cold outreach. A personal reply.