Nov 17, live webinar: ECS to EKS without freezing the roadmap for a quarter

Managed Argo CD Hosting in 2026: A Buyer's Checklist for Cost, SLAs, Credentials and Exit

What a platform team should check before handing its Argo CD control plane to Akuity, Amazon EKS, Azure, Red Hat or another vendor: how each one bills, what its SLA really covers, where your Git and cluster credentials end up, and what you keep when you leave.

Morgan Perry
Co-founder
NOV 18, 2026 · 18 MIN
Managed Argo CD Hosting in 2026: A Buyer's Checklist for Cost, SLAs, Credentials and Exit

Key points:

  • Managed Argo CD in 2026 means four different products. Akuity hosts the Argo CD UI and API and runs an agent in your clusters. The Amazon EKS capability runs Argo CD in the AWS control plane and only deploys to EKS clusters. The Azure Argo CD extension and Red Hat OpenShift GitOps run Argo CD inside your own cluster with vendor-managed packaging and updates. Compare them on where the controller runs.
  • Price the unit the vendor bills. EKS charges $0.03 per capability hour plus $0.0015 per Application hour, and counts one Application deployed to five clusters as five (AWS). Akuity Pro starts at $495 a month for 50 Applications, plus $99 per extra pack of 10 (Akuity). At 500 Applications that is roughly $570 a month on EKS and roughly $4,950 a month on Akuity Pro list price.
  • Read what the SLA covers. The EKS SLA covers "the EKS Cluster control plane run by AWS", 99.95% on the standard control plane (AWS). It does not name the Argo CD capability. Akuity's pricing page publishes no uptime figure. Get the number, the measurement and the credit in writing.
  • Credentials follow the architecture. With an agent model (Akuity), the application controller runs in your cluster and the vendor stores no cluster credentials. With the EKS capability, Argo CD runs in AWS and reaches your clusters through EKS access entries, and AWS's own example grants system:masters.
  • Your Applications are portable, the glue is not. Application, ApplicationSet and AppProject manifests move between any Argo CD. SSO and RBAC mappings, cluster registration by ARN, notification config, webhook secrets and vendor features (Kargo, AI agents, promotion dashboards) do not. Codefresh's GitOps Cloud is "no longer available" (Octopus), so plan the exit before you sign.

Qovery · Infrastructure control plane
A control plane for platform teams and their coding agents
Learn more

I have run Argo CD for our own infrastructure and watched a lot of customer platform teams run it. The pattern is always the same. Argo CD itself is free and very good. The work is everything around it: high availability, upgrades every quarter, SSO, RBAC per team, one more cluster to register, and a security advisory landing on a Tuesday.

So "should we pay someone to host it?" is a fair question. This article is the checklist I would use before signing with a managed Argo CD vendor. If you are still deciding whether to keep Argo CD at all, read our comparison of Argo CD alternatives that remove the control plane first. If you are weighing Argo CD against a commercial delivery platform, the Argo CD vs Harness breakdown covers licensing and policy. Here I assume you keep Argo CD and want someone else to run it.

What managed Argo CD options exist in 2026, and how do they differ?

There are four real options in November 2026: Akuity, the Amazon EKS capability for Argo CD, the Azure Argo CD extension for AKS and Arc, and Red Hat OpenShift GitOps. They differ most on one question, where the Argo CD components run, and that answer drives cost, credential exposure and exit effort.

Codefresh used to be the fifth name on this list. Octopus Deploy acquired Codefresh in February 2024, and its own page now states "The GitOps Cloud product is no longer available" (Octopus). Octopus now connects to Argo CD instances you run yourself: you "install an Octopus gateway in the same cluster as Argo CD" and Octopus adds environments, promotion rules and approvals on top (Octopus). That is a useful product, but it is not hosting.

OptionWhere the Argo CD UI and API runWhere the application controller runsDeployment targetsBilling unitPublished uptime commitment
Akuity PlatformAkuity's control plane (Akuity docs)Inside each managed cluster, via the agentAny conformant Kubernetes cluster with outbound HTTPSArgo CD instances plus Application packs (pricing)None on the pricing page
Amazon EKS capability for Argo CDAWS control plane, "not on your worker nodes" (AWS docs)AWS control plane"Only Amazon EKS clusters", registered by ARN (AWS docs)Capability hour plus Application hour (pricing)The EKS SLA covers the cluster control plane; the capability is not named (SLA)
Azure Argo CD extension (AKS and Arc)Your clusterYour clusterAKS and Azure Arc-enabled clusters (Microsoft Learn)Not published in the preview announcementPublic preview since 24 March 2026 (Microsoft)
Red Hat OpenShift GitOpsYour OpenShift cluster, as an operatorYour OpenShift clusterOpenShift and the clusters it managesPart of your OpenShift subscription (Red Hat)Red Hat support terms of your subscription
Self-hosted Argo CDYour clusterYour clusterAnything you can reachYour engineers' timeWhatever you build

Two of these are hosting in the usual sense (Akuity, EKS). The other two are packaging with lifecycle support: Microsoft and Red Hat ship and patch the bits, but Argo CD still runs on your nodes, uses your capacity and fails with your cluster. That is not worse. Many regulated teams prefer it. Just do not buy it expecting someone else to carry the pager for the control plane.

