psychok
v4.2
The PsychoK control plane collapses 11 fragmented dashboards and four layers of stacked YAML into a single opinionated surface — ingest, reconcile, observe, and inject failure on the clusters you actually run in production.
Built in Berlin by three ex-CoreOS platform engineers. Running in 380+ production clusters at companies including Plaid, Mercury, Ramp, Vanta, and Linear.
PsychoK runs four cooperating planes inside your cluster boundary. Each one replaces a category of vendor you would otherwise glue together with shell scripts and Slack approvals.
A drop-in OTel collector fronted by a Prometheus-compatible remote_write endpoint. In 2024 we processed 9.4 trillion container metrics and 2.1 billion distributed traces without a single dropped sample at p99 — your dashboards stay accurate even when 200 nodes page at once.
receivers:
otlp:
protocols: { grpc: {}, http: {} }
prometheus:
config:
scrape_configs: [psychok-managed]
processors:
batch: { timeout: 1s, send_batch_size: 8192 }
exporters:
psychok/ingest: { region: eu-central-1 }GitOps reconciliation against a curated catalog of 2,100+ vetted Helm charts and ArgoCD ApplicationSets, refreshed every 6 hours. Drift is detected, diffed, and either auto-pruned or surfaced for review — whichever your policy dictates. We do not own your manifests; we own the loop.
apiVersion: argoproj.io/v1alpha1
kind: ApplicationSet
metadata: { name: psychok-platform }
spec:
generators:
- list: { elements: [ { cluster: prod-eu-1, release: platform-v2024.11 } ] }
template: { releaseName: '{ {cluster} }-platform' }Pre-aggregated SLO views per workload, plus a query layer that speaks PromQL, LogQL, and SQL against the same underlying columnar store. Median query returns in 380ms across a 14-day window — so the on-call engineer gets an answer before the page times out, not after.
380ms median query, 14d window · 4.2× toolchain spend cut
ChaosK, our in-house fault-injection engine, ran 14,000+ node-failure recoveries in customer clusters during 2024 alone — pod eviction, AZ failover, etcd quorum loss, kubelet restart storms. Exercises are scheduled, scoped to namespaces, and gated on SLO budget so chaos never eats the budget it is meant to protect.
14,000+ node-failure recoveries in customer clusters, 2024
PsychoK is not a rip-and-replace. We sit beside every managed distribution and every GitOps tool your team has already vetted — extending, hardening, and unifying them under one operating layer.
First-class controllers for Amazon EKS, Google GKE, Azure AKS, Red Hat OpenShift, and vanilla kubeadm. Cluster adoption is a single CLI call; lifecycle patches, version skew, and node-pool rightsizing are handled by the control plane.
Native reconciliation via ArgoCD ApplicationSets, Flux v2 Kustomizations, and a curated catalog of 2,100+ vetted Helm charts refreshed every 6 hours. We auto-detect drift and file the PR — your reviewers approve, never author.
OpenTelemetry-native ingest, Prometheus-compatible remote_write, and a columnar store that holds 2.1B traces and 9.4T metrics without a single dropped sample at p99. Ship to your existing Datadog or New Relic if you want — or skip them and save the spend.
OPA Gatekeeper and Kyverno policies run at admission time; External Secrets Operator fans out to Vault, AWS Secrets Manager, and GCP Secret Manager. SPIFFE/SPIRE identities for service-to-service mTLS, audited end-to-end.
Open-source maintainers of chaosk, argocd-precheck, and k8s-cost-explorer — 31,400+ combined GitHub stars, founding contributor to the CNCF-sandboxed OpenFeature K8s operator (March 2024).
// 02 · Capability matrix
Six capability lanes, all GA, all in production at customer scale. Maturity is reported honestly: tier-1 lanes are core; tier-2 lanes extend the platform; tier-3 lanes are evolving.
Self-qualify: if 4 of 6 lanes match your stack, a 30-minute assessment will confirm fit. Book assessment →
// 03 · Security & trust posture
PsychoK is built for teams whose auditors read every line. Compliance is a deliverable, not a marketing badge.
Compliance
SOC 2 Type II
Audited annually. Reports available under NDA through the trust portal.
Penetration testing
3 firms
Consecutive third-party tests. Executive summaries published; full reports on request.
Customer retention
94% over 3 years
Measured across 1,840+ engineering teams on multi-year contracts.
Support SLA
4m 12s
Staffed by ex-Google, ex-Spotify, and ex-HashiCorp engineers. 99.99% uptime SLA.
// 04 · Next step
Bring one production cluster, one pain point, one pager screenshot. We'll show you the control plane on your own topology and leave you with a written findings doc whether or not you buy.
Senior engineer on the call · Median response time 4 minutes 12 seconds · No procurement loop