finds.dev← search

// the find

argoproj-labs/argocd-agent

★ 539 · Go · Apache-2.0 · updated Jun 2026

Redefining the multi cluster story of Argo CD

argocd-agent extends Argo CD with a pull-based multi-cluster architecture: lightweight agents running on remote clusters connect outbound to a central hub, instead of the hub reaching into each cluster. This inverts the network topology, which matters a lot for edge, air-gapped, or NAT-heavy environments. It's under active development from the Argo community and still pre-GA.

1. The pull model genuinely solves a real operational problem — outbound-only connections from agents mean no VPN, no firewall holes, no exposed cluster API servers. This isn't just a deployment preference, it changes what's physically possible in edge and telco scenarios. 2. Two distinct agent modes (managed vs. autonomous) with documented tradeoffs, not a one-size-fits-all design. You can mix them in the same fleet, which reflects how real organizations actually work. 3. mTLS everywhere by default, certificate-based agent identity — the security model is baked in, not bolted on. The protocol design doc shows they thought about authentication seriously. 4. Solid CI hygiene: codecov, golangci-lint, integration test workflow, CodeQL, and a chaos tester utility in hack/. That's not just README decoration.

1. Still pre-GA with no committed timeline — the README explicitly says 'we kindly ask for help to battle-test it.' For production multi-cluster GitOps, that's a real risk. Check the milestones page before committing to this for anything critical. 2. The Redis dependency per agent (visible in dev-env kustomizations) adds operational overhead at the edge. If you're deploying to resource-constrained nodes or factory floors, running Redis alongside the agent is not trivial. 3. No native Helm chart for the control plane in a single install — setup involves separate principal and agent charts plus Argo CD itself, with multiple layered kustomizations. The getting-started path is non-trivial. 4. The hub is a single point of failure for observability: if the principal goes down, agents keep running but you're flying blind. The HA docs exist but it's an additional operational concern that vanilla Argo CD doesn't force on you.

View on GitHub → Homepage ↗

// want more like this?

We dig through GitHub every week and send a few repos picked for what you actually care about — each with an honest take like this one.

Get finds in your inbox → Search again →