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
# Service-level objectives (Phase 12 — observability + governance).
#
# Every service under services/ MUST have an entry here — either an `slo:` block
# (it serves traffic and has objectives) or an `exempt:` block with a reason (a
# build-time library or a demo-only tool). scripts/check_slos.py --check fails
# CI if a service is missing, so a new service cannot ship without declaring its
# reliability posture.
#
# These are OBJECTIVES (targets), not measured SLIs. The golden-signal wiring
# that measures against them is OpenTelemetry (traces → Tempo/Jaeger) +
# Prometheus (/metrics); see docs/operations/observability.md. Error budget =
# 1 - availability_target over a rolling 30 days.
version: 1
services:
api:
availability_target: 0.999
latency_p95_ms: 300
golden_signals: [latency, traffic, errors, saturation]
description: Public REST/GraphQL API — the analyst + integration entry point.
realtime:
availability_target: 0.999
latency_p95_ms: 250
golden_signals: [latency, traffic, errors, saturation]
description: WebSocket fan-out for live alerts + graph updates.
ingest:
availability_target: 0.999
latency_p95_ms: 200
golden_signals: [latency, traffic, errors, saturation]
description: Event intake + OCSF normalization on the hot path.
fusion:
availability_target: 0.999
latency_p95_ms: 500
golden_signals: [latency, traffic, errors, saturation]
description: Correlation + promotion + alert persistence.
agents:
availability_target: 0.99
latency_p95_ms: 120000
golden_signals: [latency, traffic, errors, saturation]
description: Investigation orchestrator (LLM latency dominates; SLO is per-investigation).
connectors:
availability_target: 0.99
latency_p95_ms: 5000
golden_signals: [traffic, errors, saturation]
description: Polling scheduler + connector runtime.
enrichment:
availability_target: 0.99
latency_p95_ms: 2000
golden_signals: [latency, traffic, errors]
description: IOC + entity enrichment cache.
threatintel:
availability_target: 0.99
latency_p95_ms: 1000
golden_signals: [latency, traffic, errors]
description: Threat-intel store + actor attribution.
ueba:
availability_target: 0.99
latency_p95_ms: 1000
golden_signals: [latency, traffic, errors]
description: Behavioral baselines + anomaly scoring.
actions:
availability_target: 0.999
latency_p95_ms: 3000
golden_signals: [latency, traffic, errors]
description: SOAR executors (autonomy-gated; latency depends on the vendor API).
slack-bot:
availability_target: 0.99
latency_p95_ms: 3000
golden_signals: [latency, traffic, errors]
description: Slack ChatOps + approvals.
teams-bot:
availability_target: 0.99
latency_p95_ms: 3000
golden_signals: [latency, traffic, errors]
description: Microsoft Teams ChatOps + approvals.
honeytokens:
availability_target: 0.99
latency_p95_ms: 500
golden_signals: [latency, traffic, errors]
description: Honeytoken minting + first-touch alerting.
purple-team:
availability_target: 0.95
latency_p95_ms: 2000
golden_signals: [traffic, errors]
description: Atomic/Caldera adversary emulation + coverage.
mesh:
availability_target: 0.99
latency_p95_ms: 500
golden_signals: [latency, traffic, errors]
description: Federated threat-intel mesh hub (hashed IOC sightings + verdict signatures, k-anonymity enforced).
mcp:
availability_target: 0.99
latency_p95_ms: 2000
golden_signals: [latency, traffic, errors]
description: MCP server exposing agent tools to IDEs.
osquery-tls:
availability_target: 0.999
latency_p95_ms: 500
golden_signals: [latency, traffic, errors, saturation]
description: First-party osquery TLS plugin (endpoint telemetry ingest).
exempt:
demo-producer:
reason: Demo-only synthetic event generator; not a production service.
escalations:
reason: >-
Opt-in ack-clock notification bridge (profile `escalations`): a
single polling loop with no service endpoint to measure, designed to
retire when the native outbound-webhook emitters are wired. Its own
failure mode (a missed tick) is logged and retried next tick.
osquery-extensions:
reason: Build-time Go virtual-table library shipped with the osquery agent; no service endpoint.