Product overview

Run VMs, containers, and AI on one platform.

Starlight is built by Mainsail Industries for organizations operating where infrastructure must function with limited resources, unreliable networks, and no room for failure.

Definition

Edge-native infrastructure.

A unified platform to run virtual machines, containers, and AI workloads directly at the edge, with security and autonomy embedded into the system itself.

Unified
VMs, containers, AI. One control plane.
Edge-first
Operates without central coordination.
Autonomous
Security and resilience built in.

Why now

Built for scale, not for constraint.

Two decades of datacenter infrastructure assumed reliable networks, abundant compute, and dedicated teams. The work moved to the edge, where those assumptions break.

2000sDatacenter
Virtualization / Consolidation
Workloads consolidated in centralized environments. Assumed reliable networks and abundant compute.
2010sDatacenter
Cloud-native / Orchestration
Containers and Kubernetes scaled fleets to thousands of machines. Built for constant coordination, not constraint.
2020sEdge
Intelligence / Modern AI
Inference moves close to users and data, where centralized assumptions about network, scale, and trust break down.

Edge reality: one or two nodes, intermittent networks, constrained compute. The datacenter playbook does not fit.

Unified by design

One API, every workload.

Desktop, CLI, or automation, all pointing at the same control plane.

Workloads

Virtual machines
Containers
AI workloads

Interfaces

Desktop app
CLI
Automation / API

Identity: Same auth, same roles, across every workload type.
Policy: One enforcement model for VMs, containers, and AI alike.
Lifecycle: Deploy, observe, and retire with one set of primitives.

VMware take-out

Off VMware, on a modern platform.

Run the same VMs, plus containers and AI, under pricing you can predict and a license you can understand.

Axis	Today, on VMware	On Starlight
Pricing	Repriced; double-digit increases common	Predictable, per node. No surprise uplifts.
Licensing	Core / socket-pair counting, SKU sprawl	No core or socket-pair counting.
Renewal	Forced bundles; features you don't use	One SKU. What you run is what you pay for.
Scope	VMs only; containers and AI elsewhere	VMs, containers, and AI on one platform.

Workload compatible
Existing Windows and Linux VMs run on Starlight. Bring images as-is.
Feature compatible
Live migration, HA, snapshots, networking. The feature set teams rely on, natively.
Move at your pace
Run both during transition. Consolidate when you're ready. No forklift.

Containers

Containers without Kubernetes complexity.

A different orchestration model. Every node is autonomous: it tracks its own resources and shares state peer-to-peer with the cluster. No brittle central control plane.

Kubernetes

star · every node depends on the control plane

If the control plane is unreachable, the cluster drifts.

Starlight

peer mesh · every node autonomous, state shared by protocol

Each node runs on its own. The cluster stays consistent.

No control plane to maintain
No etcd, no scheduler, no operator sprawl. Less to run, less to break.
Runs disconnected
Nodes operate independently when the network drops. State reconverges on its own.
Same artifacts
Existing OCI images run as-is. Helm charts and Kubernetes workloads replatform incrementally.

Batteries included.

Models, drivers, protocols, and agent control ship as part of the platform. No separate AI stack, no glue work.

Models are untrusted code. Starlight enforces that.

Sandboxed execution with resource limits
Data paths and egress controlled by policy
Full observability on every run

Ship AI like you ship apps
Models are delivered as OCI container images. Same packaging, same lifecycle, same primitives as every other workload.
Ready on first boot
GPU drivers, accelerator runtimes, and inference toolchains preconfigured and version-matched. No kernel-module roulette.
Agents speak natively
A built-in Model Context Protocol server exposes Starlight to AI agents over a standard interface.
AI operates Starlight
Use agents to deploy, observe, and manage infrastructure, not only workloads. The control plane is programmable end to end.

Edge

Autonomy, not central control.

Each location operates on its own. Workloads keep running during disconnection. State synchronizes when connectivity returns.

2-node

High availability

First-class HA without a three-node cluster. Matches edge reality, not datacenter baseline.

0-ops

Disconnected operation

Nodes function independently when the network drops. No central system needed to keep workloads alive.

1×

Platform, every scale

Single node to distributed fleet, same operating model. Start small and expand without rework.

Security

Part of the architecture.

Integrated identity, policy enforcement, a hardened OS foundation, and runtime protection for every workload, including AI.

Foundation

Verified and immutable

Image-based OS, read-only root
Signed builds with full attestation
No runtime drift from deployment
DISA STIG and FIPS 140-3 alignment

Runtime

Enforced, not observed

Process-level visibility and control
Filesystem and network policy
Capability and syscall controls
Identity-driven, real time

Advanced

Ready for what's next

Confidential computing support
Quantum-resilient cryptography
AI workloads sandboxed by default
Data and egress path controls

Adoption

Bring your workloads with you.

Most teams aren't starting from scratch. Starlight is a new operating model, not a rewrite of the one you already run.

Existing container images run as-is
Kubernetes workloads replatform incrementally
Helm charts adapt without rewriting apps
CI/CD keeps delivering the same artifacts
Run both models during transition
Standardize when you're ready