Live Telemetry
Aggregate product-health and usage signals from MacParakeet installs with telemetry enabled.
Every transcription runs on-device — Parakeet, Whisper, or Nemotron — across 90+ languages, anywhere you type.
The dashboard checks every five minutes for a snapshot cached for up to fifteen minutes, separates published releases from other builds, and never includes audio, transcript text, prompt bodies, file paths, IP addresses, emails, or persistent user IDs. Anyone can opt out.
Awaiting telemetry snapshot
Counts and health rates are unknown until data loads.
Loading telemetry data...
Last 24 Hours
Main panels show the GUI app surface. macparakeet-cli is tracked separately because one-shot commands do not map cleanly to app sessions.
Hourly Pulse
Sessions and events across the last 24 UTC hours
All Time
Cumulative Growth
Sessions, transcriptions, meetings and onboarding completions share the left axis. Dictations rides a separate right axis — same linear scale, but auto-fit to its own range (the absolute count is roughly 5–10× larger), so all five lines stay readable on one chart. Transcriptions and meetings are split so each capture mode is its own trend. Hover any day for the full cumulative + daily breakdown.
Daily Usage
One line per capture mode. Dictations rides the left axis; transcriptions (file & YouTube), meetings and onboarding completions share a separate right axis — same linear scale, auto-fit to their own range (absolute counts run 10–50× smaller), so all four lines stay readable on one chart. Meetings are split out from transcriptions rather than folded in. Milestones annotated.
Engines & Languages
Speech Engines
Which on-device engine & model actually transcribed each dictation & file —
— · last 24h
Inner ring: engine family · outer ring: specific model
Languages
Most-used languages beyond English · — · last 24h
Model Builds
The exact on-device build behind each transcription, last 24h · —
Warm-up by engine
Successful model loads in the last 24 hours. Typical is the median warm-up, not a download. Engines need 10 loads in the window.
Where You Dictate
Which kind of app dictated text lands in — bucketed on-device, the app name never leaves your Mac (all-time)
Performance
Typical vs tail
Last 24 hours on published GUI builds. Typical is the median session. Mean is pulled by a few long jobs. Speed is the exception: higher is faster.
Transcription Speed
How fast completed jobs process audio for real users (24h). Hero is the median, not the mean.
By speech engine
File and URL jobs only — the same 24h window as the hero, not dictation. Speed is higher-is-faster; processing is how long the job waited. Engines need 20 jobs in the window.
Hardware
Apple Silicon Chips
Chip distribution across active GUI app sessions (24h)
macOS Versions
Active GUI app macOS versions (24h)
Worldwide
Top Countries
GUI app sessions by country (24h)
Version Adoption
Active GUI app versions (24h)
Awaiting telemetry snapshot.
Every Country
All — countries with an active GUI app session · last 24h
CLI
CLI Activity
macparakeet-cli invocations (24h)
CLI Versions
app_ver reported by CLI invocations (24h)
Dictation
Dictation Activity
Events recorded in the last 24 hours · — starts
Starts and endings are counted separately and may cross the time window. See Operation Reliability for outcome rates.
Onboarding & Activation
Activation KPIs (30d GUI) · setup step views (24h)
Setup step views
How many users reached each onboarding step (24h)
How Dictation Starts
Where the trigger came from — hotkey, idle pill, or menu bar (24h)
Hold vs Persistent
Hold-to-talk vs toggle-on / toggle-off (24h)
Dictation Pace
Words-per-minute distribution (— · dictations ≥ 5s only · 24h)
Dictation Length
How long users hold the hotkey (24h)
Transcription
Transcription Sources
Where transcriptions come from (24h)
Export Formats
How users export their transcripts (24h)
Audio Length
How long the audio people transcribe is (24h)
Diarization
Speaker detection adoption and per-job speaker counts (24h)
Meetings & AI
Meeting Recording Labs
Mic + system audio capture, transcribed locally. Trigger split, capture-track health, and notes use (24h)
Calendar Auto-Start Labs
Reminder and auto-start events recorded in the last 24 hours. These counts may come from different meetings and are not a conversion funnel.
LLM Feature Mix
Where users send LLM calls — Ask, chat, knowledge cards, and rewrite, broken down by provider (24h). Latency is successful calls only.
Latency by feature
Successful calls only. Unknown is mostly older formatter builds until those clients die; ingest now remaps formatter_dictation and formatter_transcription.
Latency by provider
Same successful calls, grouped by where the request went.
Live Ask Prompt Use
Which curated Ask surfaces and labels are used during live meetings (24h). Prompt bodies and custom user text are never sent.
Reliability
Permissions
How often each macOS permission is prompted, granted, or denied (24h). Friction here gates dictation, paste, and meeting capture.
