> ## Documentation Index
> Fetch the complete documentation index at: https://docs.wittify.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Voice Conversations
> The full call inbox for one agent. Browse every voice call across web, SIP, Telnyx, Twilio, and WhatsApp Calls. Play back recordings with click-to-seek waveforms, read synced transcripts, generate AI summaries, and export.
Voice Conversations lives at `/agents/{agentId}/voice-conversations`. Three sub-tabs (collapsed by default behind a chevron toggle on the parent row): **Inbox** / **Archived** / **Deleted**. Hidden entirely for **text-only** agents.
## Layout
Two-pane on desktop, single-pane on mobile. Same shape as Chat Conversations: list pane on the start edge, detail pane on the other side, optional right panel that slides in alongside on demand. On mobile, only one pane renders at a time.
## Source tabs
A row of 5 source filters at the top of the list pane. Each filter restricts the list to calls that came in through that route:
| Tab | Description |
| ----------------- | --------------------------------------------------- |
| **Web** | Calls placed from the website widget's voice button |
| **SIP** | SIP Trunk inbound / outbound calls |
| **Telnyx** | Calls routed via your Telnyx connection |
| **Twilio** | Calls routed via your Twilio connection |
| **WhatsApp Call** | Voice calls placed via WhatsApp |
The active tab shows the source's brand color. Each tab carries a count chip.
## Toolbar
* **Search box**, placeholder *Search calls...*. Matches name, call ID, phone number, and transcript preview.
* **Sort** dropdown: **Newest** (default), **Oldest**, **Unread first**, **Longest first**.
* **Filter** popover with multi-select chips for:
* **Status** (Open, Resolved, Handoff)
* **Sentiment** (Positive, Neutral, Negative)
* **Read** (Read, Unread)
* **Tags**
* **Duration bucket**: **Less than 1 minute**, **1 to 5 minutes**, **More than 5 minutes**
* **Rating**: 1 to 5 stars, plus a **No rating** chip
* **Assignees**
* **Active filters** chip with a **Clear** button to reset.
## Bulk actions
Same shape as Chat Conversations, with the same icon-button bar: **Mark read** / **Mark unread** / **Assign** / **Archive** (confirmation) / **Delete** (red confirmation). The select-all checkbox is on the start of the bar.
## Voice row
Each row in the list shows:
* Checkbox
* `#number`
* Customer name (or **Unknown**)
* Phone number
* Duration pill (e.g. *3:42*)
* Source brand-color dot
* Sentiment chip
* Status chip
* Rating: amber stars when `rating > 0`, otherwise nothing
* Ticket / lead pills
* Assignee chip
## Detail pane
### Detail header
* `#number` and customer name
* Source icon
* Phone
* Duration
* Tags row including rating stars
* Linked records bar (same shape as Chat Conversations, routes to the matching ticket / lead)
* On mobile, a back arrow returns to the list
### Player card
The audio player for the call recording.
| Element | Behavior |
| ---------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| **Waveform** | 60 to 120 vertical bars representing the call's amplitude. Click any bar to seek to that point in the audio. The waveform is deterministic per call (same shape on every page load). |
| **Brand-gradient Play / Pause** button | Centered. Click to start, click again to pause. |
| **Skip back 10s** / **Skip forward 10s** | Two icon buttons either side of the play button. |
| **Keyboard shortcuts** | Left / Right arrow keys skip 5 seconds. Spacebar is reserved by the page. |
| **Volume slider** | Hover to expand. |
| **Playback speed** dropdown | **0.5x**, **0.75x**, **1x** (default), **1.25x**, **1.5x**, **2x**. |
| Time display | Shows current time and total duration as `mm:ss / mm:ss`. |
### Transcript card
The full transcript synced to the audio.
* **AI / agent turns** appear on the start edge with the agent avatar.
* **Caller turns** appear on the end edge with a generic avatar.
* The **active turn** (the one being spoken right now) is highlighted with a tinted background as the audio plays. Highlighting stays in sync via per-turn timestamps.
* The card has a fixed height (around 360 pixels) and scrolls. A 64 pixel spacer at the bottom prevents the last turn from being clipped against the card border.
* **Download Transcript** button on the corner exports a UTF-8 file with BOM (so Arabic renders in Excel). The filename is sanitized from the call ID (no path separators, max 64 chars).
## AI Call Summary
A card inside the right panel's Details tab. Generated on demand, not automatically.
Three states:
| State | What you see |
| ----------- | ------------------------------------------------------------------------------------------------------- |
| **Idle** | A **Generate summary** button. |
| **Loading** | A spinner with the line *Wittify AI is summarising the call...* |
| **Ready** | A 3 to 5 sentence summary of the call's key points and outcome. A **Regenerate** button below to rerun. |
The summary is intentionally not auto-generated, calling the model on every load would be wasteful and slow. Click to summon, regenerate if the first answer misses something.
## Right panel
A drawer that slides in on demand. Three tabs:
### Details tab
Several stacked sub-cards:
* **Call Info**: ID, Source, Direction (Inbound / Outbound), Detected language
* **Performance**: Duration, Response latency, Ratings
* **Timestamps**: Started at, Ended at, Recording uploaded at
* **AI Call Summary** (described above)
* **Quick Actions**: small buttons for Mark read / Mark resolved / Add tag
### Assign tab
Same fields as Chat Conversations: **Agent**, **Team**, **Priority**, **Note**, **Save Assignment**.
### Routing tab
Voice-specific routing controls (e.g. forward future calls from this number to a specific teammate, suppress voicemail).
## Mobile layout
Single-pane like Chat Conversations. The right panel mounts inside a slide-in drawer on mobile (a single mount of the panel for the active call, never two at once).
## Recording lifecycle
What happens after a call ends:
1. **Call ends** , the server uploads the recording, computes a waveform, and starts transcription.
2. **Transcript ready** , the detail pane shows the full text.
3. **Summary requested** , click **Generate summary** to run an AI summary over the transcript.
4. **Download** , transcript export is available as soon as transcription finishes. Audio export is on the roadmap.
## Sub-tabs
Same as Chat Conversations:
| Sub-tab | What's in it |
| ------------ | -------------------------------------------- |
| **Inbox** | Live calls |
| **Archived** | Calls you've archived. Still searchable. |
| **Deleted** | Soft-deleted calls. Recoverable for 30 days. |
## Common questions
That happens when the recording uploaded but waveform generation failed. The audio still plays, you just don't get click-to-seek bars. Refresh the page or contact support.
Click **Regenerate**. The model occasionally misses nuance on the first pass. If it's still wrong, the transcript card below has the raw text, you can read the relevant section yourself.
The player streams the audio in chunks. On very slow connections, pause until the buffer fills, then resume. Click ahead in the waveform to skip past slow-loading portions.
Not yet. Transcript export (UTF-8) is available now via the **Download Transcript** button. Audio export is on the roadmap.
This means a customer dialed your WhatsApp number from inside their WhatsApp client. If WhatsApp Calls is off in [Deployment](/en/Systems/AI-Agents/Wizard/Deployment), the agent didn't answer, but the missed call still appears in the inbox so you can follow up.
## Next
The chat inbox for the same agent.
Schedule and run outbound voice campaigns.