Kllyroo learns what normal looks like for every server and service, warns you before disks fill or latency drifts, and writes the first draft of the investigation for you. The model runs on your Kllyroo server, so nothing leaves your network.
Kllyroo AI operations combine a local machine-learning engine for baselines, anomaly detection and predictive alerts with a built-in offline language model (IBM Granite or Qwen, Apache-2.0, delivered as a signed model pack) that writes incident summaries, likely causes and postmortem drafts. Cloud AI providers are optional.
kllyroo.your-company.local/ai/predictive-logs
AI Predictive Logs: predictions, anomalies and findings in one inbox
Capabilities
What you get
Anomaly detection
Learned baselines per metric raise anomaly alerts and recommendations, not just static thresholds.
Predictive alerts
“Disk full in 3 hours” style warnings so you act before the outage.
AI predictive logs
An inbox of AI findings and predictions from your log and metric streams.
Investigations & postmortems
Evidence gathering, likely cause with confidence, AI summary and postmortem drafts.
AI assistant & MCP
Chat over live monitoring data. Read-only by design — it cannot restart services or change settings. MCP server for external AI tools.
Built-in offline AI
Local model runs on CPU via a signed model pack. Or use a model server on your network, or optional Claude / OpenAI / Gemini.
Your choice of model
Private by default
AI option
Where data goes
Internet needed
Built-in offline AI (Granite 4.1 3B / 8B, Qwen3 4B)
