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
/**
* Deterministic natural-language → Sigma generator for `/tools/nl2sigma`.
*
* Runs in the browser with no LLM: it extracts high-signal artifacts from the
* plain-English description (process names, ports, MITRE ids, event verbs) and
* assembles a well-formed Sigma rule. It is intentionally a scaffold generator
* — honest about being a starting point, not a finished detection — and pairs
* with the field-map translator to emit the SIEM dialects.
*/
export interface Nl2SigmaResult {
sigma: string;
matchedKeywords: string[];
note: string;
}
const LOGSOURCE_HINTS: { re: RegExp; category: string; product: string }[] = [
{ re: /\b(powershell|cmd|process|command[- ]?line|executable|binary|spawn)\b/i, category: "process_creation", product: "windows" },
{ re: /\b(login|logon|sign[- ]?in|authentication|failed password|brute)\b/i, category: "authentication", product: "windows" },
{ re: /\b(dns|domain|resolve|query)\b/i, category: "dns", product: "" },
{ re: /\b(network|connection|port|outbound|beacon|c2)\b/i, category: "network_connection", product: "windows" },
{ re: /\b(file|write|create|drop|ransom|encrypt)\b/i, category: "file_event", product: "windows" },
];
const PROC_RE = /\b([A-Za-z0-9_-]+\.(?:exe|ps1|dll|bat|sh|py))\b/gi;
const MITRE_RE = /\bT\d{4}(?:\.\d{3})?\b/gi;
const PORT_RE = /\bport\s+(\d{1,5})\b/i;
function yamlList(items: string[], indent: string): string {
return items.map((i) => `${indent}- '${i.replace(/'/g, "")}'`).join("\n");
}
export function nlToSigma(description: string, severity = "medium"): Nl2SigmaResult {
const text = description.trim();
const matched: string[] = [];
const hint = LOGSOURCE_HINTS.find((h) => h.re.test(text));
const category = hint?.category ?? "process_creation";
const product = hint?.product ?? "windows";
if (hint) matched.push(category);
const procs = [...new Set([...text.matchAll(PROC_RE)].map((m) => m[1]!))];
const mitre = [...new Set([...text.matchAll(MITRE_RE)].map((m) => m[0]!.toUpperCase()))];
const portMatch = text.match(PORT_RE);
procs.forEach((p) => matched.push(p));
mitre.forEach((t) => matched.push(t));
// Keyword candidates for a CommandLine|contains selection: notable verbs/nouns.
const keywordCandidates = (text.match(/\b(download|invoke|encode|iex|mimikatz|whoami|net user|reg add|schtasks|vssadmin|bitsadmin|certutil|rundll32|regsvr32)\b/gi) ?? []).map(
(k) => k.toLowerCase(),
);
const uniqueKeywords = [...new Set(keywordCandidates)];
uniqueKeywords.forEach((k) => matched.push(k));
const title = text.slice(0, 70).replace(/\n/g, " ") || "Suspicious activity";
const detectionLines: string[] = ["detection:", " selection:"];
if (procs.length) {
detectionLines.push(" Image|endswith:");
detectionLines.push(yamlList(procs.map((p) => `\\${p}`), " "));
}
if (uniqueKeywords.length) {
detectionLines.push(" CommandLine|contains:");
detectionLines.push(yamlList(uniqueKeywords, " "));
}
if (portMatch) {
detectionLines.push(` DestinationPort: ${portMatch[1]}`);
}
if (!procs.length && !uniqueKeywords.length && !portMatch) {
detectionLines.push(" # No concrete artifacts detected in the description — refine the selection below.");
detectionLines.push(" CommandLine|contains: '<REPLACE_ME>'");
}
detectionLines.push(" condition: selection");
const tagsBlock = mitre.length ? `tags:\n${mitre.map((t) => ` - attack.${t.toLowerCase()}`).join("\n")}\n` : "";
const logsourceProduct = product ? `\n product: ${product}` : "";
const sigma = `title: ${title}
status: experimental
description: ${text.replace(/\n/g, " ") || "Generated from a natural-language description."}
${tagsBlock}logsource:
category: ${category}${logsourceProduct}
${detectionLines.join("\n")}
falsepositives:
- Legitimate administrative activity
level: ${severity}`;
return {
sigma,
matchedKeywords: matched,
note: "Deterministic scaffold generated in your browser — a starting point, not a finished detection. Refine the selection and false positives before deploying.",
};
}