What Seylo is
Seylo is a console for deploying to and operating Kubernetes clusters — across clouds, across teams, and specifically across the organisational boundaries that normally stop a central tool from working at all.
Most deployment tools need two things that a real fleet cannot give them:
- a network path from the tool to each cluster’s API server, and
- credentials for each cluster, held centrally.
Private clusters have no inbound path. And the platform team that owns a production cluster is frequently not willing to hand its cloud credentials to a vendor — reasonably. Seylo is built so that neither is required.
How the connection actually works
Section titled “How the connection actually works”An agent runs inside your cluster and dials out to Seylo over an outbound WebSocket. Your cluster accepts nothing inbound.
┌──────────────────────────────┐ ┌───────────────────────────┐ │ Seylo │ │ YOUR cluster │ │ │ │ │ │ API + agent gateway │ ◀───────────│ agent Deployment │ │ one port, one TLS cert │ outbound │ ServiceAccount + RBAC │ │ wss://.../agent/connect │ WSS │ loads in-cluster config │ └──────────────────────────────┘ └───────────────────────────┘ no inbound to your cluster no listener, no Service, no cloud credentials here no Ingress, no NodePortWhat follows from that diagram is most of the product:
- No firewall change. No public API endpoint, no VPN or peering, no authorised-networks entry for our IP. If your cluster can reach the internet on 443 — directly or through your existing forward proxy — it can be managed.
- No cloud credential on our side. The agent authenticates to Kubernetes with its own ServiceAccount. There is no AKS, EKS, GKE or OKE credential in our possession, so there is nothing for us to lose. Seylo stores no cloud keys at all: an optional cloud account is a read-only role you grant to Seylo’s identity and revoke by deleting it, and connecting a cluster never needs one.
- No exec, no attach, unless you opt a cluster in. No posture grants either. A cluster admin can let Nightshift, Seylo’s AI SRE, run a fixed list of read-only commands while it troubleshoots — it takes a chart value on your side and a switch in Seylo, both off by default. There is no interactive shell and no attach at any setting. See commands in containers.
- Provider-agnostic. The Kubernetes API is the same everywhere; only credential acquisition ever differed, and the agent removes that step. AKS, EKS, GKE, OKE and any conformant cluster are the same to us.
What you can do once a cluster is connected
Section titled “What you can do once a cluster is connected”The product’s menu has seven groups, each an intent rather than a resource name:
| Operate | An overview across clusters and environments; propose-and-approve on every mutating action; Automation, what Seylo may do on its own and what it did; the activity record, with a hash-chained audit trail |
| Configure | Release an image, edit manifests, environment and ConfigMaps, restart a rollout, schedule a deployment, roll back; browse resources and compare environments for drift |
| Build | The hosted coding agent and its model |
| Observability | Alerts; dashboards, metrics, logs, traces and APM; the service map; uptime |
| Nightshift | Seylo’s AI SRE, which investigates alerts and proposes fixes: incidents, fixes, skills, memory, autonomy and its settings. See Nightshift |
| FinOps | Cost & savings: what each namespace costs, what each cluster could save and where, and managed savings |
| Admin | Clusters, integrations, posture, single sign-on, access - four roles and a namespace allow/deny policy - and billing |
What Seylo deliberately does not do
Section titled “What Seylo deliberately does not do”Reading this list is faster than discovering it later.
- It does not return Secret values. Ever, to anyone, through any surface. See how secrets are handled.
- It does not give you a shell. No
kubectl execsession, noattach. The opt-in commands in containers are a fixed list Nightshift runs, not a shell anyone can type into. - It does not silently overwrite Argo, Flux or Helm. A write to a resource owned by a reconciler is refused until you explicitly acknowledge it, and the refusal names the owner.
- It does not estimate numbers it does not have. An unknown cost or metric is reported as unknown, with the reason. It is never filled in from a similar workload.
- It does not act on its own outside a signed, bounded, expiring and rate-limited envelope that you grant deliberately.
Where to go next
Section titled “Where to go next”- New to Seylo → Core concepts
- Setting up for your team → Create your organisation
- Ready to connect a cluster → Connection methods