Your cloud. Your data. Your models. Our brain.
Krenox watches the infrastructure you already run, catches degradation before users feel it, and explains incidents with cited evidence. Experimental today: proposing the fix itself, as a pull request your team approves. Telemetry, code context, and AI inference never leave your cloud.
4f2c1a9No traces from payment-gateway (no OTLP endpoint). Adding one would confirm downstream impact.
Value on what you already run. No migration, no re-instrumentation.
Read-only collectors for Kubernetes and VMs, deployed with a Helm chart and a read-only IAM role. Your existing dashboards and pipelines keep working; Krenox sits beside them, not in front of them.
eBPF observation reads syscalls, sockets, and latency at the kernel, so Krenox sees what's actually happening even when your applications don't log properly.
Deterministic detection watches for degradation (latency drift, error-rate change, saturation) and raises it while it's still a curiosity, not a page at 3 a.m.
The AI acts only through pull requests
Krenox holds no credentials to your infrastructure. Fix proposals are experimental today: where Krenox is confident, it opens a pull request in your repository. Your team reviews it, your CI/CD ships it, or you close it and it learns why. Every action is reviewable, auditable, and reversible.
deploy diff v1.18→v1.19 [git] · request volume flat [eBPF]
Allocation profiling is off. Enabling continuous pprof on search-svc would let me confirm or reject this within a day.
Every finding states its confidence, and its blind spots
Each conclusion carries the evidence it rests on, a stated confidence, and a list of what Krenox could not see. Claims without citations are rejected before a human ever reads them. When it doesn't know, it says so, and tells you exactly what would help it know.
"An AI that says 'I don't know, and here's what would help me know' beats one that's confidently wrong."
Krenox remembers
Every resolved incident, merged fix, and dismissed false alarm becomes institutional memory, in your account. The tenth similar incident is diagnosed with the first nine as evidence. Rejected theories aren't re-proposed. It knows who owned a service at the moment something broke, not just today.
Monitor everything. Log everything. Ask anything. Your bill doesn't move.
A flat, published platform fee plus a per-node unit. No per-GB ingestion, no per-seat charges, no per-question metering. The price is on the website, and it's the price.
See published pricing →How you'll evaluate Krenox
A trust ladder, not a sales funnel. Each step earns the next, and every step is reversible.
Before you install anything: user-journey probes on your public endpoints and a read-only assessment of your cloud account. You receive a real report (your estate map, risks, observability blind spots) having granted nothing but a read-only role.
Krenox deploys into your account with success criteria agreed upfront: we replay your past incidents, shadow your on-call, and measure time-to-diagnosis against your baseline. By week three it knows your baselines; the rest of the trial it proves itself on your real traffic. It closes with an evidence report the product writes itself: what it caught, what it fixed, what it cost. You pay nothing.
Convert to annual (the first five companies to do so get ~50% off their first year), or walk away: clean uninstall, your data downloaded from your dashboards, nothing retained. No lock-in is the point. Your data never left your cloud to begin with.