Call Details

Every call generates detailed data that helps you understand what happened during the conversation. The Call Details modal provides access to all information captured during the call.

Opening Call Details

Click on any call row in the Call History table, or click the View button to open the details modal.

The recording player described below (waveform, navigation, review, and tags) is a single shared component. It now also powers the recording modal in the advanced Calls view, so whichever screen you review calls from, the behavior and shortcuts documented here are identical.

The Call Details modal is organized in a three-column layout:

Panel	Content
⬅️ Left	Input Parameters (JSON)
📊 Center	Call Information
➡️ Right	Output Response (JSON)

Input Parameters

The left panel displays the input JSON that was sent to initiate the call:

{
  "phone": "+1234567890",
  "customer_name": "John Smith",
  "appointment_date": "2024-02-15",
  "custom_fields": {
    "order_id": "ORD-12345"
  }
}

Features

Copy Button - Copy the entire JSON to clipboard
Syntax Highlighting - Color-coded JSON for readability
Expandable - Scroll through large input payloads

Call Information

The center panel shows key metadata about the call:

Field	Description
Call ID	Unique identifier (monospace, copyable)
Status	Current status with colored badge
Success	✓ Yes or ✗ No indicator
Playbook	Agent name that handled the call
Duration	Call length in human-readable format
From	Origin phone number
To	Destination phone number
Email	Contact email (if provided)
Created Time	Timestamp when call was initiated
Tags	Labels applied to this call

Status Badges

Status	Color	Description
completed	🟢 Green	Call finished normally
in-progress	🔵 Blue	Call is currently active
failed	🔴 Red	Call encountered an error
transferred	🟣 Purple	Call was transferred
no-answer	⚪ Gray	Recipient didn’t answer

Analyze with Copilot

A new Analyze with Copilot button is available in the call information section that allows you to dive deeper into the conversation:

Open Copilot Analysis

Click the “Analyze with Copilot” button to navigate to the agent editor with this call’s context pre-loaded

Automatic Context Loading

The system automatically loads the call’s transcription and metadata into Copilot for immediate analysis

Ask Questions

You can now ask Copilot questions about this specific call, such as:

“What were the key topics discussed in this call?”
“How could the agent’s response be improved?”
“What patterns do you see in the customer’s behavior?”

The Copilot analysis maintains full context of the call, including transcript, input parameters, and output data for comprehensive insights.

LLM Model Usage

Meetzy runs on ElevenLabs’ LLM cascade, which can automatically fall back to a different model mid-call if the primary model is slow or unavailable. The LLM block in the Call Information panel gives you visibility into exactly which models handled the conversation.

Models Used

At the top of the block, you’ll see a list of every distinct model that participated in the call, in the order they were first used:

gpt-4o-mini → gpt-4o → claude-3-5-sonnet

If only one model appears, the cascade never needed to fall back — the primary model handled the entire call. If more than one model appears, ElevenLabs degraded to a secondary model at some point during the conversation.

For calls made before July 29, 2026, this list falls back to the single model_used field, since per-turn tracking wasn’t available yet. In that case, you’ll only see one model, even though a cascade may still have been active behind the scenes.

Per-Turn Breakdown (Cascade Detail)

Below the model list, click Show Detail to expand a collapsible, second-by-second breakdown of the cascade. This detail is collapsed by default to keep the modal compact. Each row in the expanded view represents one conversational turn and shows:

Column	Description
Time	Elapsed seconds (`t`) since the call started
Model(s)	Which model(s) generated output for that turn
Tokens	Output tokens produced by each model at that turn

If a turn used more than one model — for example, the primary model started responding but was interrupted by a fallback — both models are listed together, joined with +.

Locate the LLM block

Open any completed call’s details and find the LLM section in the center panel, below the core call metadata.

Review the models used

Check the summary list to quickly spot whether the cascade degraded during the call.

Expand the detail

Click Show Detail to reveal the per-turn table and identify exactly when the switch happened.

