Video Observability: Monitoring Your Operations Through Video
Video observability is the practice of using continuous video analysis to maintain operational awareness — the same way application observability uses logs, metrics, and traces to monitor software systems.
Observability vs monitoring vs surveillance
| Approach | Input | Output | Who acts |
|---|---|---|---|
| Surveillance | Raw recorded video | Archive for manual review | Human after incident |
| Monitoring | Motion events, alerts | Triggered notifications | Human in response |
| Observability | Continuous AI analysis | Logs + metrics + traces + search | System automatically surfaces anomalies |
The three pillars applied to video
Every detected event with timestamp, type, persons involved, and location. Queryable via AI chat: "Show me all idle events on Line 3 last week over 60 seconds."
Quantitative measures over time: cycle time per shift, utilization rate per operator, crowd density per hour. Exportable to BI tools and OTel collectors.
End-to-end process execution: from first operator action to completion, with each step timed and classified. Maps to OpenTelemetry span concepts.
VidForgeX observability infrastructure
VidForgeX logs the lifecycle of every analysis — started, completed, failed — to a centralised log service. Raw request bodies and AI payloads are never logged, and credentials are scrubbed before anything is stored.
Analysis results are stored durably in encrypted object storage (JSON artifacts) with metadata in a managed database. The notification system provides in-platform observability for analysis status. For operational use cases requiring real-time dashboards, the MCP API enables custom integrations: pollget_team_summary on a schedule and pipe results to your preferred metrics platform.