Features built for real SOC workflows

Autopilot is designed to reduce alert fatigue without creating “auto-remediation anxiety”. Plans are proposed fast, but execution stays human-gated.

Autonomous triage

Agents analyze incoming alerts, extract entities (IPs, users, hosts), and assign severity.

  • Entity extraction + normalization
  • Noise reduction via context
  • Consistent, explainable summaries

Incident correlation

Automatically link related alerts into unified cases with attack timelines.

  • Case creation + merging
  • Timeline reconstruction
  • MITRE mapping support

Response planning

Generate risk-assessed response plans with recommended Wazuh Active Response actions.

  • Plan proposals with risk notes
  • Policy guard checks
  • Execution is optional + gated

Two-tier human approval

A two-tier approval workflow ensures humans authorize every response action.

  • Approve / reject / expire
  • Audit trail for decisions
  • Safer automation at scale

Evidence packs

Structured JSON evidence packages for compliance, audits, and forensics.

  • Entities + timeline
  • Plans + actions
  • Exportable schema

Prometheus + Slack

Full observability with SOC KPIs + real-time approvals via Slack Socket Mode.

  • /metrics for KPIs and latency
  • Interactive approval buttons
  • No inbound ports required

Model-agnostic by design

OpenClaw supports multiple LLM providers. You can run paid models (for deeper reasoning) or local models (for air-gapped). Configure your provider in openclaw/openclaw.json.

Deploy (Quickstart) Air-gapped guide