How to Use the Learning Agent

Open Dashboard > Learning Agent. The header shows the brain icon and a dropdown to scope everything to a single agent or All agents. The refresh icon reloads all data.

Below the header you get two always-on sections plus a tabbed area with three views.

The first card on the page rolls up how Kai is doing right now — total calls, lead capture rate, average call quality, and trend arrows comparing this week to last. This is the same data the call quality dashboard surfaces, summarized.

The second card lists items that need your input. Two main types:

• Proposed changes — Kai noticed a recurring problem on calls and is suggesting a prompt edit. Click Approve to apply it or Reject to dismiss.
• Edge cases — specific calls where something unusual happened that Kai is not sure how to handle. Mark them Resolved once you have decided what to do.

Approving a change writes it to your agent's prompt and logs the change to your audit history.

Default tab. Each insight card describes a pattern Kai found in your data — for example, “calls between 4–6pm have a 30% higher conversion rate” or “callers asking about pricing tend to drop off when Kai pivots to scheduling.”

If no insights exist yet, click Generate Insights to ask Kai to analyze your call patterns now. Otherwise insights appear automatically once Kai has enough data to find meaningful patterns.

An inline summary of the same metrics the dedicated quality dashboard shows — alert banner, the four headline metrics, and a per-agent performance table. Click View Full Quality Dashboard at the bottom for the complete view with latency charts, cost breakdown, and time patterns.

When Kai has two variants of a prompt running side-by-side, this tab shows the comparison: how each variant is performing on the metrics that matter, and a Promote button to lock in the winner. Once promoted, the winning variant becomes the live prompt and the test ends.

When you pick a single agent (not All agents) from the dropdown, an extra panel appears at the bottom with that agent's learning-loop settings — how aggressively Kai should propose changes, what metrics to optimize for, and whether to auto-apply low-risk fixes without asking you first.

• Call Quality — full performance dashboard with latency, costs, and time patterns
• Customize Your Agent — manually edit the prompt the learning agent is improving
• Knowledge Library — give Kai more facts to work with, fewer edge cases to flag