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 Threat-Hunting types — natural-language hunt surface (T3.4).
*
* Mirrors the wire shape of `services/api/app/api/v1/endpoints/saved_hunts.py`
* and `services/api/app/api/v1/endpoints/nl_query.py`. These are the
* canonical contracts shared between the API service, the Next.js web
* app, and any future SDK consumers.
*
* The types here intentionally mirror — not import from — the runtime
* shapes in `apps/web/src/lib/api.ts`. We keep the package pure (no
* runtime imports of the web app) so non-web consumers can use it
* standalone.
*/
/** Output dialects of the deterministic NL → query translator. */
export type HuntLanguage = "esql" | "kql" | "spl";
/**
* Translator output stored alongside the original NL question.
*
* The fields are always strings (never null/undefined) — an empty
* string indicates "translator could not produce this dialect" rather
* than "field absent".
*/
export interface TranslatedQueryEnvelope {
esql: string;
kql: string;
spl: string;
explanation: string;
}
/**
* Saved natural-language hunt persisted in `aisoc_saved_hunts`.
*
* Distinct from the hypothesis-driven hunt model in
* `services/api/app/api/v1/endpoints/hunts.py` (which has its own
* heavyweight schema with MITRE mappings, multi-platform queries, and
* findings rollups). This lighter-weight type backs the "Saved hunts"
* sidebar on the `/hunt` page.
*/
export interface SavedHunt {
id: string;
name: string;
/** Original English question the analyst asked. */
nl_query: string;
/** Snapshot translation taken at save time (and refreshed on each run). */
translated_query: TranslatedQueryEnvelope;
/** Preferred dialect to render in the editor when the hunt re-opens. */
language: HuntLanguage;
/**
* Optional 5-field cron string (`min hour dom mon dow`). When set,
* the API-side `hunt_scheduler` worker fires the hunt on cadence.
*/
schedule: string | null;
/** ISO-8601; null until the first run completes. */
last_run_at: string | null;
created_at: string;
updated_at: string;
/** UUID of the creating user, or null if the hunt was system-seeded. */
created_by: string | null;
}
export interface CreateSavedHuntRequest {
name: string;
nl_query: string;
language?: HuntLanguage;
schedule?: string | null;
}
/**
* Synchronous response from `POST /v1/saved-hunts/{id}/run`.
*
* The endpoint *re-translates* the NL question (so improvements to the
* translator land automatically) and stamps `last_run_at`. It does not
* actually execute ES|QL — that path is owned by `/v1/nl-query/execute`.
*/
export interface RunSavedHuntResponse {
id: string;
name: string;
nl_query: string;
translated_query: TranslatedQueryEnvelope;
last_run_at: string;
}
/** Body for `POST /v1/nl-query/translate`. */
export interface NlQueryTranslateRequest {
question: string;
index_pattern?: string;
time_range_hours?: number;
}
/** Response shape for `POST /v1/nl-query/translate`. */
export interface NlQueryTranslateResponse {
request_id: string;
question: string;
esql: string;
spl: string;
kql: string;
explanation: string;
created_at: string;
/** `deterministic` for the in-process grammar, `llm` for LLM-enhanced. */
engine: "deterministic" | "llm";
grammar_validated: boolean;
}