Conversation Details is the full record of one conversation. It shows who the customer was, where the conversation ran, the transcript and the flows it passed through, what AI agents, guardrails and HTTP calls did along the way, and the conversation's AI Insights and Custom Attributes. This article covers reading that page and exporting a single conversation. Finding conversations, filtering them and exporting the list are covered in Conversations.
For the date range, retention and which conversations the reports include, see Reports.
Opening a conversation
- Open Insights > Conversations.
- Click the conversation's Date. Only the Date cell opens the conversation. Clicking anywhere else on the row does nothing.
If you already have the ID, type it in the Conversation/Session ID search. When exactly one session matches, that conversation opens directly.
Previous and Next step through the rows on the list page you opened the conversation from.
The page
1. Conversation and Traces tabs: Conversation shows the transcript. Traces shows how each message was processed. Traces appears only for users with the Traces permission. Without that permission, the page shows only the transcript.
2. Previous / Next: Moves to the previous or next conversation in the list you came from.
3. Execution Details (gear icon): Switches that add AI agent, guardrail and HTTP call details to the transcript.
4. Export: Exports this one conversation.
5. Left card: Customer, project, AI Insights and Custom Attributes for the conversation.
6. Transcript: The messages, in order.
7. Flow timeline: The flows the conversation passed through, with a dot for each customer message.
The left card
From top to bottom, the card holds these blocks.
| Block | What it shows |
|---|---|
| CUSTOMER INFO | Name, Phone Number, E-mail, Id and Customer Sentiment. Hover the sentiment icon to see its score, from -1 to 1. |
| PROJECT INFO | Project, Date, Channel, Conversation ID, Session ID and Duration. The two IDs each have a copy icon. Rating Score and Rating Comment appear here too if the conversation was rated. |
| AI Insights | The insight values generated for the conversation. |
| CONVERSATION SUMMARY | A summary you generate on request. |
| Custom Attributes | The attribute values your flows recorded. |
Some blocks appear only under certain conditions:
- CUSTOMER INFO is hidden when the conversation has no customer information. Customer Sentiment appears only if text sentiment is enabled for your tenant. It is off by default.
- AI Insights appears only while AI Insights is on for the project. Turning it off also hides values that were already stored.
- CONVERSATION SUMMARY appears only if conversation summarization is enabled for your tenant and the conversation has no AI Insights Summary value. Click Generate Summary to create a summary. It is saved with the conversation.
1. AI Insights: Free text insights such as Summary and Topic show as text. Category insights show as chips. An insight with more than three values shows the first three and then +N. Click +N to see the rest and Show less to collapse them again. If insight generation failed for this conversation, Generate AI Insights appears under the heading. AI Insights explains when the button appears and what it does.
2. Custom Attributes: Displayed the same way as the insights. To define attributes, see Custom Attributes.
The transcript and flow timeline
- Time on each bubble: How much time had passed since the conversation started when the message was sent.
- Who replied: When a live agent or an AI agent answered, its name appears above the bubble.
- Show/Hide Details: Appears on messages that have intent matches, multiple matches, speech recognition fallback confidence or extracted entities. Click it to show or hide them.
- Original / Translated: Appears when messages were translated. It switches the transcript between the original and the translated text.
- Voice: Each voice message can be played from its bubble. IVR conversations also have Start Conversation / Stop Conversation on the timeline, which plays the whole call.
The flow timeline along the bottom has one segment per flow. Hover over a segment to see the flow's name. Each customer message is a dot, coloured by its sentiment. Click a dot to jump to that message in the transcript.
Execution Details
Click the gear icon to the left of Export and turn on the details you want to see in the transcript.
| Switch | Available |
|---|---|
| Show AI Agent tool call details | When AI agents are available |
| Show guardrail details | When AI agents are available |
| Show HTTP call details | Always |
| Show AI Agent Invoke Failure | When AI agents are available |
All four are off by default. Your choices are remembered in your browser, for each project separately.
📝 Note: These are the same switches as the monitoring switches in Test Chat. Turning one on or off here does the same in Test Chat, and the other way round.
What each detail shows
Each detail appears in the transcript at the moment it happened, as a row you can expand.
Guardrail details: One row for each guardrail check. On the customer's turn, maskedInput is the message as the AI agent's model received it, after PII masking. On the agent's turn, unmaskedOutput is the answer as the model wrote it, before output masking. The customer received the masked answer shown in the transcript. violated and violationType show whether the guardrail blocked the turn, and why.
Tool call details: Tool Call Request shows the tool's name and the parameters the AI agent passed to it. Tool Call Result shows what the tool returned to the agent.
HTTP call details: HTTP Request shows the method and URL. Under Headers it lists the headers typed into the HTTP node; a credential selected in the node is never listed. For a call made with mTLS, it shows that mTLS was used and the certificate's name. HTTP Response shows the status code, the duration and the response body. To keep secrets out of the flow, see Securing HTTP Requests.
After the retention period for execution details has passed, the page shows "Execution details for this conversation have expired and are no longer available." and the switches are greyed out.
Traces
1. Trace header: The trace ID, the date and time, and the total duration in milliseconds.
2. Steps: A waterfall of the steps that processed the message, each with its own duration.
To go straight to one message's trace, click the trace icon on its bubble (the tooltip reads "Click to view the message trace"). Only messages that have a trace show this icon. The page switches to Traces and highlights that trace.
Traces come from the monitoring system and follow its retention, not the conversation's. If none are left, the tab shows "No traces data available".
Exporting one conversation
- Open the conversation.
- If the messages were translated, set Original / Translated to the text you want. The export uses whichever version is selected.
- Click Export. The file is prepared in the background, and a notification tells you when it is ready.
The export is an .xlsx file with one sheet, Transcript. It contains:
- the page's header fields
- the summary, the last intent and the rating
- the customer fields
- the messages
- the conversation's AI Insights and Custom Attributes
If the conversation includes voice, you get a ZIP file instead. It also contains an audio_files folder with the recordings as .wav files.
Execution details are not included in the export.
When masking is on
With masking on, this page masks message text, the summary, AI Insight values and customer fields.
Masking a conversation takes some time after it ends. Until masking is finished, the text reads "Masking in progress" and you cannot use Export, Execution Details, Generate AI Insights or Generate Summary. When masking is complete, all four are available again.