Skip to content

ApplicationSets (autonomous agents)

Overview

ApplicationSets work with argocd-agent by generating Applications that are synced through the normal agent protocol. The argocd-applicationset-controller can run on the control plane (for managed agents) or on workload clusters (for autonomous agents). In both cases, only the resulting Application resources are synced between principal and agent -- ApplicationSet resources themselves are never synced.

Autonomous Mode: ApplicationSets on Workload Clusters

In autonomous mode, run the argocd-applicationset-controller on the workload cluster alongside the agent and application controller. The controller generates Applications locally, and the agent syncs them back to the principal for central observability.

Local generators such as list, git, pull request, and matrix work as expected. Generators that require cross-cluster state (e.g., the cluster generator spanning multiple clusters) are limited to the local cluster's perspective.

The Kind getting-started guide includes an optional install step for deploying the applicationset-controller on autonomous agents.

Limitations

  • ApplicationSet resources are not synced. Only the Application resources generated by an ApplicationSet are synced between agent and principal. The ApplicationSet resource itself stays on whichever side created it.
  • No cross-side generators in autonomous mode. An autonomous agent's applicationset-controller only sees its own cluster, so generators that depend on multi-cluster state (e.g., cluster generator listing all clusters) are scoped to the local cluster.