Two platforms, different scope
Northflank runs services, jobs, databases and preview environments on Northflank’s cloud or in your own cloud account. It is polished for developers, and strong on running untrusted code, with microVM sandboxes for AI-generated code.
Ankra covers that ground and the platform underneath it. It builds the Kubernetes cluster on the provider you pick or takes the one you already run, ships your code to it, and keeps operating everything on it with vulnerability ranking, cost, backups and an AI that waits for your approval.
The biggest differences are where the cluster can come from, how much of the work after the first deploy is built in, and what the platform holds.
Where the cluster comes from
Northflank creates clusters in your account on EKS, GKE, AKS, Oracle OKE, Civo, CoreWeave, Nebius and AWS European Sovereign Cloud, and on Enterprise it lists fully managed clusters on bare metal. Anything else is imported, and its import requirements are strict. The cluster must use Cilium as its CNI, Northflank strongly recommends importing a new, dedicated cluster, and it says not to import clusters that run production workloads. It does not provision infrastructure for imported clusters, so their node pools are not autoscaled.
Ankra builds kubeadm or k3s clusters on Hetzner, OVHcloud, UpCloud, DigitalOcean, AWS EC2 and Proxmox VE, with a private network, a bastion and the storage driver. It creates EKS, GKE, AKS, DOKS, OVHcloud MKS and UpCloud UKS. And it imports any running, conformant cluster as it is, with one Helm command. The agent only connects outbound, so the cluster can stay behind a firewall or NAT. If you would rather not have a cloud account at all, Ankra Cloud Kubernetes, with control planes run by the Ankra team on Ankra’s own EU cloud, is in closed beta.
If your plan is Hetzner, OVHcloud, UpCloud, Proxmox VE or a cluster you already run, Northflank leaves that cluster to you outside Enterprise. Ankra builds it, or takes it as it is.
After the first deploy
Most of a cluster’s life comes after it is created. On clusters it builds, Ankra runs in-place Kubernetes upgrades with etcd snapshots, node groups and power schedules that stop a cluster outside working hours. On the five public clouds it adds node autoscaling and rolling OS security patching one host at a time. On UpCloud and Proxmox VE, Cluster Mesh connects services across clusters with Cilium ClusterMesh over Ankra’s WireGuard overlay.
Shipping your code
Both products take a repository to a running service. Northflank builds from a Dockerfile or Buildpacks, runs Workflows built in a visual editor with approval nodes, and provisions full-stack preview environments for pull requests.
Ankra’s Applications read your repository and open a setup pull request with the Dockerfile, manifests and pipeline, or keep the CI you already have and deploy the image it publishes. Ankra Pipelines run in your own cluster, with tests, rootless builds and Semgrep, Checkov and Trivy scans, then a gate that decides which image may be published.
- Build once. A merge that carries exactly what a pull request run already built, scanned and tested publishes that tested image instead of rebuilding it.
- A merge train. On GitHub, the merge train tests each pull request as the exact commit its merge will produce, so two green pull requests cannot break the main branch together.
- Signed provenance. With verified provenance in enforce mode, an application refuses any image that your organisation’s pipeline gate did not publish, such as one built on a laptop or pushed by hand.
- Previews. Pull-request previews exist for GitHub applications, but Ankra’s own documentation says not to rely on them yet while a known issue is fixed. Full-stack previews are a Northflank strength today.
Pipelines trigger on GitHub, GitLab and Bitbucket Cloud. Applications start from GitHub repositories, and some documented pipeline stage kinds, including in-pipeline approval, do not run yet.
Platform add-ons and fleets
Northflank’s templates are edited as JSON with two-way Git sync, and cover its own services and add-ons.
Ankra’s stacks install any Helm chart or manifest in dependency order, so the database comes up before the service that needs it, and deploy waves order whole stacks. The catalogue covers ingress, cert-manager and monitoring, and security building blocks such as External Secrets, Sealed Secrets, Vault and Kyverno. Stack profiles version a setup and roll each new version across a fleet of clusters, and two-way GitOps keeps every change as a commit in your repository.
Data and backups
Northflank’s managed databases are a strength, with PostgreSQL, MySQL, MongoDB, Redis, MinIO and RabbitMQ and automatic backups.
Ankra protects stacks on a schedule into a bucket you own. The cluster writes its volumes and database dumps straight to the bucket, and the platform never holds the data. Restore points restore volumes and PostgreSQL or MySQL dumps in place. Operator databases such as CloudNativePG are captured but cannot yet be restored in place. A stack can be cloned to another cluster with its data, and ankra migrate moves a Docker Compose deployment onto Kubernetes with its PostgreSQL, MySQL or MariaDB data. Ankra’s own managed services, PostgreSQL, Valkey, NATS, OpenSearch and S3-compatible storage among them, run on your cluster and are in closed beta.
Security of what runs
On its own cloud, Northflank isolates workloads in Kata Containers microVMs by default. Since September 2026 you can also turn on microVM or gVisor isolation on clusters in your own account, choosing it per service, job, add-on or build. Its container scanning finds vulnerabilities and secrets in images you build on Northflank, once you enable it on a build or service, and for new builds only. Enterprise adds automatic scanning with policy enforcement.
Ankra’s Security Center reads the vulnerabilities in what is actually running across your clusters, including images Ankra never built, once the security baseline is on. It ranks them by CISA’s Known Exploited Vulnerabilities catalogue and FIRST’s EPSS scores, so the ones attackers use come first, and it scores how far each cluster’s network policies over-allow traffic. Compliance Management scores the fleet against the Pod Security Standards, and against CIS and NSA/CISA where those inputs are enabled, tracks exceptions with owners and review dates, and produces an evidence report. Ankra’s security and compliance material is on its trust page.
