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
"""Pre-fetched ContextBundle for sub-agents (T2.1).
Sub-agents historically called graph / memory / TI tools as their *primary*
discovery step, which serialised four to six round-trips on the critical path
of every investigation. The :class:`ContextBundle` is built once, before the
fan-out to sub-agents, and carries:
* the entity neighbourhood walked from the alert (configurable depth, default 2),
* historical similar-case verdicts pulled from ``aisoc_institutional_memory``,
* peer-entity UEBA baselines for each principal in the alert,
* threat-intel matches for each IOC.
Tools the sub-agents still call become *enrichments* of the bundle, not
primary discovery.
See ``services/agents/tests/test_context_bundle.py`` for the latency gate
(p95 < 5s across the 200-incident eval set).
"""
from .bundle import (
ContextBundle,
ContextBundleBuilder,
EntityNeighborhood,
EntityRef,
HistoricalCase,
ThreatIntelMatch,
UEBABaseline,
build_context_bundle,
extract_entities,
)
__all__ = [
"ContextBundle",
"ContextBundleBuilder",
"EntityNeighborhood",
"EntityRef",
"HistoricalCase",
"ThreatIntelMatch",
"UEBABaseline",
"build_context_bundle",
"extract_entities",
]