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.
AiSOC examples
Each scenario in this directory is a small, self-contained alert fixture you can walk through the AiSOC agent funnel two ways:
Real ingest —
aisoc submit examples/alerts/<name>.jsonposts the events to the running stack (started bymake up). The CLI ispackages/aisoc-cli; install it withpip install -e packages/aisoc-clifrom the repo root. It POSTs to/api/v1/alerts/submit, which writes the alert row directly, so the result lands in the Investigation Rail within a second without waiting on the Kafka pipeline.Offline simulator —
aisoc-sandbox demo --scenario <name>runs the same scenario through a zero-dependency in-memory simulator. No Docker, no API key, completes in < 5 s. Install path:pip install -e packages/aisoc-sandboxfrom the repo root. The PyPI package is built on every tag but not published — the upload is blocked on registry credentials — sopip install aisoc-sandboxwill not resolve.
The five scenarios cover the most common attack patterns in modern SOC alert queues; collectively they exercise eleven MITRE ATT&CK techniques across initial access, credential access, lateral movement, privilege escalation, and exfiltration.
| ID | Title | MITRE | Severity | Walkthrough |
|---|---|---|---|---|
lateral-movement |
Impossible-travel Okta sign-in | T1078, T1078.004 | high | walkthrough |
aws-credential-exfil |
IAM keys used from new ASN, then s3:GetObject flood |
T1552, T1567, T1078.004 | critical | walkthrough |
phishing-payload |
Click-through to credential-harvest page | T1566, T1566.002 | high | walkthrough |
kubernetes-privesc |
Namespace SA bound to cluster-admin |
T1098, T1078 | critical | walkthrough |
github-token-theft |
PAT leaked — six private repos cloned in 11 s | T1078, T1555, T1567 | high | walkthrough |
Layout
examples/
├── README.md # this file
├── alerts/
│ ├── lateral-movement.json # 5x production-faithful OCSF/native-shape
│ ├── aws-credential-exfil.json
│ ├── phishing-payload.json
│ ├── kubernetes-privesc.json
│ └── github-token-theft.json
├── lateral-movement.md # walkthroughs — what the agent does, step by step
├── aws-credential-exfil.md
├── phishing-payload.md
├── kubernetes-privesc.md
└── github-token-theft.md
The bundled offline scenarios live in
packages/aisoc-sandbox/src/aisoc_sandbox/scenarios/ — same IDs, simplified shape so the
simulator stays a single small package.
Contributing a new scenario
Open a good first issue
or send a PR that adds:
examples/alerts/<id>.json— production-shape fixture with at least one MITRE technique and a clear narrative in the_description.examples/<id>.md— walkthrough using the existing pages as a template.packages/aisoc-sandbox/src/aisoc_sandbox/scenarios/<id>.json— simplified sandbox variant soaisoc-sandbox demo --scenario <id>works.- Update the table in
examples/README.mdand (if relevant) the tables inREADME.mdandpackages/aisoc-sandbox/README.md.
A maintainer will pair the scenario with at least one detection rule
in detections/ so the production funnel can
actually fire on it.