Agent Installation

The PulseStream agent is a lightweight service that runs in your Kubernetes cluster. It collects telemetry (logs, events, metrics) and streams it to the PulseStream control plane for AI-powered analysis.

Prerequisites

Kubernetes 1.27+
Helm 3.10+
A PulseStream account with a workspace

Platform support

All PulseStream agent images are published as multi-architecture manifests supporting both linux/amd64 and linux/arm64. No special configuration is required. Kubernetes automatically pulls the correct image for your node architecture.

Architecture	Supported	Examples
x86_64 (amd64)	Yes	Standard EC2 instances, GCE, AKS nodes
ARM64 (aarch64)	Yes	AWS Graviton (m7g, c7g, r7g), GKE Tau T2A, Ampere

Mixed-architecture clusters (amd64 and arm64 nodes) work out of the box. All bundled components (PostgreSQL, OpenSearch, Vector, NATS) also support both architectures.

Provisioning flow

PulseStream uses token-based provisioning — you never need to copy-paste API keys:

Generate a provisioning token in the PulseStream dashboard (valid for 30 minutes)
Apply the YAML manifest to your cluster (creates a Kubernetes Secret)
Install the agent via Helm — it reads the secret automatically
The agent exchanges the token for permanent credentials and connects

Install with Helm

Step 1: Generate a provisioning token

Go to Settings → Connection in the PulseStream dashboard and click Generate Provisioning Token. Copy the YAML manifest.

Step 2: Apply the provisioning secret

kubectl apply -f agent-provision.yaml

Step 3: Install the agent

helm install pulsestream-agent oci://public.ecr.aws/b2g7w6t0/charts/pulsestream-agent \
  --version 0.4.0 \
  --namespace pulsestream-agent \
  --create-namespace

The agent reads the provisioning secret, exchanges it for permanent credentials, and connects to the control plane automatically.

Resource footprint

No configuration is required. The agent works out of the box with sensible defaults. Plan for approximately 2 CPU cores and 4 GiB memory available in your cluster, plus a small per-node overhead for the log-collection DaemonSet (~100m CPU and ~128 MiB memory per node).

Verify connection

After installation, check the agent status:

kubectl get pods -n pulsestream-agent
kubectl logs -n pulsestream-agent deployment/pulsestream-agent --tail=20

In the PulseStream dashboard, navigate to Settings → Connection. The status should show as connected.

Upgrading

helm upgrade pulsestream-agent oci://public.ecr.aws/b2g7w6t0/charts/pulsestream-agent \
  --version <new-version> \
  --namespace pulsestream-agent

Uninstalling

The standard uninstall removes the agent and every resource it owns — including all data.

helm uninstall pulsestream-agent --namespace pulsestream-agent
kubectl delete namespace pulsestream-agent

What this removes

Resource	Removed by
Deployments, StatefulSets, DaemonSets, Services	`helm uninstall`
ConfigMaps and Secrets owned by the chart	`helm uninstall`
ClusterRoles and ClusterRoleBindings	`helm uninstall`
PersistentVolumeClaims (PostgreSQL, OpenSearch, Analyzer) and the underlying PersistentVolumes	`kubectl delete namespace` (cascading)
`agent-provision` secret you applied in step 2 of install	`kubectl delete namespace`
The namespace itself	`kubectl delete namespace`

Verify the uninstall

# Namespace gone
kubectl get namespace pulsestream-agent

# No chart-owned cluster-scoped leftovers (should print nothing)
kubectl get clusterrole,clusterrolebinding | grep pulsestream-agent

If the namespace finalizer hangs for more than a minute, list what’s blocking:

kubectl get namespace pulsestream-agent -o json | jq '.status.conditions'

Next steps

Troubleshooting Having connection issues? Check common problems and solutions.