Collapse when done

Click Hide Detail to collapse the breakdown again and keep the modal easy to scan.

Cascade degradation is not necessarily an error — it means ElevenLabs automatically routed the conversation to keep latency low. Use this data to spot patterns (e.g., a model that degrades frequently) and adjust your agent configuration if needed.

Output Response

The right panel displays the output JSON containing all data captured during the call:

{
  "transcription": "Agent: Hello, this is Sarah from...",
  "summary": "Customer confirmed appointment for...",
  "sentiment": "positive",
  "output_params": {
    "appointment_confirmed": true,
    "preferred_time": "morning",
    "callback_requested": false
  },
  "evaluations": {
    "successful_greeting": true,
    "collected_requirements": true
  }
}

Output Contents

Transcription

Full conversation transcript between the agent and the caller, with speaker labels.

Summary

AI-generated summary of the call’s key points and outcomes.

Output Parameters

Structured data extracted from the conversation based on your Output Settings configuration.

Evaluations

Results of automated evaluations defined in your agent configuration.

Tool Calls

If the agent invoked any tools or webhooks, their requests and responses are included.

Webhook Information

If your agent has a webhook configured, the bottom of the right panel shows:

Field	Value
Method	POST, GET, etc.
URL	Webhook endpoint
Auth	Authorization header (masked)

The webhook section only appears if a webhook was configured and triggered for this call.

Recording Playback

If call recording is enabled, clicking the recording icon opens a dedicated Call Recording modal with a full-featured player on one side and a rich, per-turn transcript on the other.

This is the same recording modal used across Meetzy — including the advanced Calls view — so the waveform player, keyboard navigation, and review/tag editing described below all work identically no matter which list you opened it from.

Navigating Between Calls

The recording modal no longer traps you on a single call. When you open a recording from a filtered list (for example, from Call History), the modal knows about the calls immediately before and after the one you’re viewing, and lets you move between them without closing the modal and reopening a different row.

Previous / Next buttons

Two chevron buttons sit in the modal header, right next to the close button, so they’re always visible and never overlap the waveform or transcript below. Click ‹ to load the previous call’s recording and transcript, or › to load the next one, all without leaving the modal.

Keyboard shortcuts

Press ← to go to the previous call and → to go to the next call, without touching the mouse. Hovering the header buttons also shows these shortcuts as tooltips (“Previous call (←)” / “Next call (→)”).

The previous/next buttons are automatically disabled (dimmed, non-clickable) at either end of the list — for example, if you’re already viewing the first call, the “previous” button (and the ← shortcut) has no effect. The buttons only render at all when at least one adjacent call is available.

Arrow key navigation is automatically disabled while you’re typing in the review note or tag input fields inside the modal, so pressing ← or → to move the cursor in a text field never accidentally jumps to another call.

Switching calls triggers a short slide transition (the content pane animates left or right depending on the direction you navigated) so it’s clear a new recording has loaded, and reloads the waveform, audio, transcript, review status, and tags for the newly selected call.

Call ID Badge Next to the Transcript

To make it easy to reference a specific call while reviewing recordings — for example, when filing a bug report or pasting the ID into a support ticket — the Call ID is now shown as a small badge right above the transcript panel, next to the “Transcript” heading.

Always visible

You no longer need to scroll back up to the Call Information panel or return to the table to find the ID of the call you’re currently listening to — it’s shown directly next to the transcript.

One-click copy

Click the badge to copy the Call ID to your clipboard, exactly like the copy buttons elsewhere in the modal.

Because the badge reloads together with the rest of the panel, navigating to the previous/next call with the header buttons or arrow keys automatically updates it to the new call’s ID — you’ll never accidentally copy the wrong one.

