Dashboard

The dashboard is the “health + activity” overview for admins.

It typically answers:

• What is happening right now?
• Are knowledge bases healthy and up-to-date?
• Is usage increasing or dropping?
• Are we seeing quality issues (scores/feedback)?

Common Metrics

Exact labels can vary, but dashboards often include:

• Total knowledge bases
• Total sessions
• Total queries/messages
• Unique users
• Score summaries (if scoring is enabled)
• Recent activity feed

Drill-down Panels

Many dashboard cards open panels for deeper inspection, for example:

• recent sessions
• recent queries
• recent users
• score distributions

Operational advice:

• Use dashboard filters (KB filter, date range) before drawing conclusions.
• Always validate a dashboard spike by opening a few sessions/traces behind it.

Troubleshooting via Dashboard

A KB is “ready” but activity is zero

Possible causes:

• users aren’t discovering the KB
• chat link isn’t being shared
• KB is hidden from public listing (if applicable)

Recommended checks:

• open /ai-chat and verify the KB appears (if public listing is enabled)
• run a test query and confirm responses

Scores suddenly dropped

Possible causes:

• ingestion changes (extraction quality changed)
• index changes (chunking/embedding changed)
• upstream model change

Recommended workflow:

• compare traces before/after the drop
• inspect retrieved chunks for a representative set of queries