Transcripts¶
Status: Current | Last reviewed: 2026-10-12
Every conversation an agent handles is recorded as a transcript: the user side, the agent side, the timing, and any tools or workflows the agent invoked. Use transcripts to debug a specific call, audit agent behavior, and find examples for your QA process.
Open Conversation Search¶
On Home, choose the Conversations timeframe and select its number or label to open conversation search for that window, including test calls. Insights is deferred; it is no longer a navigation destination.
The list is filterable by:
- Date range
- Agent
- Agent version
- Status
- Channel (voice, chat, SMS, Teams, embedded)
- Caller
- Outcome
- Recording availability
- Archive state
- Test traffic
The default global view excludes test traffic. Turn on Include test when you need emulator, test chat, or test call sessions.
Per-Agent History¶
Open an agent and choose Activity → Conversation History. This modal shows a simple paginated list across all its versions and draft tests. Each row shows its version number or Draft, with no version selector in the modal. Select a conversation to view its transcript, then return to the same list page. Use global conversation search when you need detailed filters.
Read a transcript¶
Click any conversation to open its detail. You'll see:
- Header — agent name, version, channel, duration, outcome.
- Transcript pane — turn-by-turn, with timestamps. User turns are on the left, agent turns on the right.
- Audio playback (voice agents only) — listen to both sides of the call, including emulator test calls. Recordings are encoded for compactness; expect slight latency-skew vs the live call. Chat-only agents never show the audio section; a voice call with no recording shows a "no audio recording" notice instead.
- Tool calls — every tool the agent invoked, with inputs and outputs. Click one for the full payload.
- Workflows — every workflow triggered during the call, with run status.
Use Show Events in the transcript pane when you need the internal timeline. It interleaves persisted tool calls, tool results, pipeline markers, and subagent/active-agent events with the saved transcript. Leave it off for the basic conversation view.
Searching¶
The search box supports transcript full-text search. Use the filter controls for structured fields:
- Free text searches the transcript content.
- Caller narrows to known phone numbers, emails, or external caller IDs.
- Agent version filters to a specific config version.
- Archive state separates hot conversations from archived or purged detail rows.
Search results use cursor pagination so large tenants can move through long history without slow offset pages.
Archive State¶
Conversation rows show whether detail is Hot, Archived, Purged, Failed, or currently archiving. Purged rows keep searchable metadata in the list, but transcript, event, and recording detail is loaded from the archived bundle when opened.
Privacy and PII¶
If your agent has the pii_redaction workflow node enabled, the transcript stores the redacted version. Original content does not survive past the redaction step. For agents without redaction, the raw transcript is stored — apply your tenant's retention policy accordingly.
Note: Transcript read access follows the AI Manager agent-view permission. Users without access cannot open the list, even if you send them a direct link.
Exporting¶
The Export button on the transcript list produces a JSONL file for your filter. Use it for QA spot-checks, model-eval datasets, or compliance archives. Exports are throttled to once per minute per user.