Quick Verdict (Pending / OK / KO)

Reviewing a batch of recordings usually ends with the same decision for each call: was this a good call or not? Instead of closing the modal and marking the call from the table, you can now record that verdict right from the recording player. The verdict control is tri-state, not a simple on/off toggle:

State	Meaning
Pending	Default state — the call has not been reviewed yet, or the verdict was cleared
OK	The call was handled correctly by the agent
KO	The call needs attention (bad handling, missed data, wrong tone, etc.)

OK / KO control

A three-way control in the modal, next to the review checkbox, lets you mark the call currently loaded as OK or KO, with the active option visually highlighted.

Click again to deselect

Clicking the currently active option a second time clears it, returning the call to Pending — there’s no need to hunt for a separate “clear” or “undo” action.

The verdict is saved against the call you’re viewing and travels with it — navigating to the previous/next call loads that call’s own verdict state (Pending, OK, or KO); it never carries over from the call you just left.

Open a recording

Open any call’s recording from Call History or another filtered list. Its verdict starts at whatever state was last saved for it — commonly Pending if it hasn’t been reviewed yet.

Review the call

Listen to the audio and read the transcript to judge whether the agent handled the call well.

Set the verdict

Click OK or KO in the modal to record your judgment for this call. Click the same button again to reset the verdict back to Pending if you change your mind.

Move to the next call

Use the › button or the → key to jump straight to the next call in the list and repeat the process — no need to reopen the modal.

Combining the tri-state verdict with previous/next navigation and the arrow-key shortcuts turns batch call review into a fast, keyboard-friendly workflow: listen, judge, move on. The Pending state also makes it easy to filter for calls that still need a human decision.

Setting a verdict updates the call’s review status immediately. If you need to also leave a written explanation, use the review note field alongside the verdict control — the two are independent and both persist with the call.

Full review editing now lives inside the recording modal itself, so you don’t need to go back to the table row to finish reviewing a call.

Reviewed checkbox

A Reviewed checkbox marks the call as looked-at, independent of the OK/KO verdict. Unchecking it also clears the verdict back to Pending, keeping the two fields consistent — the same behavior as the checkbox in the calls table.

Review note

A text field lets you type a free-form note explaining your verdict (e.g. “Agent missed the callback time”). It saves automatically a moment after you stop typing, so there’s no separate save button to remember.

Follows the call

Just like the verdict, the Reviewed checkbox, note, and tags all reload for whichever call you navigate to with the previous/next controls — nothing is carried over between calls.

Because tags and review fields are edited on the same object the table row displays, updates made in the modal show up in the calls list the moment you close it — no manual refresh needed.

Waveform Player

Unlike a plain <audio> progress bar, the player renders a real, interactive waveform generated directly from the decoded audio — not a placeholder animation. When the recording loads, the audio is decoded via the Web Audio API and sampled into 100 amplitude bars, so the shape you see actually reflects the call’s audio (silences, loud segments, etc.).

Click-to-seek

Click anywhere on the waveform to jump to that point in the call. The played portion is highlighted in a different color as the call progresses.

+/-10s skip

Use the back/forward buttons to jump exactly 10 seconds, ideal for re-listening to a specific exchange without scrubbing manually.

Playback speed

Choose between 0.5x, 0.75x, 1x, 1.25x, 1.5x, and 2x to review calls faster or slow down a hard-to-understand segment.

Download

Download the original recording file for offline access, sharing, or permanent storage.

The waveform is computed once per call and cached for the session — reopening the same recording doesn’t require re-decoding the audio.

Enriched Transcript

Next to the player, the transcript is rendered turn-by-turn (not as a single wall of text), with the Call ID badge sitting right next to the “Transcript” heading. Each turn shows its elapsed time in the call, and agent turns carry two kinds of expandable detail sourced directly from the underlying ElevenLabs conversation: a response time breakdown and, when applicable, tool call details.

