Documentation

# Alerts, events & the incident database

Watch classifies analyses as safe, suspicious, or critical. Review live results in the log and saved critical incidents in the database.

## Three states

**Safe**

No condition requiring action was detected.

**Suspicious**

Worth review; more confirmation may be needed.

**Critical**

Reached the configured notification threshold.

## Set visual alerts and sounds

1.  Open Settings → Monitoring mission and find the alert settings below the mission.
2.  Choose the visual treatment and sound for suspicious and critical results separately. Use the play button beside the sound to check your computer’s volume and output device.
3.  Return to the camera, test the mission with a known scene, and compare the log severity with the visual alert.

These settings control how an alert looks and sounds. Change the [mission conditions](https://watch.tensordynamics.org/docs/missions?lang=en) to change what counts as suspicious or critical. A successful sound test confirms local audio playback only.

## Notification destinations

In Settings → API & integrations, use the read-only API or expand Advanced event delivery to configure a webhook. Forwarding sends created events matching its source and severity filters. Standard v0.39 does not expose Agent → Channels. Ask support about availability before using optional chat or email channels. Verify delivery for each configured destination.

## Review saved events

1.  Open Database in the sidebar. Enter a camera, start/end time, or event text, then select Search.
2.  Review matching records, descriptions, and available snapshots. If no event appears, clear filters or widen the time range.
3.  Select Export Excel or Export JSON to download records matching the current filters. Open the file and check its timestamps and cameras.

In v0.39, live monitoring stores confirmed critical incidents and subsequent evidence in the incident database, with the time, camera, space, analysis, and any available snapshot. Safe and suspicious analyses are visible live in Watch, but are not saved individually as permanent incidents. Use the Database page to search saved critical records and export the evidence you need.

In v0.39, a confirmed critical condition creates a live-camera event. Recovery does not create another event; check the latest analysis for the current state. An empty history means no saved records match your filters. Also check the camera’s live picture and latest analysis time.

To read events from another system, follow the [v0.39 read-only API quick start](https://watch.tensordynamics.org/developers/quickstart?lang=en). See that guide for the data available through the LAN API.

Critical snapshots are retained for 14 days by default. Adjust or disable retention to match your privacy and data-retention obligations.

## Reducing false alarms

Specify the area and behavior to watch on each camera, then refine the mission using site tests. Watch confirms critical conditions before escalating an alert.

[

Next

Analyze recorded video

](https://watch.tensordynamics.org/docs/offline-analysis?lang=en)

---
Source: https://watch.tensordynamics.org/docs/alerts
More Watch guides: https://watch.tensordynamics.org/llms.txt