Agent Studio

Conversations (HITL)

flowchart LR
  A["Filter conversations<br/>status, sentiment, gaps"] --> B["Inspect detail<br/>timeline, actions, analysis"]
  B --> C{"Issue found?"}
  C -->|"Knowledge gap"| D["Add to Knowledge Base"]
  C -->|"Action bug"| E["Debug handler"]
  C -->|"Looks good"| F["No action needed"]
  style A fill:#e8ecff,stroke:#1842ef,color:#0a0e1a
  style B fill:#e8ecff,stroke:#1842ef,color:#0a0e1a
  style C fill:#e8ecff,stroke:#1842ef,color:#0a0e1a
  style D fill:#dbe2ff,stroke:#1842ef,color:#0a0e1a
  style E fill:#fef3c7,stroke:#d97706,color:#0a0e1a
  style F fill:#d1fae5,stroke:#059669,color:#0a0e1a

Path: Agent Studio → Conversations

Conversations shows every real interaction users have had with your agent. It’s your primary tool for quality review and debugging.

Summary metrics

The top bar summarizes activity across all conversations:

Metric	What it means
Total Conversations	All sessions where a user interacted with the agent
Questions Asked	Total user messages sent
Conversation Handoffs	Times the agent escalated to a human
Actions	Total action calls executed by the agent

Conversation list

The left panel lists conversations sorted by recency. Each row shows the user’s name, the auto-generated conversation topic, and the time. Click a row to open the full conversation.

Filters

Use the filter bar to narrow the list by:

Resolution status — resolved, unresolved, or escalated.
User Feedback — thumbs up/down ratings from users.
Actions — conversations where a specific action was triggered. Use this to see how a published action behaves in the wild, then open The action lifecycle to iterate. Connected A2A agents appear in the same dropdown by name, tagged with an A2A badge — select one and click Apply to see every conversation that delegated to it.
Conversation Signals — auto-detected tags on each conversation. Values include User Sentiment, Information Request, Action Request, and Reported Issue — plus the two that drive your improvement worklist:
- Agent Knowledge Gap — the agent lacked the content to answer. Each one points to a Knowledge Base article to add or refresh.
- Agent Action Gap — the user asked for something the agent had no action for. Each one is a candidate for a new action.
Source — where the conversation came from: Live Agent, Slack, Trust Lab, or SDK Test. Select SDK Test to surface test mode sessions, which are hidden by default so real user activity stays clean.

Conversation detail

The right panel shows the full message-by-message exchange. At the bottom of each agent message there’s a View Analysis link — click it to see how the agent interpreted the request, which knowledge it used, and which actions it called.

User details and session replay

The right-side panel shows User Details (name, email, product) along with any Session Replay recording associated with the conversation.

When you spot a problem, the fix usually lives elsewhere: add missing content in the Knowledge Base, or correct a capability in Actions.

Conversation metrics: the events that record volume, resolution, and escalation across every conversation.
Session replay: watch the session behind a conversation.