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
id: ai-gateway
name: AI Gateway / MCP Runtime
version: "1.0.0"
plugin_type: connector
tier: stable
description: >
Pulls request logs from an LLM gateway or MCP server (LiteLLM, Portkey,
Helicone, or an in-house proxy) so a SOC can see which agents call which
tools and models, on whose behalf. One connector covers every application
behind the gateway, which matters because the apps most in need of
visibility are the ones nobody will retrofit an SDK into. For applications
you do control, packages/aisoc-ai-sdk is the lower-friction push path.
Routine activity is archived to the event lake and hunted; a guardrail
violation becomes an alert.
author: AiSOC Core Team
tags: [ai, ai-spm, llm, agents, mcp, guardrails, runtime, connector]
homepage: https://github.com/beenuar/AiSOC/tree/main/services/connectors/app/connectors/ai_gateway.py
license: MIT
min_aisoc_version: "8.0.0"
config_schema:
type: object
required: [base_url, api_key]
properties:
base_url:
type: string
description: "Gateway base URL including scheme (e.g. https://llm-gateway.internal)"
api_key:
type: string
description: "Gateway API key with read access to request logs"
secret: true
approved_models:
type: string
description: >
Comma-separated list of approved models. Any model outside the list
raises an unapproved_model finding. Leave blank to disable the check
rather than flag every model.
verify_tls:
type: boolean
default: true
description: "Disable only for an internal gateway using a private CA"