Helm Control

Helm NodeLocal inference management

One machine.
A capable foundation.

Manage your local AI workloads from a calm, focused workspace. Your models, runtimes and connections, together.

Prepare your first nodeSee what’s inside
Technical illustration of a workstation and an open-top rackmount server with visible motherboard, memory, cooling and GPU.

Model library

Register model files, manage downloads and keep variants and capability information organized.

Runtime profiles

Save GPU placement, context and concurrency. Start, observe and drain supported model services.

API access

Connect your applications with scoped inference tokens and clear endpoint details.

FROM MODEL TO CLIENT

Four steps.
One connected workflow.

  1. 01

    Register a model

    Add a supported model from your mounted storage or download a model file.

  2. 02

    Configure a profile

    Choose the engine, GPU placement, context and concurrency.

  3. 03

    Start an endpoint

    Bring the profile online and check readiness and logs.

  4. 04

    Connect your client

    Create a scoped API token and use the connection details.

CAPABILITIES, CLEARLY DEFINED

Built on a
practical foundation.

Model and GPU compatibility depends on the runtime, recipe, architecture, format and available memory.

Current capabilities
  • Local model and runtime management
  • GPU placement and memory reservations
  • Scoped API access and persistent usage history
PLANNED

Broader machine and process management, alongside fleet coordination.

Product direction, not a release commitment.

RUNTIME COMPATIBILITY

Model choice.
Details in view.

Choose a compatible engine and recipe for the model and hardware you actually have.

llama.cpp

Default engine

Supported GGUF models with matching backend runtimes. Deployment examples cover Intel SYCL and NVIDIA CUDA.

vLLM

Restricted recipes

Configured GGUF and prepared Flash-Next NVFP4 recipes through a separately configured controller. Model and GPU restrictions apply.

YOUR FIRST NODE

Bring your hardware online.

Start with the host, runtime and access requirements.

Read the setup guide

BEFORE YOU START

A few practical answers.

Does Helm Node manage more than one machine?

Helm Node manages model services on one machine. Helm Fleet provides a separate read-only view of enrollment, licensing capacity and audit activity. Fleet-wide orchestration is planned.

Can I run any model on any GPU?

Compatibility depends on the runtime, recipe, architecture, format, quantization and memory. Registering a model does not establish that it will run on your hardware.

How do applications connect?

Use the Connections view to find the shared or profile-specific inference URL, then create a token with the required scopes. Supported API behavior depends on the runtime and model.

Where can I see the current interface?

View the current Helm Node interface, captured locally with sample data. The application has its own release cycle.

PART OF SOMETHING CONNECTED

The Helm family
Helm FleetHelm NodeHelm ComputeHelm RuntimeHelm VaultPlannedHelm FlowPlanned