Estimated Crash-Free Sessions
Distinct crash incidents compared with observed GUI sessions (24h, versions with ≥5 sessions)
Operation Reliability
GUI app terminal outcome events with true failures separated from cancelled, empty, and unavailable states (24h)
Every Event
Event Breakdown
Top 20 GUI app event types (24h) · —
Failure Event Log
Explicit failure breadcrumbs, sanitized error buckets, and crash reports; normal terminal outcomes are excluded
What We Collect
MacParakeet sends anonymous, session-scoped telemetry. Event props are bounded product metadata: counts, durations, broad sources, outcomes, and error buckets. They never include audio content, transcription text, prompt bodies, file names, file paths, API keys, emails, or persistent user identifiers.
App Lifecycle
app_launchedApp openedapp_quitSession length on quitonboarding_stepSetup step reachedonboarding_completedSetup completed
Dictation
dictation_startedRecording starteddictation_completedDuration, words, engine, languagedictation_cancelledUser cancelleddictation_emptyNo speech detecteddictation_failedFailure bucketdictation_operationWide outcome rowdictation_insertStop to Cmd+V posted, excluding overlay pausedictation_deletedHistory item removeddictation_undo_usedUndo window used
Transcription
transcription_startedSource and audio durationtranscription_completedSpeed, words, diarization, languagetranscription_cancelledStage cancelledtranscription_failedStage and failure buckettranscription_operationWide outcome rowtranscription_deletedLibrary item removedtranscription_favoritedFavorite toggled
Meeting Recording
meeting_recording_startedManual, hotkey, or calendar triggermeeting_recording_completedDuration and live word countmeeting_recording_cancelledUser discarded recordingmeeting_recording_failedFailure bucketmeeting_operationCapture plus transcription outcome
Speech Engine
model_download_startedModel install beganmodel_download_completedInstall durationmodel_download_failedInstall failure bucketmodel_loadedLocal model load timemodel_operationDownload, warm-up, repair, and cache outcomesspeech_engine_switch_operationEngine switch outcomes and blocked reasonsdiarization_startedSpeaker detection begandiarization_completedSpeaker count and durationdiarization_failedDiarization failure bucket
LLM Integration
llm_prompt_result_usedPrompt-library run succeededllm_prompt_result_failedPrompt-library failure bucketllm_chat_usedTranscript chat succeededllm_chat_failedChat failure bucketllm_transform_usedRewrite/transform succeededllm_transform_failedTransform failure bucketllm_formatter_usedDictation/transcript formatting succeededllm_formatter_failedFormatter failure bucketllm_operationProvider, feature, outcome, duration
Live Ask
ask_menu_openedQuick-prompt menu openedask_prompt_firedCurated label, source, and group onlychat_conversation_createdPersisted chat thread created
Feature Use
export_usedExport formatcopy_to_clipboardCopy sourcehistory_searchedSearch usedhistory_replayedAudio replay usedkeystroke_snippet_firedSnippet action firedfeedback_submittedFeedback categoryfeedback_operationFeedback delivery outcomeauto_save_operationAuto-save outcomecli_operationCLI command outcome
Vocabulary & Prompts
custom_word_addedDictionary entry addedcustom_word_deletedDictionary entry removedsnippet_addedSnippet createdsnippet_deletedSnippet removedprompt_createdPrompt template createdprompt_updatedPrompt template editedprompt_deletedPrompt template removed
Settings & Permissions
permission_promptedPermission prompt shownpermission_grantedPermission became availablepermission_deniedPermission unavailablesetting_changedSetting name only, not value contenthotkey_customizedHotkey changed, not key comboprocessing_mode_changedRaw/clean mode changedtelemetry_opted_outFinal event before telemetry stops
Calendar & Recovery
calendar_reminder_shownMeeting reminder surfacedcalendar_auto_start_triggeredCalendar auto-start acceptedcalendar_auto_start_cancelledAuto-start cancelledcalendar_auto_start_failedAuto-start failedmeeting_recovery_discoveredInterrupted sessions foundmeeting_recovery_startedRecovery beganmeeting_recovery_completedRecovery succeededmeeting_recovery_discardedRecovery intentionally discardedmeeting_recovery_failedRecovery failure bucket
Errors & Licensing
error_occurredSanitized app errorcrash_occurredCrash type and sanitized crash detailslicense_activatedLicense activation succeededlicense_activation_failedActivation failure buckettrial_startedTrial begantrial_expiredTrial expiredpurchase_startedPurchase flow beganrestore_attemptedRestore flow beganrestore_succeededRestore succeededrestore_failedRestore failure bucket
Per-event metadata
Every event has a small shared envelope plus optional event-specific props. Operation events also carry short-lived workflow IDs so one dictation, meeting, CLI command, or LLM call can be debugged as a single flow.
e.g. dictation_completed
e.g. 0.5.5
Major.minor only, e.g. 15.3
e.g. Apple M3 Pro
Language preference, e.g. en_US
From Cloudflare edge, not IP
Random UUID, reset each launch
e.g. word count, duration, outcome - never content
How to Opt Out
Don't want to be part of these stats? No hard feelings. Open MacParakeet, go to Settings, and toggle off telemetry. That's it — the app stops sending telemetry immediately after one final opt-out event.
Telemetry is opt-out (on by default) because it helps us fix bugs and understand usage. We respect your choice either way.