VergeIQ: Orchestrate GPUs, Host Your Own LLMs and Put Them to Work
VergeOS, the private cloud OS, manages GPUs the same way it manages compute, storage and networking. Pass-through, NVIDIA vGPU and MIG run from one interface, and every GPU becomes a shared, protected infrastructure resource. On that foundation, VergeIQ hosts the LLMs you choose on hardware you own, and your organization puts them to work.
GPUs Become Infrastructure, Not a Specialty
GPU hardware is easy to buy. Running it is the hard part. Driver compatibility, MIG partitioning at the command line and GPU-aware disaster recovery sit outside the standard sysadmin skill set, so many projects stall waiting for a specialist.
VergeOS removes that barrier. GPUs are first-class VergeOS resources, scheduled, monitored, snapshotted and replicated alongside compute and storage, and managed from the same interface or the REST API.
- Pass-through, NVIDIA vGPU and MIG from one point-and-click form, with no nvidia-smi
- Upload the driver once and VergeOS deploys it, bit for bit, to every guest VM
- Snapshots and replication include each VM’s GPU configuration, so DR includes the GPUs
- NVIDIA supports VergeOS as a vGPU platform, with joint support from both companies
One driver upload covers every VM. MIG splits a GPU into hardware-isolated slices, each serving its own workload.
| GPU mode | Best for | Sharing | Isolation |
|---|---|---|---|
| Pass-through | Full-GPU simulation, scientific compute, large model hosting | One VM per GPU | Complete |
| NVIDIA vGPU | VDI, multi-user inference, developer GPU access | Time-sliced across VMs | Software-level |
| MIG | Multi-tenant and regulated workloads with guaranteed resources | Hardware-partitioned slices | Enforced in silicon |
| MIG + time slicing | Maximum VM density with per-tenant isolation | Time slicing inside each MIG slice | Hardware-enforced at the MIG tier |
NVIDIA vGPU software licenses are required for vGPU operation and are available through NVIDIA-authorized partners. See GPU infrastructure on VergeOS for supported hardware.
Run the Models You Choose on Hardware You Own
With the GPUs orchestrated, VergeIQ turns them into a private AI service. It is part of VergeOS, so it is ready once the platform is installed. VergeIQ hosts popular open models such as Llama, Qwen, Phi and Gemma, allocates GPU resources to them dynamically and loads models automatically.
Every model sits behind an OpenAI-compatible service. Applications built on OpenAI libraries and tools such as VS Code, Continue and AnythingLLM point at an internal endpoint and work without code changes.
- Prompts, records and model weights stay on your cluster
- Run fully disconnected, with no internet or cloud connection
- No per-token bill and no egress fee, a capital cost you size
- You decide which models run and when they change
VergeIQ loads the models onto the GPU pool and serves them through one OpenAI-compatible endpoint.
Three Ways to Use Your Private LLMs
Once the models run on your own GPUs, they serve the infrastructure, the people who run it and everyone else in the organization.
Automate the infrastructure
An AI assistant works the platform through Verge CLI, within limits your administrators set.
02 · ProductivityMake the IT team more productive
Faster root cause, a shorter learning curve and quicker migrations.
03 · The OrganizationRoll AI out to everyone
Private models for every user, with no token meter and data that stays home.
Automate the Infrastructure in Plain Language
Verge CLI ships and is supported as part of VergeOS. It has three parts. The Verge CLI maps to the full VergeOS API, an MCP server connects an AI assistant to the environment over the open Model Context Protocol, and agent skills supply the know-how of an experienced VergeOS engineer.
Together they let Claude Code, OpenAI Codex or a local model you run on VergeIQ build networks, deploy workloads and set up environments from a plain-language request. Administrators decide what the assistant runs on its own and what waits for approval, and it never acts outside those limits.
- One request spans compute, storage, networking and data protection
- Agent skills apply the firewall rules and design an engineer would
- Pairs with Packer, Terraform and Ansible for code you keep in Git
A request flows through the MCP server and agent skills to Verge CLI. Changes outside the approval limit wait for an administrator.
Give the IT Team Its Time Back
The same assistant that automates the platform makes the people running it faster. It reasons against the VergeOS documentation through the MCP server, so its answers follow how the platform works instead of a model’s guess. With one API across the stack, it traces a problem across compute, storage and networking in one pass.
- Root cause from real log lines, across the whole stack
- Advanced features explained in plain language, so the team uses more of the platform sooner
- Guidance through VMware migration and the validation that follows
- Routine work like VM creation, workload moves and network changes handed off
The administrator stays in charge. The assistant gives the team a faster read on the platform and proposes the work, and your people decide what runs.
