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
# Prometheus scrape config for the AiSOC docker-compose stack.
#
# Phase 2.3 — reconciled against reality. Each `job_name` here MUST
# correspond to a service that actually exposes a `/metrics`
# endpoint *and* is reachable on the documented container hostname
# + INTERNAL container port (the left side of `8088:8003` is the
# host port; Prometheus runs inside the `aisoc` Docker network so
# it sees the right-hand-side port).
#
# Pre-Phase-2.3 the file shipped five scrape jobs, four of which
# were broken:
#
# - aisoc-core-api - hostname "core-api" did not exist (the
# compose service is named "api").
# - aisoc-enrichment- enrichment is a Go connector that has NEVER
# had a /metrics handler. Prometheus would
# have logged "connection refused" forever.
# - aisoc-agents - hostname correct, but agents/app/main.py
# has no /metrics route. The Python tests
# for the agents service don't import
# prometheus_client.
# - aisoc-actions - same story; no /metrics route in
# services/actions/app/main.py.
# - aisoc-ingest - this one was always right.
#
# The result was "we scrape five jobs; one returns metrics". Every
# dashboard built against `up{job=...}` showed four orange tiles.
#
# Below: only services with a verified /metrics handler at the
# advertised port are scraped. The audit lives in
# scripts/audit_prometheus_targets.py and is wired into CI so this
# file can't drift again silently.
#
# Service Source code reference Port
# ------------ -------------------------------------------- -----
# api services/api/app/main.py @app.get("/metrics") 8000
# ingest services/ingest/internal/server/server.go 9090
# r.Get("/metrics", promhttp.Handler()...)
# fusion services/fusion/app/api/router.py 8003
# @router.get("/metrics")
# threatintel services/threatintel/app/main.py 8005
# app.mount("/metrics", metrics_app)
global:
scrape_interval: 15s
evaluation_interval: 15s
# Phase 2.4 — alert rule files. Each file is loaded at Prometheus
# start-up; reloading rules at runtime requires SIGHUP'ing the
# container. The shipped ruleset lives under `infra/docker/alerts/`
# so adding a rule = adding a file + including it here.
rule_files:
- /etc/prometheus/alerts/*.yml
# Phase 2.4 — Alertmanager wiring. Prometheus forwards firing /
# resolved alerts to the in-cluster Alertmanager defined in
# `docker-compose.yml`. Both run inside the `aisoc` network so the
# DNS name `alertmanager:9093` resolves cluster-internal.
alerting:
alertmanagers:
- static_configs:
- targets:
- alertmanager:9093
scrape_configs:
# FastAPI core API — uvicorn worker on container port 8000.
- job_name: 'aisoc-api'
static_configs:
- targets: ['api:8000']
metrics_path: /metrics
# Go ingest service — dedicated metrics listener on METRICS_PORT
# 9090 (HTTP traffic is served on 8080 internally; metrics live
# on a separate listener so a stalled request handler doesn't
# block scrapes).
- job_name: 'aisoc-ingest'
static_configs:
- targets: ['ingest-worker:9090']
metrics_path: /metrics
# Fusion service — FastAPI on container port 8003. The /metrics
# route is registered through `app.api.router`, not at top level.
- job_name: 'aisoc-fusion'
static_configs:
- targets: ['fusion:8003']
metrics_path: /metrics
# Threat-intel service — FastAPI on container port 8005. Uses
# `app.mount("/metrics", metrics_app)` so the prometheus-client
# ASGI handler runs in its own sub-app (cheaper than a route).
- job_name: 'aisoc-threatintel'
static_configs:
- targets: ['threatintel:8005']
metrics_path: /metrics
# LiteLLM gateway (issue #478) — third-party image that exports
# per-task LLM latency, tokens, cost, errors, retries and fallbacks
# on /metrics via its prometheus callback (see
# infra/litellm/config.yaml). Not a service in our `services/` tree,
# so it's allowlisted in scripts/audit_prometheus_targets.py under
# THIRD_PARTY_HOSTS. Container listens on 4000 inside the network.
- job_name: 'aisoc-litellm'
static_configs:
- targets: ['litellm:4000']
metrics_path: /metrics
# Phase 2.4 — self-scrape Alertmanager so we have visibility
# into the notification pipeline (delivery failures show up as
# `alertmanager_notifications_failed_total > 0`).
- job_name: 'aisoc-alertmanager'
static_configs:
- targets: ['alertmanager:9093']
metrics_path: /metrics
# NOTE: agents, actions, enrichment, ueba, honeytokens,
# purple-team, connectors, osquery-tls, realtime, slack-bot, and
# teams-bot do NOT currently expose /metrics. They MUST be wired
# up before they appear in this file (and the audit script will
# remind any future contributor who adds the route to also add
# the scrape target here).
# The auto-triage worker. Its counters were already being kept and were
# readable only from a shutdown log line — dead_lettered rising means
# alerts are dropped rather than triaged, and deterministic climbing while
# llm stays flat means the LLM path is failing open to the fallback, which
# looks fine from outside because alerts still get verdicts.
- job_name: aisoc-agents
static_configs:
- targets: ["agents:8084"]