Operate your OpenStack cloud with AI diagnosis.
Cortex collects metrics and logs from every node, learns what normal looks like on your own infrastructure, groups related alerts into one incident through a live topology graph, and explains it in plain language. It proposes the fix — you approve it.
Anomaly score
4.2σ
Example · compute-02 iowait vs. its weekday-hour baseline
Correlated incident · example
18 alerts collapsed into 1 incident — probable cause on compute-02.
Root-cause likelihood
Scrape interval
20 s
Autonomy
Level 0 · you approve
- Metric scrape interval
- 20 s
- Weekday-hour baselines
- 7 × 24
- Max forecast horizon
- 90 days
- Every fix needs approval
- Level 0
Fleet
Every node,
one health picture
Controller, compute, storage, network and monitoring nodes report through node_exporter. Each card shows what Cortex surfaces for a node — health, anomalies against its baseline, capacity forecasts. Illustrative fleet; yours appears once nodes are registered.
controller-01
Keystone · Neutron L3 · Glance · MariaDB
cpu 34%
compute-01
nova-compute · RabbitMQ · libvirt
cpu 51%
compute-02
nova-compute · ovs-br-ex
iowait 4.2σ
storage-01
cinder-volume · node-exporter
88% in 4d
Built for OpenStack operators
From raw telemetry to
an approved fix.
Metrics, logs, topology and your own runbooks feed a team of specialised agents. Detection, diagnosis and remediation are one workflow, with a person on the last step.
Detection
Baselines that know your Tuesday 10:00
Robust median/MAD baselines per node, per metric, per weekday and hour. Deviations are scored 2 / 3 / 4σ as medium, high and critical, so normal rhythms stay quiet and real drift stands out.
Correlation
One incident, not twenty alerts
Alerts are walked across the live topology graph — hosts, services, networks, instances — and grouped into a single incident with a probable root cause and a plain-language narrative.
Forecasting & capacity
Saturation seen in advance
Pooled quantile forecasts for CPU, memory and disk give a time-to-threshold ETA. A what-if simulator, quota and budget alerts, and VM right-sizing turn it into capacity decisions.
Security
Security signals in context
SSH auth-log anomalies, security-group drift, exposed ports, CVE matching and Keystone token audit — plus kernel-level eBPF signals from Falco or Tetragon when a sensor is deployed.
Copilot
Ask the cloud, in plain language
A LangGraph orchestrator routes your question to the monitoring, anomaly, prediction, network, security and knowledge agents, with answers grounded in your runbooks and the OpenStack docs indexed in Qdrant.
Remediation
Fixes you approve, never surprises
The remediation agent proposes a concrete fix and simulates its impact on the topology graph. You approve, reject or ask for more information, then run it with an explicit click via Ansible.
Built on
Guardrails
Built to be trusted with production
An AI that can propose changes to your cloud needs hard limits. These are enforced in the product, not left to convention.
Autonomy
Nothing runs without your click
Cortex sits at Level 0 of the confidence ladder: it diagnoses and proposes, a person approves, and only an explicit click executes the fix — even for low-risk changes.
Simulation
Impact shown before you decide
Every proposal is simulated read-only against the topology graph, so you see which services, containers and instances a command would touch before you approve it.
Audit
Every decision on the record
Diagnosis, proposal, approval or rejection, and execution are written to a timestamped audit trail you can browse from the console.
Access
Role-based by default
Admin, operator and viewer roles. Operators handle low and medium-risk fixes, higher-risk ones need an admin, and viewers are read-only.
Alerting
Routed by severity
Medium alerts stay on the dashboard. High and critical alerts are emailed and escalate if nobody acknowledges them. A weekly digest summarises capacity, security and what the agents caught.
Resilience
Degrades, never goes dark
If Prometheus, Loki or the language model is unreachable, each agent and dashboard widget falls back on its own with a clear degraded state instead of blanking the page.
Get started
Stop correlating alerts by hand.
Let Cortex explain the incident.
Register your nodes, point Cortex at Prometheus and Loki, and get baselines, correlated incidents, forecasts and reviewed fix proposals in one console.
Sign in to CortexSelf-hosted · Built for OpenStack 2024.1