Mock Caller in Preview

Status: Current | Last reviewed: 2026-10-05

Mock Caller lets an AI play the customer using an ElevenLabs voice. It follows your scenario, listens to your agent's replies, and generates its next turn. You can nudge the caller during the conversation or ask them to wrap up. Use it to rehearse a demo or record the phone screen with both voices.

Configure the caller

  1. Open a voice agent's Preview, maximize it, and turn on Mock Caller in the header. This opens the settings modal. The pencil button inside the header's Mock Caller control reopens it. The same button is available in the Mock Caller widget; both are disabled during a call or recording.
  2. Write a Caller prompt describing the person, their reason for calling, facts they know, and when they should finish.
  3. Select an ElevenLabs integration and Caller voice. Search the voice list by name using the standard searchable picker, then select Play sample to audition it. Stop sample stops playback; closing settings or changing voices also stops it. The caller uses ElevenLabs v4.
  4. Expand Advanced to adjust call timing if needed; it is collapsed by default. Review Recording Options for silence trimming.
  5. Select Done and start the call from the phone.

The tenant needs an active OpenAI integration for the caller's decisions. The agent uses its own selected draft or published version, models and voices. Caller settings are saved separately for this agent, user and tenant in this browser. Repeating the setup preserves the scenario; generated wording can vary.

Minimizing Preview turns all switches off and stops a running mock call or recording. Maximize it again to start from the default Preview. Your saved caller prompt, voice, and timing settings remain available when you enable Mock Caller again.

For the local Consumer Cellular demo, enter +12025550147 as ani in Preview variables before calling. The demo uses fictional Margaret Wilson.

The caller prompt does not set the calling number. Leave ani blank only when you want to test an unknown caller; account lookup and a greeting by name need the test ANI above.

Transcripts follow speech time even if the agent's response appears before the caller's transcription is ready. A delayed caption is inserted into its correct position in both Thinking and the phone's Transcript view.

Delivery cues such as [coughs softly], [chuckles softly] and [warmly] can appear in the prompt. Ask for occasional cues and a natural apology rather than a cough in every turn.

Understand the timing

Expand Advanced to see the timing fields. Hover over or focus each field's help icon for its explanation.

Call time limit (seconds) is a maximum elapsed time, including connection setup. The default is 180 seconds. The caller is asked to wrap up about 25 seconds before the limit, or after eleven turns. The call stops at the limit; the setting does not force the conversation to last that long.

Wait for agent to finish (seconds) defaults to 2 seconds. The caller waits for this much quiet after the agent's speech or caption activity before planning its response. A longer wait gives the agent room to continue speaking. The caller's AI response generation and ElevenLabs speech generation happen after this wait, so it is not the total response latency.

Live calls also include connection, agent processing, speech detection and audio buffering time. Recording edits can shorten quiet gaps in the exported video without changing how the two agents take turns live.

Direct the call

For a more patient caller, an agent setup link ending in ?mockCallerPacing=patient applies a 3-second wait when Preview opens and adds guidance for short turns without long mid-sentence pauses. It preserves an existing scenario, selected voice and recording options. The link does not start a call. You can change these settings afterward in the usual modal.

With Mock Caller enabled, a compact panel appears inside the Preview canvas, to the left of the phone on wider layouts or above it when space is tighter. On wider layouts, it sits at the top of its column in both Preview and Recording. With Thinking open, the two side columns have equal widths and the phone is centered between them. The controls fill the left column, with compact spacing around their status, direction field, and buttons. The theater masks icon beside Mock Caller identifies the widget. With Show Thinking enabled, desktop layouts use three columns: caller controls, phone, and a compact Thinking panel that matches the phone's full height. Both widgets use the same rounded corners. Actions appear inline with the transcript; expand an action to see its inputs, result, and workflow details. On smaller screens, Thinking moves below the phone and controls. The header groups the Mock Caller toggle and pencil button inside one border. Once the call connects, use either control:

Directions apply when the caller plans its next turn. They go privately to the caller AI; only the resulting spoken reply enters the call. The controls sit inside the canvas but outside the phone/Thinking recording area, so private directions are excluded from downloads. The latest pending direction replaces an earlier one.

The phone's end-call button stops immediately. Closing Preview also stops the simulated call. One Mock Caller call can run per user and tenant at a time, for at most ten minutes.

Telephony is disabled while Mock Caller is on because IVR simulation is not supported yet. Turn Mock Caller off to enable Telephony.

Mock Caller is for direct voice-agent Preview. It is not available in Telephony, translation or copilot Preview. It simulates the caller only: your agent's tools and workflows still run with their configured data.

Record a demo

  1. Configure Mock Caller, expand Preview, and turn on Recording.
  2. Optionally enable Show Thinking to include the transcript and actions.
  3. In the Recording widget, select Test & Record and share the current browser tab.
  4. Use Wrap up when the story is complete, or let the caller finish.
  5. Wait for recording to finish, then use the same widget to select Download MP4, or Download trimmed MP4 if you enabled recording edits.

The Recording widget contains status, start/stop controls, and download buttons. It shares the left column with Mock Caller and stays outside the video. Enabling Recording keeps the same Preview header.

Caller audio streams directly from ElevenLabs into the agent. Preview monitors both original audio streams and uses them for recording. It does not request a microphone or run Krisp on simulated caller audio. Browser tab sharing captures the picture; tab/system audio is not the recording's audio source.

Shorten waiting time in the export

Before recording, open Mock Caller settings → Recording Options and enable Shorten long silences in downloads. This is off by default. Enabling it also automatically trims long quiet sections at the start and end, leaving a little space around the first and last speech. Its help icon explains this combined edit and how audio and video stay synchronized while live playback keeps its original timing.

The defaults match the pause timing used in the earlier Consumer Cellular demo edit. That recording went from 182.6 to 124.2 seconds, removing 58.4 seconds of quiet lead-in, tail and long gaps. It did not isolate the individual provider or network contributions to those gaps.

The export cuts the same intervals from the phone video and audio. Speech is not sped up or regenerated. Short pauses and overlaps remain; background sound can prevent a gap from being treated as silence. The on-screen call timer and Thinking timestamps may jump over removed time.

The selected download button shows a spinner and Preparing MP4… (or Preparing original MP4…) while the file is being prepared. Wait for your browser to start the download before selecting another version.

With editing enabled, Download original MP4 is also available. Settings are captured when recording starts, and the original stays available if processing fails. Silence trimming supports recordings up to 15 minutes within the normal 100 MB conversion limit. Keep Preview open until downloads finish.

Live microphone recordings continue to use their normal caller cleanup. See recording and Thinking for capture and download details.