Open-source AI Security Operations Center: alert fusion, LLM-agent triage, MITRE ATT&CK investigation, and a replayable decision ledger for every agent step. Self-hostable, runs with no API keys, MIT licensed. Ships an MCP server for Claude, Cursor and Continue.
Python agentic-aiai-securityalert-triageblue-teamcybersecuritydetection-engineeringincident-responselanggraphllm-agentsmcpmitre-attackmodel-context-protocolpurple-teamsecurity-operationsself-hostedsiemsoarsocthreat-detectionthreat-intelligence
# Phase 2.4 — Alertmanager configuration for the AiSOC dev stack.
#
# The dev compose stack ships with a stub receiver that writes
# notifications to a local webhook endpoint (a tiny FastAPI handler
# in `services/api` would catch these in a future iteration; for
# now they bounce to stdout via Alertmanager's own logs). Operators
# replacing this in production swap the receiver block for their
# real channel (PagerDuty, Slack, Opsgenie, email, Teams) — every
# environment variable below has a sensible default that keeps the
# dev stack from crashing if the operator hasn't provisioned the
# downstream system yet.
global:
# Wait this long before resending a still-firing alert. Aligns
# with the rules' `for:` windows so we don't double-page.
resolve_timeout: 5m
# Routing: severity-based fan-out, team-aware grouping. This is
# the production-shaped tree, NOT a single catch-all receiver —
# even the dev stack benefits from realistic routing so debugging
# the rules surfaces the right notification.
route:
receiver: default
# Group by alertname + team so a "service down" page covers all
# affected services in one notification, but different teams'
# alerts don't get merged.
group_by: ['alertname', 'team']
group_wait: 30s
group_interval: 5m
repeat_interval: 4h
routes:
- matchers:
- severity = "critical"
receiver: critical-page
group_wait: 10s
repeat_interval: 1h
- matchers:
- severity = "warning"
receiver: warning-ticket
group_wait: 1m
repeat_interval: 12h
receivers:
# Production note: Alertmanager does NOT expand shell-style
# `${VAR:-default}` substitutions — `amtool check-config` will
# reject them. Operators replacing these receivers for prod
# should swap the entire file via a docker volume mount or by
# rendering a templated config at deploy time (e.g. with
# `envsubst` in the entrypoint or a Helm template). The dev
# URLs below point at the API's internal alert sink, which
# logs every payload for inspection.
- name: default
# In dev, we just want Alertmanager to log the alert. The
# receiver below is a no-op webhook; Alertmanager will still
# surface the alert in its UI (http://localhost:9094/#/alerts).
webhook_configs:
- url: http://api:8000/internal/alerts/sink
send_resolved: true
- name: critical-page
# Production operators replace this with PagerDuty / Opsgenie.
# In dev we route to the same sink + tag the payload so the
# API can distinguish severity tiers in its incoming-alert
# webhook handler.
webhook_configs:
- url: http://api:8000/internal/alerts/sink?severity=critical
send_resolved: true
- name: warning-ticket
# Production operators wire this to Jira / ServiceNow.
webhook_configs:
- url: http://api:8000/internal/alerts/sink?severity=warning
send_resolved: true
# Inhibit a warning if a critical from the same group is already
# firing — keeps the on-call from seeing two pages for the same
# root cause (e.g. service down ALSO triggers high error rate).
inhibit_rules:
- source_matchers:
- severity = "critical"
target_matchers:
- severity = "warning"
equal: ['team', 'alertname']