Everything your fleet needs, one lightweight agent.
VantraOps replaces the four-tool stack a small team usually assembles by hand — a metrics system, an alerting layer, a cost tool, and a diagnostics assistant — with a single read-only agent and one dashboard your whole team already knows how to read. Here's what's actually included.
The six pillars
Metrics monitoring
Live CPU, memory, and network usage streamed per node, namespace, and workload. Data lands in seconds, not on a five-minute Prometheus scrape interval — which matters most exactly when you're chasing a spike that's already passed by the time a stale dashboard refreshes.
Health checks
Issue detection with debounced open and resolve states: a pod that restarts twice and settles isn't the same as one stuck in CrashLoopBackOff for twenty minutes, and the system treats them differently instead of paging you for both.
Alerts & events
Rule-based alerts on the conditions you define, plus a live Kubernetes events feed that's deduplicated and grouped by workload — not the raw, timestamp-sorted firehose kubectl gives you by default.
Cost visibility
Spend broken down by namespace and workload, mapped to actual node utilization rather than an evenly-divided guess at the monthly bill. See which service is actually driving cost, not just which one exists.
Architecture health
An ongoing Well-Architected-style review — single-AZ exposure, missing PodDisruptionBudgets, containers running as root — that reruns continuously instead of going stale a week after a one-time audit.
Multi-cluster fleet view
Every cluster you connect, side by side, with a single health signal per tile. Problems in a cluster nobody's actively watching still surface before a customer notices.
Bring your own provider. Every time.
VantraOps detects the issue — an OOMKilled loop, a scheduling failure, a cost spike. Your own AI provider explains it in plain language and proposes a fix. Nothing routes through a shared model or a VantraOps-held key: VantraOps sends only the anonymized signal — events, metric deltas, resource specs — never full logs or secrets.
Each tenant's provider key is encrypted with its own data key, isolated the same way cluster credentials are. See the full model on the security page.
- → Anthropic, AWS Bedrock, Azure OpenAI, or Google Vertex — switch any time
- → Choose your model per provider — Claude Opus vs. Sonnet, GPT-4o vs. GPT-4o-mini
- → Turn it off entirely, tenant by tenant, and fall back to rule-based health checks
A walkthrough of the exact memory-curve diagnosis AI analysis automates — including the case behind the example above.
Connect a cluster in three steps.
Name your cluster
From the portal, register a cluster and choose its cloud provider. This generates a single-use enrollment token with a 15-minute TTL, scoped to that cluster only.
Install the agent with Helm
Run one command against the cluster you want to connect. It deploys the agent with a read-only RBAC role and nothing else — no inbound access to your cluster is ever required.
helm repo add vantraops https://charts.vantraops.com
helm install vantraops-agent vantraops/agent \
--set enrollmentToken=<generated-in-portal> \
--set apiBaseUrl=https://api.vantraops.comWait for the first heartbeat
The agent exchanges the token for a long-lived credential on boot, stores it as a Kubernetes Secret in its own namespace, and heartbeats roughly every 30 seconds from then on. No manual verification step.
The agent's ServiceAccount is read-only by default (get / list / watch). Enrollment tokens are single-use and expire in 15 minutes whether or not they're consumed.
See it running on your own cluster.
Free tier includes one cluster, real-time metrics, and health alerts — no card required.