Kubernetes (Helm)

The official Helm chart deploys HEBBS as a StatefulSet with persistent storage, health checks, and optional monitoring integration.

Prerequisites

Kubernetes 1.26+
Helm 3.12+
A StorageClass that supports dynamic provisioning

Quick Install

helm install hebbs oci://ghcr.io/hebbs-ai/charts/hebbs \
  --namespace hebbs \
  --create-namespace

Custom Values

Create a values.yaml to override defaults:

replicaCount: 1

image:
  repository: ghcr.io/hebbs-ai/hebbs
  tag: "0.1.0"

server:
  grpcPort: 50051
  metricsPort: 9090
  logLevel: info

storage:
  size: 50Gi
  storageClass: gp3

resources:
  requests:
    cpu: "2"
    memory: 4Gi
  limits:
    cpu: "4"
    memory: 8Gi

config:
  embedding:
    batchSize: 32
    acceleration: auto
  index:
    hnswM: 16
    hnswEfConstruction: 200
    hnswEfSearch: 100
  reflect:
    provider: openai
    model: gpt-4o-mini
    enabled: true

tls:
  enabled: false
  certSecretName: hebbs-tls

auth:
  enabled: false
  tokenSecretName: hebbs-auth

serviceMonitor:
  enabled: false

Install with custom values:

helm install hebbs oci://ghcr.io/hebbs-ai/charts/hebbs \
  --namespace hebbs \
  --create-namespace \
  -f values.yaml

Values Reference

Core

Key	Default	Description
`replicaCount`	`1`	Number of replicas (single-writer architecture)
`image.repository`	`ghcr.io/hebbs-ai/hebbs`	Container image
`image.tag`	Chart appVersion	Image tag

Storage

Key	Default	Description
`storage.size`	`20Gi`	PVC size for data volume
`storage.storageClass`	`""` (default)	Kubernetes StorageClass

Resources

Key	Default	Description
`resources.requests.cpu`	`1`	CPU request
`resources.requests.memory`	`2Gi`	Memory request
`resources.limits.cpu`	`2`	CPU limit
`resources.limits.memory`	`4Gi`	Memory limit

TLS

Key	Default	Description
`tls.enabled`	`false`	Enable TLS for gRPC
`tls.certSecretName`	`""`	Kubernetes Secret containing `tls.crt` and `tls.key`

Monitoring

Key	Default	Description
`serviceMonitor.enabled`	`false`	Create Prometheus ServiceMonitor
`serviceMonitor.interval`	`30s`	Scrape interval

Scaling Considerations

HEBBS uses a single-writer architecture. Horizontal scaling is planned for future releases. For now:

Vertical scaling — increase CPU and memory resources for higher throughput.
Storage scaling — expand PVC size as memory count grows.
Read replicas — not yet supported; planned for a future release.

Persistence

The Helm chart creates a PersistentVolumeClaim for HEBBS data. Data survives pod restarts and rescheduling. To ensure data durability:

Use a StorageClass with reclaimPolicy: Retain
Configure regular backups (see Production Checklist)

Upgrading

helm upgrade hebbs oci://ghcr.io/hebbs-ai/charts/hebbs \
  --namespace hebbs \
  -f values.yaml

The chart performs a rolling update. HEBBS handles in-place upgrades with automatic data migration when needed.

Uninstalling

helm uninstall hebbs --namespace hebbs

The PVC is not deleted by default. To also remove data:

kubectl delete pvc -n hebbs data-hebbs-0