Help & FAQ

Click the record button in the main window, or press ⌘R (or Space when the WaveLog window is focused). WaveLog will immediately begin capturing audio and transcribing in real time. Click the button again (or press ⌘R) to stop.

Open Settings → Recording and select your preferred microphone or system audio source. You can also enable system audio capture to record meetings from video-call apps like Zoom or Teams.

Yes. Drag and drop an audio file (MP3, M4A, WAV, etc.) onto the WaveLog window, or use File → Import Audio. The file will be transcribed locally using the same on-device ASR engine you selected for live recording (Whisper, Qwen3-ASR, or SherpaASR).

Never. All transcription is performed on-device using your chosen local engine (Whisper, Qwen3-ASR, or SherpaASR) with Metal GPU acceleration. Your audio recordings and transcripts stay entirely on your Mac and are never uploaded to any server.

Whisper large-v3-turbo supports 99 languages with automatic detection — just start recording. Qwen3-ASR and SherpaASR have different language coverage. You can pin a specific language or switch engines in Settings → Transcription.

Yes. Transcription, recording, speaker diarization, and playback all work fully offline. Internet is only needed for account login, license validation, and optional cloud integrations.

WaveLog supports three local ASR engines — Whisper large-v3-turbo (99 languages, highest accuracy), Qwen3-ASR (Apple-Silicon-tuned low latency), and SherpaASR (lightweight SenseVoice). Choose per session in Settings → Transcription. Accuracy may decrease with heavy background noise, strong accents, or highly technical vocabulary.

WaveLog uses on-device ONNX models to create voice embeddings and automatically segment the transcript by speaker. Two models are available in Settings → Analysis: ECAPA-TDNN (fast, lower resource) and Sherpa-pyannote (higher accuracy for overlapping speech). Speakers are labelled Speaker 1, Speaker 2, etc.

Yes. Click any speaker label in the transcript to rename it. The name change applies to all segments from that speaker across the entire session.

Yes. Open the Tools panel (Tools button, top-right of a session) and choose Speaker tools. You can merge two speakers into one using the dropdown selectors, or split a single speaker track by clicking Split — drag across speech bubbles in the Messenger view to select the segments you want to reassign, then confirm to create a new speaker from them. You can also re-run speaker analysis at any time from the same panel.

WaveLog can distinguish up to 8 speakers per session. For best results, make sure speakers have distinct voices and avoid cross-talking where possible.

Open any completed session and click the AI Summary tab. WaveLog generates meeting summaries, action items, and key points in three modes. Configure your AI provider and model in Settings → Analysis.

WaveLog connects to the AI CLIs you've already installed and signed into on your Mac — including Claude CLI (Anthropic), Antigravity CLI (Google), and Codex CLI (OpenAI). Pick the provider and model in Settings → Analysis or the AI Summary tab. Authentication is handled by each CLI's own app, so WaveLog never stores or forwards your API keys.

The output language follows your Settings → Analysis → LLM Settings preference. By default it matches the transcript language. You can override it to always produce summaries in a specific language regardless of the meeting language.

WaveLog collects the meeting transcript and a system prompt, then calls each AI CLI exactly once in one-shot (non-interactive) mode and receives the result. • Claude CLI: claude --print --output-format json --model <model> --system-prompt <system> (user prompt is passed via stdin) • Antigravity CLI (agy): agy -p "<system + user prompt>" (passed as a positional argument) • OpenAI Codex CLI: codex exec --json --skip-git-repo-check --model <model> - (combined prompt is passed via stdin) All three CLIs run on your Mac under your own credentials and never pass through WaveLog's servers.

It depends on your provider and authentication method. • Antigravity (agy) — Included in your Google Antigravity account at no extra charge. No separate pay-as-you-go policy currently. • OpenAI Codex — Included with a ChatGPT Plus ($20/month) or Pro subscription at no additional cost. • Claude — Starting June 15, 2026, Anthropic will deduct non-interactive calls such as claude --print from a separate "Agent SDK credit" pool (Pro $20 / Max 5x $100 / Max 20x $200 per month), distinct from interactive usage. After credits are exhausted, standard pay-as-you-go API rates apply. If the ANTHROPIC_API_KEY environment variable is set, pay-as-you-go billing applies regardless of plan or date. Charges are billed directly to you by each AI provider (Google / OpenAI / Anthropic). WaveLog's servers are not involved.

WaveLog supports exporting action items and summaries to: Jira (create issues), Notion (append to page), Confluence (create page), Linear (create issues), Slack (post message), and Google Docs (append to doc).

Go to Settings → Integrations and click Connect next to the service you want. You will be redirected to the service's OAuth login page. Once authorised, WaveLog stores your access token securely in the macOS Keychain.

Exports use OAuth 2.0 with tokens stored in the macOS Keychain. WaveLog sends only the text content you choose to export — never your raw audio or full transcript unless you explicitly select it.

Press ⌘R to toggle recording from anywhere in the app. Space also works when the main window is focused. (⌘+Space is the macOS Spotlight shortcut and is not used by WaveLog.)

Click any sentence or segment in the transcript view and playback will immediately jump to that timestamp. You can also use the waveform scrubber to seek manually.

Press ⌘/ at any time to open the keyboard shortcuts dialog, which lists every available shortcut in the app.

Open System Settings → Privacy & Security → Microphone and make sure WaveLog is toggled on. If it doesn't appear in the list, try quitting and reopening WaveLog. macOS sometimes requires the app to request permission before it appears.

The Whisper large-v3-turbo model is approximately 1.6 GB — typically 3-10 minutes on a fast connection. Qwen3-ASR and SherpaASR have separate downloads with different sizes. The download for each engine happens once; after that, transcription is fully offline.

Make sure the matching AI CLI (Claude CLI, Antigravity CLI, or Codex CLI) is installed on your Mac and you are signed in through its own app. After signing in, restart WaveLog so it can detect the available providers. If the issue persists, try re-installing the CLI and granting any requested system permissions.

This happens the first time WaveLog reads each stored token after an OS update or app re-install. Click 'Always Allow' to prevent the dialog from appearing repeatedly. If it appears more than twice for the same item, try disconnecting and reconnecting the integration in Settings.

Speaker diarization can be confused by echo or bleed between microphones. For best results, use a directional microphone and ensure only one person speaks at a time. You can always manually correct speaker labels by clicking them in the transcript.