The assistant connects symptoms across layers and hands the administrator a root cause and a proposed fix.
Private AI for Everyone, at a Cost You Set
IT exists to deliver services to the business, and AI is now one of them. Roll the models you host out to every department through the same internal endpoint. Finance, legal, engineering and support use the tools they already know, and nothing they type leaves the building.
- Analyze documents, review proprietary code and draft content against internal data
- Prompts, records and model weights stay on your cluster, and VergeIQ can run fully disconnected
- No per-token bill and no egress fee, an open-ended operating cost becomes a capital cost you set
- One GPU pool serves IT and the business, sized by your team
The same GPUs that help run the infrastructure now serve the whole company. Hardware, power and people remain real costs. What goes away is the token meter.
Departments reach private models through one internal endpoint. The token meter and the egress fee stay outside.
Renting AI Compared to VergeIQ
| Question | Public cloud AI | VergeIQ on VergeOS |
|---|---|---|
| What does it cost? | Per-token billing plus egress, with no ceiling you set | A capital cost you size, with no token meter and no egress fee |
| Where does the data sit? | On the provider’s hardware, governed by its terms of service | On your cluster, behind your firewall, with a fully disconnected option |
| Who picks the models? | The provider decides price, access and availability | You choose open models and decide when they change |
| Can it help run IT? | Separate tools with no view into your platform | Verge CLI, the MCP server and agent skills connect AI to the whole platform through one API |
| Who runs the GPUs? | The provider, on its hardware | Your IT team, through the same VergeOS interface as compute, storage and networking |
Resources
Run Your Infrastructure by Conversation
How Verge CLI, the MCP server and agent skills work together.
On-Demand WebinarPrivate, Enterprise AI Demonstration
Watch VergeIQ deliver private generative AI on VergeOS.
GPU InfrastructureGPU Infrastructure Without the Complexity
Pass-through, vGPU and MIG on VergeOS, validated with NVIDIA.
Frequently Asked Questions About VergeIQ
What is VergeIQ?
VergeIQ is the private AI capability of VergeOS. VergeOS orchestrates the GPUs in your servers, and VergeIQ hosts the LLMs you choose on them, so the models serve your infrastructure, your IT team and the rest of the organization on hardware you own.
Which GPU modes does VergeOS support?
VergeOS supports NVIDIA GPU pass-through, NVIDIA vGPU and Multi-Instance GPU (MIG), plus MIG with time slicing, all from one interface or the REST API. NVIDIA vGPU software licenses are required for vGPU operation and are available through NVIDIA-authorized partners.
Do I need GPU specialists to run it?
No. GPUs are managed through the same VergeOS interface as compute, storage and networking. MIG configuration is point-and-click with no nvidia-smi, drivers are uploaded once and deployed to every guest VM, and snapshots and replication include each VM's GPU configuration.
Which models can VergeIQ host?
VergeIQ hosts popular open models such as Llama, Qwen, Phi and Gemma. You choose the models and decide when they change. Model choice and workload drive GPU sizing, which a VergeIO technical design report covers for your environment.
Can our existing AI tools connect to VergeIQ?
Yes. VergeIQ provides an OpenAI-compatible service. Tools such as VS Code, Continue and AnythingLLM, and applications built on OpenAI libraries, point at an internal endpoint and work without code rewrites.
Does my data leave the building?
No. Prompts, records and model weights stay on your VergeOS cluster, behind your firewall. VergeIQ can run fully disconnected, with no internet or cloud connection required.
Is private AI free?
No. Hardware, power, GPUs and the people who run them are real costs. VergeIQ removes the per-token bill and the egress fee, turning an open-ended operating cost into a capital cost you size and control.
How does AI help automate VergeOS?
Verge CLI ships as part of VergeOS with three parts: a command-line interface that maps to the full VergeOS API, an MCP server built on the open Model Context Protocol and a set of agent skills. Together they let an AI assistant, including a local model hosted on VergeIQ, build networks, deploy workloads and set up environments from a plain-language request.
Does the AI change my infrastructure on its own?
Only within the limits your administrators set. Administrators decide which actions the assistant runs on its own, which need approval and which it cannot take at all. The assistant never acts outside those limits.
Does VergeIQ train models?
No. VergeIQ runs inference on the models you select. Training is a different workload with different hardware demands. VergeIQ hosts the models, serves them to your users and keeps the data behind your firewall.
Take Control of Your AI
Run private AI on hardware you own, put it to work for the IT team and roll it out to the whole organization at a cost you set.