Skip to content

AI Chat events

AI Chat Events is platform-administrator telemetry for the portal's AI assistant. Use it to diagnose routing, access and verification decisions, query safety, returned-row counts, execution performance, and user-reported AI failures.

Open General Administration and select AI Chat Events. The view is global rather than site-scoped and shows the newest records matching the selected Status, Capability, and row limit.

Filter and refresh

Status choices are All, success, error, clarification, access denied, verification failed, and simple AI. Capability choices are All, site, visitor, logistics, access, LPR, and inventory. Row limits are 50, 100, 250, or 500; the default is 100.

Change the filters and select Refresh. There is no date, site, user, request-ID, thread-ID, or text-search filter in this dialog. An empty result can also occur when telemetry storage cannot be queried; correlate it with application health and server logs if known events are missing.

Interpret the grid

Column Meaning
Created Stored telemetry timestamp, displayed without an explicit time-zone label.
Status High-level processing result.
Capability Functional area selected for the request.
Intent / Domain Interpreted request and data domain.
User Internal user identifier, not the display name.
Scope Site-scope type and number of sites considered.
Rows Number of rows returned by the data operation when applicable.
Duration End-to-end recorded duration in milliseconds.
SQL Blocked Safety controls prevented generated SQL from executing.
Mode Execution route used for the response.
Error Recorded processing or validation failure.
Message Preview Truncated or preview form of the user's request.

SQL Blocked is a safety outcome, not a reason to bypass controls. Review the request, capability, site scope, and supported AI behavior. Rephrase or correct the authorized workflow rather than attempting to force unsafe SQL.

Rows Returned alone does not establish answer quality. A small result can be correct for a narrow scope, while a truncated result or unexpected scope can make an otherwise successful response incomplete. Some additional fields—including truncation, generated or retried SQL, time window, route names, metric source, request ID, thread ID, and full telemetry JSON—are available only in the Events CSV.

Export events

Events CSV runs a new query using the currently selected Status, Capability, and row limit. It includes the grid data plus request ID, thread ID, route names, time window, metric source, truncation flags, SQL-generation and retry flags, and the complete stored telemetry JSON.

For consistent review, select filters, refresh the grid, and then export. If filters were changed without refreshing, the export follows the current controls and can differ from the visible grid.

Memory CSV first selects the currently filtered recent telemetry events, then exports AI memory rows sharing their user ID and thread ID. It can include up to 5,000 memory rows and contains the full stored user query and AI response, ordered by memory-row ID.

Memory CSV can therefore contain conversations beyond the one previewed event when several selected events share a thread. Events without both a user ID and thread ID do not contribute memory rows.

Danger

AI exports can contain personal data, operational details, full prompts and answers, internal identifiers, route metadata, and diagnostic JSON. Export only for an authorized purpose, store the file in an approved location, restrict access, and delete it according to the applicable retention process.

AI Chat Events is diagnostic only. It cannot edit a conversation, rerun a request, change scope, grant access, or modify AI configuration.