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
-- AiSOC Detection-as-Code lifecycle migration (Wave 2 — w2-dac)
--
-- Adds the `detection_rule_proposals` table that backs the
-- propose → review → eval-gated → promote flow. Every proposal carries
-- the candidate rule body, MITRE coverage, the eval baseline that will
-- be compared against, and the decision audit trail. When a proposal
-- is promoted it materialises a row in `detection_rules` and links back
-- via `promoted_rule_id` so the catalog UI can render lineage.
--
-- Multi-tenant RLS is applied: each tenant only sees its own proposals.
-- Platform-wide proposals (tenant_id IS NULL) are visible to operators
-- with the platform role and are how built-in detections get reviewed.
-- ============================================================
-- detection_rule_proposals
-- ============================================================
CREATE TABLE IF NOT EXISTS detection_rule_proposals (
id UUID PRIMARY KEY DEFAULT uuid_generate_v4(),
tenant_id UUID REFERENCES tenants(id) ON DELETE CASCADE,
-- Optional pointer to an existing rule when the proposal is an edit.
base_rule_id UUID REFERENCES detection_rules(id) ON DELETE SET NULL,
-- Materialised rule once the proposal is promoted.
promoted_rule_id UUID REFERENCES detection_rules(id) ON DELETE SET NULL,
-- Rule definition (uses the same naming convention as the ORM model).
name VARCHAR(255) NOT NULL,
description TEXT,
rule_language VARCHAR(30) NOT NULL,
rule_body TEXT NOT NULL,
category VARCHAR(100) NOT NULL,
severity VARCHAR(20) NOT NULL DEFAULT 'medium',
confidence INTEGER NOT NULL DEFAULT 50,
mitre_tactics JSONB NOT NULL DEFAULT '[]'::jsonb,
mitre_techniques JSONB NOT NULL DEFAULT '[]'::jsonb,
tags JSONB NOT NULL DEFAULT '[]'::jsonb,
-- Lifecycle: proposed → in_review → eval_passed | eval_failed → promoted | rejected.
status VARCHAR(20) NOT NULL DEFAULT 'proposed',
-- The most recent eval result so the UI can render gate verdicts
-- without re-running the suite. Shape:
-- {"baseline": {...}, "candidate": {...}, "delta_pp": -1.2, "passed": false, "ran_at": "..."}
eval_result JSONB NOT NULL DEFAULT '{}'::jsonb,
-- Free-form review thread. Shape: [{"actor_id": "...", "comment": "...", "at": "..."}].
review_comments JSONB NOT NULL DEFAULT '[]'::jsonb,
-- Authoring + decision audit trail.
proposed_by_id UUID REFERENCES users(id) ON DELETE SET NULL,
decided_by_id UUID REFERENCES users(id) ON DELETE SET NULL,
decision_comment TEXT,
decided_at TIMESTAMPTZ,
created_at TIMESTAMPTZ NOT NULL DEFAULT NOW(),
updated_at TIMESTAMPTZ NOT NULL DEFAULT NOW()
);
CREATE INDEX IF NOT EXISTS idx_detection_proposals_tenant
ON detection_rule_proposals(tenant_id);
CREATE INDEX IF NOT EXISTS idx_detection_proposals_status
ON detection_rule_proposals(status);
CREATE INDEX IF NOT EXISTS idx_detection_proposals_base
ON detection_rule_proposals(base_rule_id)
WHERE base_rule_id IS NOT NULL;
ALTER TABLE detection_rule_proposals ENABLE ROW LEVEL SECURITY;
ALTER TABLE detection_rule_proposals FORCE ROW LEVEL SECURITY;
DROP POLICY IF EXISTS detection_proposals_tenant ON detection_rule_proposals;
CREATE POLICY detection_proposals_tenant ON detection_rule_proposals
USING (
tenant_id = current_tenant_id()
OR tenant_id IS NULL
OR current_tenant_id() IS NULL
);
-- updated_at touch trigger (mirrors the agent_approvals pattern).
CREATE OR REPLACE FUNCTION detection_proposals_touch_updated()
RETURNS TRIGGER LANGUAGE plpgsql AS $$
BEGIN
NEW.updated_at = NOW();
RETURN NEW;
END;
$$;
DROP TRIGGER IF EXISTS trg_detection_proposals_touch ON detection_rule_proposals;
CREATE TRIGGER trg_detection_proposals_touch
BEFORE UPDATE ON detection_rule_proposals
FOR EACH ROW EXECUTE FUNCTION detection_proposals_touch_updated();
-- ============================================================
-- detection_eval_baselines
-- ============================================================
-- Stores the snapshot of MITRE accuracy / alert reduction / completeness
-- / response_quality the gate compares candidate proposals against.
-- A new baseline row is written by `scripts/run_evals.py --record-baseline`
-- and the latest active row per tenant (NULL = platform) is what the
-- gating endpoint reads.
CREATE TABLE IF NOT EXISTS detection_eval_baselines (
id UUID PRIMARY KEY DEFAULT uuid_generate_v4(),
tenant_id UUID REFERENCES tenants(id) ON DELETE CASCADE,
-- Eval suite this baseline scopes (mitre_accuracy, alert_reduction, ...).
suite VARCHAR(64) NOT NULL,
-- Aggregate score (e.g. 0.823 for 82.3 % MITRE accuracy).
score DOUBLE PRECISION NOT NULL,
-- Full eval result for diffing — same shape produced by run_evals.py.
payload JSONB NOT NULL DEFAULT '{}'::jsonb,
is_active BOOLEAN NOT NULL DEFAULT TRUE,
recorded_by_id UUID REFERENCES users(id) ON DELETE SET NULL,
created_at TIMESTAMPTZ NOT NULL DEFAULT NOW()
);
CREATE INDEX IF NOT EXISTS idx_eval_baselines_tenant_suite
ON detection_eval_baselines(tenant_id, suite, is_active);
ALTER TABLE detection_eval_baselines ENABLE ROW LEVEL SECURITY;
ALTER TABLE detection_eval_baselines FORCE ROW LEVEL SECURITY;
DROP POLICY IF EXISTS eval_baselines_tenant ON detection_eval_baselines;
CREATE POLICY eval_baselines_tenant ON detection_eval_baselines
USING (
tenant_id = current_tenant_id()
OR tenant_id IS NULL
OR current_tenant_id() IS NULL
);