Skip to main content

​
Overview

Every call and chat your agents have is logged as a conversation — recording, transcript, extracted variables, tool calls, and tags. The Conversations section is where you search and inspect them, independent of any single lead or process.
Open Conversations from the top navigation. The search bar accepts free text (it searches transcript text, lead ID, phone, or call SID) or a structured filter. Conversation Search list — call duration, turn count, playback, status, and tags per conversation The Summary panel on the right gives project-wide totals for the selected window: Total Conversations, Analyzed Conversations, Average Turns, and Average Duration.

​
Filters

Click Filters to build a structured query against fields including Lead ID, Call SID, Flow, Disposition, Tags, Tool Call, Duration, Turns, Updated At, Process, and phone. Conversation Search filter fields

​
Tags

Click Tags to filter by automatically-detected conversation issues — useful for quality review without listening to every call: Filter by Tags — Incomplete Call, AI Missed Turn, Short Disconnected Call, Disconnected Call, Voicemail Detected, AI Unresponsive, One Sided Call, High Negative Sentiment, High Positive Sentiment

​
Sampling

The More menu includes Sampling shortcuts (Best 1%, Worst 1%, Mixed 1%) to quickly pull a representative sample for review, and Export CSV to download the current result set.

​
Conversation Detail

Click into any conversation to open its detail page:
  • Recording & playback — the call audio, with duration and timestamp.
  • Overview / Conversation Messages tabs — the Overview tab shows call metadata; Conversation Messages shows the full turn-by-turn transcript, with a Translate option.
  • Tool Calls Made — every tool the agent invoked during the call (e.g. a price lookup), with the time offset it was called at.
Tool Calls Made panel on a conversation detail page
  • Conversation Variables — every variable extracted from or used in the call (lead fields, dynamic prompt variables, AI-extracted values like commentSummary or nextCallbackReasoning).
  • Token Usage — model token consumption for the call.
  • Text Simulator — replay the conversation against a different agent or model without making a real call.
Conversation transcripts and variables routinely contain real customer names, phone numbers, and other PII spoken or extracted during the call. Treat conversation detail pages as sensitive — don’t screenshot or export them outside your normal data-handling policy.

​
Feedback

To rate a conversation and leave notes for quality review, see Sharing Feedback on Conversations.

​
See Also