Turn Latency Breakdown

Every agent turn shows a “Replied in Xs” summary next to its message. Clicking it expands a per-turn latency breakdown that answers the question auditors actually care about: the agent took this long to answer — where did that time go? The breakdown splits the response time into a stacked bar with the following stages, shown in the order they occur in the pipeline:

Stage	Meaning
Understanding speech	Speech-to-text (ASR) latency — how long it took to transcribe what the caller said
Model thinking	Time to first byte from the LLM — how long the language model took to start generating a response
Generating voice	Text-to-speech (TTS) latency — how long it took to synthesize the agent’s audio
Other overhead	Any small residual not accounted for by the stages above, rather than being silently dropped

A separate “Waiting on purpose” segment may also appear below the stacked bar. This is not slowness — it’s the agent deliberately holding back before speaking (for example, waiting a few seconds before asking “are you still there?” or pausing to let the caller finish a thought). It’s called out separately, with an explanatory note, so a deliberate pause isn’t mistaken for a performance problem. The headline “Replied in Xs” always excludes this deliberate wait time, so it reflects only actual processing latency.

Each stage in the breakdown shows its duration and percentage of the total, along with the LLM model that generated the response for that turn. This turns a single opaque latency number into an actionable diagnosis:

A response dominated by Understanding speech may point to background noise or a caller who talks quickly.
A response dominated by Model thinking may point to a slow or overloaded LLM — worth checking against the LLM Model Usage section above to see if a cascade fallback happened on that same turn.
A response dominated by Generating voice may point to a longer-than-usual agent reply that takes more time to synthesize.
A visible Waiting on purpose segment simply confirms the agent chose to pause — no action needed.

Scan the turn

Each message bubble shows its elapsed time in the call, plus a “Replied in Xs” summary on agent turns.

Expand the latency breakdown

Click Replied in Xs to reveal the stacked bar showing how the response time splits across understanding speech, model thinking, generating voice, and any deliberate wait.

Check for deliberate waits

If a Waiting on purpose segment is shown, treat it as intentional agent behavior rather than latency to fix.

Expand tool calls

Click a 🔧 tool name badge to reveal the exact parameters sent to the tool and the result (or error) it returned.

Tool Call Details

If the agent invoked a tool or function during a turn, a 🔧 tool name badge appears next to that turn. Expanding it shows:

Parameters — the exact JSON payload sent to the tool for that call
Result — the value the tool returned, or the error details if the call failed

This lets you verify not just that a tool was called, but exactly what was sent and what came back, without leaving the recording view.

Color-Coded Tool Activity

To make tool activity scannable at a glance, the transcript now uses a consistent color code for anything related to tools — both in the badges next to a turn and in the expanded detail cards below it:

Color	Meaning
🟠 Amber	A tool call — the request the agent sent to a tool, with the tool name and parameters
🔵 Blue	A tool response — the result the tool returned for that call
🔴 Red	A tool error — the call failed, and the card shows the error details instead of a result

This color coding is applied consistently across the modal: the small 🔧 tool name badge next to a turn, and the expanded parameters/result card underneath it, both use the same amber/blue/red scheme. A quick glance down the transcript is enough to spot which turns involved tools and whether any of those calls failed.

This view mirrors what ElevenLabs’ own conversation view shows, since this per-turn detail (latency metrics, model attribution, tool payloads) isn’t persisted in Meetzy’s database — it’s fetched live from ElevenLabs when you open the recording.

If the underlying ElevenLabs conversation is no longer available (for example, an old or purged call), the modal automatically falls back to a plain transcript view — the message bubbles are still shown, but without latency breakdowns, model detail, or expandable tool call/result cards. A short notice appears above the transcript in this case. The same amber/blue/red color coding for tool calls, responses, and errors is preserved in this fallback mode wherever that information is still present in the raw transcript, so tool activity remains easy to spot even without the rich per-turn metrics.

Call recordings are stored for a limited time based on your plan. Download important recordings for permanent storage.