How do you compare the cost of managed Argo CD with running it yourself?

Compare at your real Application count, counted the way each vendor counts, then add the people cost of self-hosting. Managed Argo CD is cheap against engineer time at a few hundred Applications. Per-Application pricing can overtake a part-time engineer once you pass a few thousand.

What do the published prices look like?

The EKS capability bills two hourly components: $0.03 per Argo CD capability hour and $0.0015 per Argo CD Application hour in us-east-1. AWS's own example puts 100 Applications at $131.40 a month (AWS EKS pricing). The counting rule matters: "Each Application is counted per target cluster deployment - if you deploy one Application to 5 different clusters, this counts as 5 Applications", and every Application generated by an ApplicationSet counts as one.

Akuity Pro "starts at $495/month" and includes one Argo CD instance, one Kargo instance, 50 Argo CD Applications, 50 Kargo stages and 25 million AI tokens. Extra capacity comes in packs of 10 Applications, 10 Kargo stages and 5 million tokens at $99 a month. Enterprise is "contact sales", and self-hosted or on-premise Akuity is Enterprise only (Akuity pricing).

Red Hat OpenShift GitOps has no separate price on Red Hat's product page, which describes it as "a tool included in OpenShift" (Red Hat). You pay for OpenShift. The Azure extension had no published price in its preview announcement.

Monthly list price at...100 Applications500 Applications2,000 Applications
EKS capability (one capability, us-east-1, 730 h)$131$569$2,212
Akuity Pro (list, before any discount)$990$4,950$19,800
Self-hosted Argo CD, software$0$0$0

I computed the EKS rows from AWS's two published rates and the Akuity rows from the $495 base plus $99 per 10 Applications. Akuity Enterprise pricing is negotiated, and nobody at 2,000 Applications pays list. The table shows the shape: EKS scales almost linearly and stays cheap, Akuity bundles more product (Kargo, dashboards, AI agents) and costs more per Application.

What does self-hosting really cost?

Self-hosting costs you three things: nodes, upgrades and incident time. The software is free under Apache 2.0 and Argo is a CNCF graduated project.

  • Nodes. The HA install "will require at least three different nodes due to pod anti-affinity rule in the specs" (Argo CD docs). Above roughly a thousand Applications you also tune controller sharding and processor counts.
  • Upgrades. Argo CD ships a minor release "four times a year, once every three months", and "versions older than the three most recent minor versions are considered EOL and will not receive bug fixes or security updates" (Argo CD release process). Skip three releases and you are running an unpatched control plane with write access to every cluster.
  • Advisories. On 7 October 2026 the project published seven security advisories, four rated critical, including command execution in the repo-server through Kustomize (GitHub advisories). Someone on your team reads those the same day and patches.

People cost dominates. The US median wage for software developers was $135,980 in May 2025 (BLS). If Argo CD care and feeding takes a quarter of one engineer, that is about $34,000 a year in salary alone, before overhead and before the on-call weekends. At 100 or 500 Applications, both EKS and Akuity Pro cost less than that. At 2,000 Applications on Akuity list price, the gap closes, and the decision becomes about features and risk rather than money.

What do managed Argo CD vendors actually promise on SLAs and support?

Very little is published, so the SLA is the first thing to negotiate. Check four things: what component the uptime figure covers, how it is measured, what the credit is, and whether support response times are in the contract or on a marketing page.

Amazon EKS. The EKS SLA defines "EKS Cluster" as "the EKS Cluster control plane run by AWS". It commits to 99.95% monthly uptime on the standard control plane, measured in 5-minute intervals, and 99.99% on Provisioned Control Plane, with credits of 10%, 25% or 100% of EKS charges depending on how far uptime falls (AWS EKS SLA). The Argo CD capability runs in the AWS control plane, but the SLA text I read does not name capabilities. Ask your AWS account team in writing whether a capability outage counts against the EKS service commitment. Credits also only apply to future EKS charges, which is small money next to a day without deploys.

