Alert Agent Service#
The Alert Agent Service is a generic, multimodal alert action dispatcher microservice. It receives alert events from upstream detection pipelines (such as video analytics, audio sensors, or IoT devices), reasons over the alert context using an LLM-powered agentic backend, and dispatches one or more configurable action tools — such as logging, webhook notifications, MQTT publishing, and snapshot saving.
Key Features#
Multimodal Payloads: Accept text, image, audio, video, and binary media artifacts attached to alert events.
ADK-Powered Dispatch: LLM-based agentic reasoning via Google ADK and OpenVINO™ Model Server for intelligent tool selection.
Rule-Based Dispatch: Deterministic tool invocation without LLM overhead when
AGENT_MODE=false.Subscription Config: YAML-driven default routing so upstream callers only need to send minimal request fields.
Deduplication: Suppress repeated alerts within configurable time windows to avoid notification floods.
Escalation: Automatically invoke additional tools (e.g., MQTT) after a configurable number of consecutive detections.
MCP Integration: Dynamically extend the tool set with tools from external MCP servers without code changes.
SSE / WebSocket Streaming: Real-time event fanout to monitoring dashboards and downstream consumers.
Hot-Reload: Refresh tool and MCP configurations at runtime without restarting the service.
Use Cases#
This microservice is ideal for:
Video analytics pipelines requiring automated alert responses (webhook notifications, MQTT publishing, snapshot archiving)
Edge AI applications that need LLM-guided alert triage
Multi-sensor environments (cameras, microphones, IoT) with unified alert handling
Monitoring dashboards consuming real-time alert event streams