Kubernetes (Helm)

The Helm chart is the supported path for production deployments at scale. It handles worker autoscaling, pod-level isolation, and rolling deploys.

Add the repo

helm repo add tone https://charts.tonehq.dev
helm repo update

Install

helm install tone tone/tone \
  --namespace tone --create-namespace \
  --values values.yaml

A minimal values.yaml:

ingress:
  enabled: true
  host: tone.example.com
  tls:
    enabled: true
    issuer: letsencrypt-prod
 
api:
  replicas: 2
 
worker:
  replicas: 4
  autoscaling:
    enabled: true
    minReplicas: 4
    maxReplicas: 32
    targetCallQueueDepth: 20
 
postgres:
  external: true
  url: postgres://...
 
secrets:
  twilio:
    accountSid: ...
    authToken: ...
  openai:
    apiKey: ...

Worker autoscaling

The worker pods scale on call queue depth, not CPU. This matches the actual unit of work (a call) rather than a derived metric. The default target is 20 calls per worker — tune based on your turn-time characteristics.

Upgrades

The chart follows semver. Patch upgrades are zero-downtime by default; minor upgrades may require a migration step that's listed in the release notes. Major upgrades are rare and always documented.

Observability

Every component emits OpenTelemetry traces and Prometheus metrics. The chart ships a ServiceMonitor for Prometheus Operator and ready-made Grafana dashboards in deploy/dashboards/.