# Workflows and Agents: What Runs Today

Start with the operational result and choose the smallest workflow that delivers it. An interactive answer, a saved notification and a custom agent are different levels of implementation.

<table id="bkmrk-job-documented-route"><thead><tr><th>Job</th><th>Documented route</th><th>Boundary</th></tr></thead><tbody><tr><td>Find when a unit was last seen</td><td>Interactive assistant query</td><td>A tool call retrieves an observation. It does not start ongoing monitoring.</td></tr><tr><td>Review activity or visit durations</td><td>Supported server-side analytics</td><td>Requires suitable records, scope and eligibility; incomplete visits remain distinct.</td></tr><tr><td>Tell me when a unit is observed again</td><td>Detection alert</td><td>Matches newly stored detections. Delivery is to your own verified account email.</td></tr><tr><td>Send a regular activity breakdown</td><td>Scheduled aggregate digest</td><td>Supports documented count metrics and schedules. It does not provide dwell or occupancy digests.</td></tr><tr><td>Update another system or perform a business action</td><td>Separately scoped workflow or agent</td><td>Availability, runtime, external access and approval policy must be agreed.</td></tr></tbody></table>

## Example: find a trailer

> Using SiteTrax.io, find the latest observation of trailer \[complete identifier\] at \[project\]. Show the recorded time and timezone, the project, the record link and any missing evidence.

Expected result: the last recorded observation and a way to check it. If there is no match, check the identifier, project access and time window before concluding the unit is absent.

## Example: prepare a detection alert

> Prepare a one-time alert for the next new detection of trailer \[complete identifier\] at \[project\], expiring in 30 days. Show the exact filter, recipient and expiry, then ask me to confirm before creating it.

Creating or cancelling a notification requires `sitetrax:notifications:write` in addition to read access. Listing notifications and delivery history uses read access. A saved notification ID confirms configuration; a delivery-history entry records dispatch status. Neither proves the email was read.

## Example: scope an exception workflow

**Design example, not a claim of an available packaged agent.** Compare observed equipment with an expected list from a customer system, prepare an exception and route it to an operator for review. Agree the external source, identifier matching, freshness limit, approval step and destination before implementation.

## What to specify for a custom workflow

<table id="bkmrk-decision-example-tri"><thead><tr><th>Decision</th><th>Example</th></tr></thead><tbody><tr><td>Trigger</td><td>New eligible observation, an agreed schedule, or an operator request.</td></tr><tr><td>Evidence</td><td>SiteTrax.io project, equipment family, time window and supporting records.</td></tr><tr><td>Business context</td><td>Expected list, contract or schedule supplied by an authorized system.</td></tr><tr><td>Action</td><td>Draft an exception for review, or an explicitly permitted system update.</td></tr><tr><td>Control</td><td>Who approves, what happens on missing evidence, and how duplicate actions are prevented.</td></tr><tr><td>Success measure</td><td>Time to answer, reviewed exceptions resolved, or time from observation to approved action.</td></tr></tbody></table>

For exact alert types, schedules, metrics and limits, use [Alerts and Scheduled Digests](https://docs.sitetrax.io/books/operational-use-cases/page/alerts-and-scheduled-digests).