Skip to main content

kubeswarm - Orchestrate Agents on Kubernetes

Orchestrate AI agents at swarm scale.

kubeswarm is a Kubernetes operator that manages LLM-powered agents as first-class resources. Define your agents in YAML, connect MCP tools, deploy with kubectl apply and operate with the same Kubernetes tooling you already use for services.

What kubeswarm does for agent teams​

  • Manages agent pods - replicas, KEDA autoscaling, health checks, rolling restarts
  • Connects MCP tools - bearer auth, mTLS, per-tool trust levels, custom headers
  • Orchestrates agent teams - pipeline DAGs, dynamic delegation, LLM-routed dispatch
  • Enforces guardrails - tool allow/deny lists, token budgets, network policies
  • Tracks every execution - immutable run records with token usage and step outputs
  • Integrates natively - OpenTelemetry, Prometheus, KEDA, Kubernetes RBAC

Install kubeswarm on Kubernetes​

helm repo add kubeswarm https://kubeswarm.github.io/helm-charts
helm install kubeswarm kubeswarm/kubeswarm \
  --namespace kubeswarm-system --create-namespace

Deploy your first agent on Kubernetes​

apiVersion: kubeswarm.io/v1alpha1
kind: SwarmAgent
metadata:
  name: hello-agent
spec:
  model: claude-sonnet-4-6
  prompt:
    inline: "You are a helpful assistant."
  apiKeyRef:
    name: provider-api-keys
    key: ANTHROPIC_API_KEY
kubectl apply -f agent.yaml
kubectl get swagent -w

Next steps​