Akuity. The pricing page lists audit record retention (6 months on Pro, 1 year on Enterprise) but no uptime figure and no support response times (Akuity pricing). I could not find a public SLA document. Treat that as a contract item: put the uptime number, the support severity matrix and the incident notification window in the order form.

Microsoft. The Argo CD extension was announced in public preview on 24 March 2026, with opt-in "automatic patch releases" (Microsoft Tech Community). Azure previews generally come without an SLA, so check the extension's status on the day you buy before you put production on it.

Red Hat. Support comes through your OpenShift subscription, and each OpenShift GitOps release supports a specific range of OpenShift versions. For example, OpenShift GitOps 1.21.1 runs on OpenShift 4.18 through 4.22 (Red Hat release notes). Your OpenShift upgrade plan becomes your Argo CD upgrade plan.

Argo CD going down does not take your applications down. Running workloads keep running. What stops is change: no syncs, no rollbacks through Argo CD, no drift correction. So the uptime number matters less than two other questions. How fast can you roll back during a vendor outage? And does the vendor have a documented break-glass path, such as applying manifests directly with kubectl while the control plane is down?

Where do your Git and cluster credentials go in each architecture?

In an agent architecture, the cluster dials out and the vendor never holds cluster credentials. In a hub-and-spoke architecture, the central Argo CD holds a credential for every target cluster, so whoever runs the hub holds the keys to the fleet. Know which one you are buying, and know where your Git credentials and rendered manifests live.

Akuity (agent). Akuity hosts the Argo CD frontend, and "the Argo CD Application Controller, however, is running inside of the managed workload clusters". Its docs state "No cluster access means no cluster credentials" (Akuity architecture). The agent needs only outbound TCP 443 to Akuity's domains, with separate US and EU endpoints (Akuity networking). One detail to check: the location of the repo-server, ApplicationSet controller and Image Updater "is configurable". If the repo-server runs in Akuity's control plane, your Git credentials and your rendered manifests are processed there. If you render Secrets in manifests, that matters.

Amazon EKS capability (hub in AWS). Argo CD runs in AWS, so "your worker nodes don't need direct access to Git repositories or Helm registries" (AWS). Cluster access goes through an IAM capability role and EKS access entries on each target cluster. AWS's own multi-cluster example creates that access entry with --kubernetes-groups system:masters, followed by a note to "consider using more restrictive Kubernetes groups" for production (AWS considerations). Do that. AWS also connects to fully private EKS clusters for you, which removes a VPC peering project but means traffic to your private API servers is brokered by AWS.

Azure extension and OpenShift GitOps (in your cluster). Everything runs on your nodes. Microsoft's extension supports workload identity federation for ACR and Azure DevOps and Entra ID SSO (Microsoft Learn). Credential exposure looks like self-hosted Argo CD, because it is.

Open source agent mode. If you like the agent model but not the vendor, the argocd-agent project brings a principal-and-agents design to upstream Argo CD, with "managed" and "autonomous" modes where agents connect back to a central hub. You still run it yourself.

Exposure questionAkuityEKS capabilityAzure extension / OpenShift GitOpsSelf-hosted hub-and-spoke
Who can write to your clusters?Agent in each cluster, with its own ServiceAccountAWS-run Argo CD through access entries you grantArgo CD in your clusterCentral Argo CD with a credential per cluster
Cluster credentials stored by the vendorNo, per Akuity docsIAM role you create; AWS operates the service using itNoNo vendor; your hub stores them
Where Git credentials are usedWherever the repo-server runs (configurable)AWS control plane, from your accountYour clusterYour hub
Inbound network access to clustersNone, outbound 443 onlyNone to configure; AWS brokers private accessNoneHub must reach every API server
SSO optionsSSO configured in AkuityAWS Identity Center onlyEntra ID (Azure); OpenShift OAuthDex or OIDC of your choice
Data residency choiceUS or EU endpointsYour AWS RegionYour cluster's regionYours

Whatever you pick, the audit question remains yours. A managed control plane gives you a sync history, not the full chain from pull request approval to the change in the cluster. We wrote a separate guide on joining Git, Argo CD and Kubernetes audit logs to answer "who approved that deploy".

Already running several Argo CD instances?
Qovery connects the Argo CD instances on your clusters and shows their applications next to your other services, with logs, pod status, shell access and sync status, while deployments stay in Git and Argo CD. Talk to us about your setup.

What do you keep if you leave a managed Argo CD vendor?

