Live Video Alert Agent#

Deploy AI-powered video alerting using OpenVINO Vision Language Models to process RTSP streams, generate real-time alerts from natural language prompts, and respond automatically through a configurable agentic action pipeline.

Use Cases#

Real-time Video Analytics: Monitor security cameras, industrial equipment, or public spaces with AI-powered scene understanding and automatic alerting.

Safety Monitoring: Deploy prompts like “Is there a fire?” or “Is anyone wearing a safety vest?” to trigger immediate notifications via webhook or MQTT.

Agentic Alert Response: Use Google ADK or a local LLM to reason about alert severity and automatically select which tools to invoke — snapshot, webhook, or MQTT.

Custom Alerts: Use natural language to define what constitutes an alert without retraining a model.

Key Features#

Dynamic Alert Prompts: Define and modify alerts (prompts) in real-time through the UI or REST API without redeploying.

Agentic Tool Dispatch: When an alert fires, an action agent decides which tools to invoke — powered by Google ADK, an OVMS-hosted LLM, or deterministic rule-based execution.

Alert State Management: Per-stream, per-alert cooldowns and consecutive-detection escalation suppress noise while ensuring persistent conditions trigger escalated responses.

Built-in Action Tools: log_alert, capture_snapshot, trigger_webhook (HMAC-signed), publish_mqtt (MQTTv5).

Concurrent Multi-Camera: Each camera stream runs in its own independent asyncio task — slow or stalled cameras do not block others.

Real-time Event Broadcasting: SSE delivers analysis results and alert_action events instantly to the dashboard with low latency.

Observability Endpoints: /health, /ready, /metrics for liveness probes, readiness checks, and CPU/memory monitoring.

Intel® Hardware Optimized: Designed for high-performance inference on Intel® CPUs and GPUs via OpenVINO.