Secrets, and the October 2026 incident
On 30 October 2026 Northflank disclosed a security incident after a threat alert from Wiz. Its status page told customers to “Assume all environment variables have been decrypted. Disable/rotate all keys.” It later wrote that “Users have reported unauthorised use of AI API keys supplied to the Northflank platform”, and on 31 October it enabled audit logs for all customers, including free users. On 2 November it moved the incident to monitoring, said evidence suggested the attack was focused on secrets held in its control plane, that customers on forward-deployed or dedicated control planes were unaffected, and that it had patched CVE-2026-98432 across its fleet.
On 6 November Northflank added that the first known decryption of secrets took place between 02:39 and 03:08 UTC on 30 October, about nine hours before the first alert at 11:43 UTC. It said it patched the exploit within 60 minutes of being notified, and that it had found no evidence that customer workloads or workload data were accessed. As of 9 November the incident is still in monitoring, an independent forensic review is under way, and Northflank says it will share a root cause analysis once its findings are confirmed. Wiz has tracked incidents at AI developer platforms since October 2026, and by 9 November its list named more than a dozen platforms, Northflank among them. Wiz says no single vulnerability or dependency links them.
If you used Northflank, rotate first. Revoke each key with the provider that issued it, not only in Northflank, and check your AI providers’ usage and billing for activity you did not start. Northflank also advises reviewing provider logs and setting IP restrictions and budget limits.
Northflank does offer a further control for stored secrets: since September 2026 its customer-managed envelope encryption lets you wrap the keys with your own cloud KMS key.
Then ask every platform, Ankra included, what it holds. Ankra answers on one page, What Ankra can see and do.
- Ankra’s inventory holds digests, not values. The agent replaces every Secret value with a SHA-256 digest before cluster data leaves the cluster, and the AI never sees the values. Reading one takes its own permission, and each reveal is written to the audit log, which is kept for 365 days.
- You choose what the platform holds. Your workloads run in your own cluster, so their secrets can be created there or synced from your own secret manager with External Secrets or Vault. Ankra’s inventory then holds only their digests.
- Ankra’s AI starts without your AI key. By default it runs on Ankra’s own model access within your plan’s monthly allowance, and falls back to the DevOps AI model by Ankra when the allowance runs out. You can also connect your own provider key or your own endpoint.
- Pipeline secrets are files, not environment variables. A pipeline secret arrives as a file at run time, reaches only the stages that list it, is never written into the pipeline itself, and pull requests from forks get none.
- What you give Ankra is sealed. Cloud, Git and registry credentials, and your organisation’s SOPS key for Git, are stored encrypted in an OpenBao or HashiCorp Vault secret store in the EU and decrypted only when an action needs them. Values you store with
ankra secretsgo in and never come out.
Cost
Ankra’s Cloud Cost estimates spend per cluster, namespace and stack from list prices, points out waste and savings, tracks budgets, and reprices a whole cluster on other providers before you move it. Northflank bills usage to the second, with no seat fees.
AI that runs code, or AI that runs the platform
Northflank’s AI work gives agents a safe place to run. Sandboxes execute AI-generated code in microVMs, Cloud Harnesses run coding agents such as Claude Code or Codex, and a hosted MCP server lets your coding agent deploy and inspect resources.
Ankra’s AI operates the platform. The assistant answers from your clusters, AI Insights analyses every firing alert, the AI board runs work through role agents with a person approving, and the AI Gateway reviews pull requests and fixes failing ones. Ankra’s own AI asks before every write by default. The MCP server gives your own agents more than 500 tools across the same operations. Its writes are audited, and your MCP client decides how it confirms them.
Price
For clusters in your own cloud, Northflank’s pay-as-you-go plan charges $0.02083 per node vCPU an hour. On Northflank’s own cloud it charges $0.01667 per vCPU and $0.00833 per GB of memory an hour, compute included.
Ankra’s Free tier includes 30 worker vCPU. Beyond that, the Startup plan charges €0.01 per worker vCPU an hour, about €7.20 a month. Controller nodes are never billed, and your provider bills the servers.
When to choose Ankra, and when to choose Northflank
Choose Ankra when you want Kubernetes on Hetzner, OVHcloud, UpCloud, DigitalOcean or your own servers built and operated for you, need to bring production clusters you already run, want the vulnerabilities attackers exploit found across every cluster, or want security, cost, backups and AI operations in the same product as delivery.
Choose Northflank when
- you want to run on Northflank’s own cloud, with no cluster of your own
- managed databases are the centre of your stack
- you want approval steps inside visual release workflows, and full-stack previews, today
- you run untrusted or AI-generated code that needs microVM isolation
- you need a forward-deployed or air-gapped control plane under an Enterprise contract
Compare the work that remains
- Create the foundation. On your target provider, record who builds the cluster and who scales its nodes.
- Bring what you run. Try to import a cluster that already runs your workloads and note what each platform asks you to change first.
- Ship a release. Include tests, scans, a preview and a rollback.
- Add a platform component. Install an operator or monitoring stack and note how it is versioned across clusters.
- Investigate a vulnerability. Find every affected workload, across every cluster.
- Rotate a secret. Note every place the platform kept a copy of it.
- Compare total cost. Include the infrastructure, the platform fee per vCPU and the work each route leaves to you. See Northflank’s pricing and Ankra’s pricing.
Reviewed 9 November 2026 against Northflank’s documentation, pricing, changelog, security page and status page. Product scope and plans can change.