You keep everything that lives in Git: Application, ApplicationSet and AppProject manifests, Helm charts, Kustomize overlays and your sync policies. You lose, or rebuild, everything the vendor configured around them. Count the rebuild list before you sign, because that list is your real lock-in.

AWS documents the migration into its capability in ten steps, including scaling self-managed controllers to zero, exporting Applications, migrating repository and cluster credentials, and updating destination.server to cluster names or EKS ARNs (AWS comparison). Reverse those steps and you have the exit plan. The same page lists what the capability does differently from upstream, and each difference is a piece of glue you will redo:

  • RBAC is three fixed roles (ADMIN, EDITOR, VIEWER) mapped to AWS Identity Center groups, configured through the capability, not argocd-rbac-cm.
  • Only AWS Identity Center works for SSO, up to 1,000 identities per capability.
  • Targets are EKS cluster ARNs, not API server URLs.
  • Config Management Plugins, UI extensions and most argocd-cm keys are unsupported, and the sync timeout is fixed at 120 seconds.
  • Notifications read config only from a labelled argocd-notifications-cm and drop the email, teams, webhook, rocketchat, nats and alertmanager services.
  • The ApplicationSet webhook lives on a different path, and webhook secrets live in a different Secret.

That list also works in the other direction. If your current self-hosted Argo CD relies on a Config Management Plugin or a Teams notification, the EKS capability cannot run it today.

With Akuity, Argo CD itself is upstream-compatible, so your Applications move. The work is in what you adopted on top: Kargo promotion pipelines, Akuity AI agents, dashboards and audit records. Those are Akuity features. Export audit records before the contract ends, because Pro keeps them six months.

The Codefresh story is the reason I put "exit" on every evaluation now. A team that standardized on GitOps Cloud keeps its Argo CD Applications after the product ends, but its promotion flows and dashboards have to move somewhere else (Octopus). That is the best case of lock-in with an open source core, and it is still a project nobody planned for.

Run an exit drill during the trial. Stand up a plain upstream Argo CD in a test cluster, point it at the same repositories, apply your Applications and see what breaks. If it takes a day, you are in good shape. If it takes a sprint, write that number in the business case.

What should go on the evaluation checklist?

Use this as the scoring sheet in your trial. Every item is a question you can ask a vendor and verify yourself in a week.

AreaQuestion to askWhy it matters
ArchitectureWhere do the application controller and repo-server run?Decides who holds cluster and Git credentials
ArchitectureWhich cluster types can be targets (EKS only, any Kubernetes, on-prem, air-gapped)?EKS capability targets EKS only
CostWhat is the billing unit, and is an Application on five clusters one or five?EKS counts five
CostWhat does our count look like in 12 and 24 months, with ApplicationSets and preview environments?Generated Applications each count on EKS
CostWhich features are Enterprise only (self-hosting, SSO providers, audit retention)?Akuity self-hosting is Enterprise only
SLAWhich component does the uptime figure cover, how is it measured, what is the credit?EKS SLA names the cluster control plane
SLAWhat is the support response time per severity, in the contract?Not published by Akuity on its pricing page
UpgradesHow far behind upstream does the vendor run, and can we pin or delay a version?Upstream supports three minor versions
SecurityHow fast did the vendor patch the 7 October 2026 critical advisories?Their past patch time predicts the next one
SecurityCan we restrict the vendor's cluster permissions below cluster-admin?AWS example uses system:masters
IdentityWhich SSO providers and RBAC model, and how many identities?EKS: Identity Center only, 1,000 identities
CompatibilityDo our plugins, notification services and argocd-cm settings work?EKS drops CMPs and several notifiers
AuditHow long are sync and audit records kept, and can we stream them out?6 months on Akuity Pro
ExitCan we export everything we configured outside Git, and in what format?Your rebuild list
ExitWhat happens to our instances if the product is discontinued?GitOps Cloud ended
ResidencyWhich regions hold control plane data?Akuity offers US and EU
Break-glassHow do we deploy or roll back during a vendor outage?Change stops when the control plane stops

When should you keep running the Argo CD control plane yourself?

Keep running it yourself when you already have two or more engineers who know Argo CD well, when your clusters are air-gapped or sovereign, or when you depend on features a managed offering does not support. Self-hosting is also the cheapest option at very large Application counts if you already staff a platform team.

The cases where I would not buy managed Argo CD:

  • Air-gapped or no-external-control-plane policies. Akuity self-hosted is Enterprise only, and the EKS capability runs in AWS. In-cluster packaging (OpenShift GitOps, the Azure extension) fits better.
  • Heavy plugin use. If you generate manifests with Config Management Plugins, check compatibility before anything else.
  • Mixed estates. If you run EKS, GKE, AKS and on-prem clusters, an EKS-only control plane gives you a second Argo CD to run anyway.