Enhanced Analytics Workflow

The new Copilot integration streamlines your analysis workflow:

Quick Analysis

Jump directly from call details to Copilot analysis without losing context

Contextual Insights

Get AI-powered insights based on the specific call’s transcription and data

Pattern Recognition

Identify trends and patterns across similar calls using Copilot’s analytical capabilities

Agent Improvement

Use call-specific analysis to refine agent prompts and behaviors

Copying Data

Copy Buttons

Copy Input - Copies the entire input JSON
Copy Output - Copies the entire output JSON
Copy Call ID - Copies just the call identifier

Feedback

When you copy data:

The button changes to “Copied!” with a checkmark
Reverts to normal after 2 seconds

Common Use Cases

Debugging

Review input params and output to diagnose why a call didn’t perform as expected. Use the color-coded tool cards in the transcript to quickly spot which tool call failed.

Quality Assurance

Listen to recordings, check transcriptions, and mark each call Pending/OK/KO — plus add a note and tags — to build a fast, keyboard-driven QA workflow entirely inside the recording modal.

Integration Verification

Verify that webhook data was sent correctly and contains expected values.

AI-Powered Analysis

Use Copilot to analyze call transcriptions and generate improvement recommendations.

Keyboard Shortcuts

Shortcut	Action
`Esc`	Close the modal
Click outside	Close the modal
`←`	In the recording modal: jump to the previous call
`→`	In the recording modal: jump to the next call

The ← and → shortcuts only apply while the Call Recording modal is open, and are automatically ignored while typing in a text field (such as the review note or tag input).

There is no dedicated shortcut for the verdict control, the Reviewed checkbox, or tags — use the on-screen controls in the modal. This keeps review edits a deliberate click/keystroke rather than something that could be triggered accidentally while navigating between calls with the arrow keys. Clicking an already-selected OK or KO button deselects it back to Pending.

Next Steps

Call History

Return to searching and filtering calls.

Copilot Analysis

Analyze calls with AI-powered insights.

Reporting

Generate reports from your call data.

​Call Details

​Opening Call Details

​Modal Layout

​Input Parameters

​Features

​Call Information

​Status Badges

​Analyze with Copilot

​LLM Model Usage

​Models Used

​Per-Turn Breakdown (Cascade Detail)

​Output Response

​Output Contents

​Webhook Information

​Recording Playback

​Navigating Between Calls

Previous / Next buttons

Keyboard shortcuts

​Call ID Badge Next to the Transcript

Always visible

One-click copy

​Quick Verdict (Pending / OK / KO)

OK / KO control

Click again to deselect

​Reviewed Status, Notes, and Tags — Right From the Modal

Reviewed checkbox

Review note

Tags

Follows the call

​Waveform Player

Click-to-seek

+/-10s skip

Playback speed

Download

​Enriched Transcript

​Turn Latency Breakdown

​Tool Call Details

​Color-Coded Tool Activity

​Enhanced Analytics Workflow

Quick Analysis

Contextual Insights

Pattern Recognition

Agent Improvement

​Copying Data

​Copy Buttons

​Feedback

​Common Use Cases

Debugging

Quality Assurance

Integration Verification

AI-Powered Analysis

​Keyboard Shortcuts

​Next Steps

Call History

Copilot Analysis

Reporting

Call Details

Opening Call Details

Modal Layout

Input Parameters

Features

Call Information

Status Badges

Analyze with Copilot

LLM Model Usage

Models Used

Per-Turn Breakdown (Cascade Detail)

Output Response

Output Contents

Webhook Information

Recording Playback

Navigating Between Calls

Call ID Badge Next to the Transcript

Quick Verdict (Pending / OK / KO)

Reviewed Status, Notes, and Tags — Right From the Modal

Waveform Player

Enriched Transcript

Turn Latency Breakdown

Tool Call Details

Color-Coded Tool Activity

Enhanced Analytics Workflow

Copying Data

Copy Buttons

Feedback

Common Use Cases

Keyboard Shortcuts

Next Steps