Deployments / Self-Hosted

ManyLayers on your Kubernetes cluster — fully air-gapped.

One Helm chart. Deploy ManyLayers to your own infrastructure, activate with a license file, and operate with complete data sovereignty — no outbound connections required after installation.

What you get

Full control, no compromises.

Minimal operational footprint

Gateway, Workspace, and the Deploy module ship as one deployable unit. Helm chart available for Kubernetes. No microservices to coordinate, no dependency sprawl.

Air-gapped operation

After the initial image pull, ManyLayers requires no outbound internet connections. License file activation works completely offline. All inference and logs stay on-premises.

Full data sovereignty

Provider API keys, embeddings, vector indexes, conversation history, and audit logs never leave your infrastructure. No telemetry, no usage callbacks.

On-prem model serving

Deploy open-weight models to your GPU nodes via the Deploy module. Models register automatically as Gateway routing targets — same API surface as cloud providers.

Kubernetes native

Helm chart with configurable resource quotas, GPU node affinity, and horizontal pod autoscaling. Rolling upgrades with zero downtime included.

License file activation

License activation is a single file checked in at startup. No internet handshake required. Licenses are issued per cluster and tied to your organisation.

Compare deployment options

Cloud, Hybrid, and Self-Hosted side by side.

Feature SaaS Cloud Hybrid Self-Hosted
Infrastructure management Fully managed Control plane managed; data plane yours Your team
Data residency Our cloud (region-locked) Your VPC / on-prem Your data centre
Prompt & data egress Leaves your network Stays in your network Stays in your network
Upgrades Automatic Automatic (data plane) Self-managed
Air-gapped operation Not available Partial — control plane requires reach Supported
SSO / RBAC Included Included Included
Audit log Included, exportable Written locally to your storage Included, exportable
On-prem model deployment Not available Supported (Deploy module on your nodes) Supported (Deploy module)
Pricing model Subscription Subscription + data-plane infra cost License + your infra cost
Setup time Minutes Hours Hours to days
Requirements

What you need to run ManyLayers self-hosted.

Minimum compute 4 vCPU, 8 GB RAM (Gateway only); GPU nodes required for Deploy module
Kubernetes 1.27+ recommended; bare-metal Linux supported via docker compose
Database Postgres 15+ (provided as a dependency in the Helm chart)
Object storage S3-compatible (MinIO, AWS S3, GCS) for audit log export and model weights
License License file issued per cluster; offline activation, no internet required
Air-gap support Full — no outbound connections after initial image pull

Deploy ManyLayers on your infrastructure today.