Platform / Cloud-Native Infrastructure

Kubernetes-native, standard dependencies, air-gap ready.

ManyLayers deploys via Helm onto your existing Kubernetes clusters. Standard dependencies — Postgres, Redis, Kafka, Qdrant — and zero exotic infrastructure. Air-gap capable, horizontally scalable, and upgradeable with standard Helm commands.

Book a demo Self-hosted docs

Kubernetes-Native Deployment

ManyLayers ships as a Helm chart. Every component — gateway, workspace, inference serving, vector store — runs as standard Kubernetes workloads. Horizontal Pod Autoscaler handles scale-out. No custom operators or non-standard resources required.

  • Helm chart with values for single-node and multi-node topologies
  • HPA configured for gateway replicas based on request throughput
  • Supports EKS, GKE, AKS, and bare-metal Kubernetes equally

Kubernetes-Native Deployment — key behaviours

01 Helm chart with values for single-node and multi-node topologies
02 HPA configured for gateway replicas based on request throughput
03 Supports EKS, GKE, AKS, and bare-metal Kubernetes equally

Standard Dependencies Only

The full ManyLayers stack runs on Postgres, Redis, Kafka, and Qdrant. These are widely understood, mature, and available as managed services on every major cloud. No proprietary databases, no exotic message queues, no data planes that require specialised skills to operate.

  • Postgres for config, audit log, and semantic cache
  • Redis for rate limiting counters and session state
  • Qdrant for vector storage and similarity search

Standard Dependencies Only — key behaviours

01 Postgres for config, audit log, and semantic cache
02 Redis for rate limiting counters and session state
03 Qdrant for vector storage and similarity search

Air-Gap Ready

All container images are publishable to a private registry. The ManyLayers gateway operates without any outbound network calls — no telemetry, no license checks, no upstream callbacks. Kafka and object storage connections are internal. Verified on fully air-gapped networks.

  • All images published to your private registry before deploy
  • Zero outbound connections required after initial install
  • Model weights cached locally — no internet access during inference

Air-Gap Ready — key behaviours

01 All images published to your private registry before deploy
02 Zero outbound connections required after initial install
03 Model weights cached locally — no internet access during inference

Upgrade Path

ManyLayers follows a versioned Helm chart release cycle. Upgrades apply via standard Helm upgrade commands with pre-upgrade hooks that validate schema migrations. Rolling deployments ensure zero-downtime upgrades for the gateway and workspace components.

  • Rolling upgrades with readiness probes prevent traffic routing to unready pods
  • Pre-upgrade hooks run schema migrations and validate config compatibility
  • Rollback to prior chart version restores previous state cleanly

Upgrade Path — key behaviours

01 Rolling upgrades with readiness probes prevent traffic routing to unready pods
02 Pre-upgrade hooks run schema migrations and validate config compatibility
03 Rollback to prior chart version restores previous state cleanly
Component matrix

Components, dependencies, scaling, and air-gap support.

Component Dependencies Scaling mechanism Air-gap capable
Gateway Postgres, Redis HPA / replicas ✓
Workspace Postgres, Redis, Kafka HPA / replicas ✓
Vector store Qdrant StatefulSet ✓
Inference serving GPU nodes (Kubernetes) HPA / GPU metrics ✓
Object storage S3-compatible or MinIO Managed / self-hosted ✓

Deploy ManyLayers on your Kubernetes infrastructure today.

Book a demo Self-hosted docs