The cases where I would buy it: a small platform team spending most of its week on Argo CD upgrades and HA, a fleet that is mostly on one cloud, and a security team happy with the vendor's credential model. For a team on EKS only, the capability is hard to beat on price. For a multi-cloud fleet that wants Kargo, Akuity is the most complete hosted Argo CD.

Where does Qovery fit if you already run several Argo CD instances?

Qovery is an Agentic DevOps platform that deploys applications to Kubernetes in your own AWS, GCP, Azure or Scaleway account, or on clusters you already run. It does not host Argo CD, and it can sit next to the Argo CD instances you keep.

The Qovery Argo CD integration connects an Argo CD instance running on a cluster added to Qovery. Qovery imports its applications into a project, creates one environment per destination namespace and one service per Argo CD application, and shows pod status, metrics, live logs, shell access, sync and health status, and a manifest view comparing live and target state. The integration needs an Argo CD API token with read access to applications and logs, and for multi-cluster setups the Qovery stack on every destination cluster.

An earlier post described what was planned. As of today, the docs say Argo CD-managed services are "largely read-only in Qovery - configuration and manifests stay in Git", and that "Sync and Rollback actions are coming soon". Every deployment still goes through Argo CD and Git. Because the documented setup expects Argo CD to run on a cluster registered in Qovery, check with us before assuming it covers a control plane hosted outside your clusters, such as the EKS capability. The original announcement is in Keep ArgoCD. Get Qovery.

For a platform team, this gives one place where developers see Argo CD applications, Qovery-native services, Terraform modules and jobs together, under one RBAC model and one audit log, without migrating a single Application. It does not host or upgrade your Argo CD. If your goal is "stop running the control plane", your options are a vendor from this article, or moving app teams to Qovery's own git-push deployments and preview environments while Argo CD keeps the platform resources.

Frequently asked questions
What should a platform team evaluate before picking a managed Argo CD hosting vendor?

Evaluate five things: where the application controller and repo-server run, how the vendor counts billable Applications, what the SLA covers and how it is measured, which credentials the vendor holds, and what you must rebuild if you leave. Run an exit drill during the trial by applying your Applications to a plain upstream Argo CD. Compare the total against the engineer time you spend on self-hosted upgrades and HA today.

How much does managed Argo CD cost compared with self-hosting?

The Amazon EKS capability costs $0.03 per capability hour plus $0.0015 per Application hour, about $131 a month for 100 Applications. Akuity Pro starts at $495 a month for 50 Applications plus $99 per extra 10. Self-hosted Argo CD is free software but needs at least three nodes for HA, a minor upgrade every three months and fast patching of advisories, which typically costs a meaningful slice of an engineer's time.

Do managed Argo CD vendors publish an uptime SLA?

Few do. The Amazon EKS SLA commits to 99.95% for the EKS cluster control plane on the standard tier and does not name the Argo CD capability. Akuity's pricing page lists no uptime figure. Ask for the covered component, the measurement interval, the credit and the support response times in the contract.

Is the Amazon EKS Argo CD capability real, and what are its limits?

Yes. AWS runs Argo CD as a managed EKS capability in the AWS control plane, billed per capability hour and per Application hour. It only targets Amazon EKS clusters registered by ARN, supports only AWS Identity Center for SSO, and does not support Config Management Plugins, UI extensions or most argocd-cm keys.

What do I keep if I leave a managed Argo CD vendor?

You keep your Application, ApplicationSet and AppProject manifests, Helm charts and Kustomize overlays, because they live in Git and work on any Argo CD. You rebuild SSO and RBAC mappings, cluster registrations, notification configuration, webhook secrets and any vendor-only features such as promotion pipelines or dashboards. Export audit records before the contract ends.

Can Qovery manage my existing Argo CD instances?

Qovery connects Argo CD instances that run on clusters added to Qovery and shows their applications as Qovery environments and services, with logs, pod status, shell access, sync and health status. The integration is read-only today: sync and rollback are announced as coming soon, and deployments stay in Argo CD and Git. Qovery does not host or upgrade Argo CD for you.

Morgan Perry
About the author
Morgan Perry

Morgan co-founded Qovery and leads engineering. He writes about Kubernetes architecture, DevOps best practices, and building resilient infrastructure at scale.

Next step

Already running several Argo CD instances?

Qovery connects the Argo CD instances on your clusters and shows their applications next to your other services, with logs, pod status, shell access and sync status, while deployments stay in Git and Argo CD. Talk to us about your setup.