Yes, but local transcription is only one part of a cloud-free meeting workflow. Audio capture, speech recognition, summaries, storage, and connected features can follow different data paths. Check each one before assuming that a local transcript means the entire meeting stays on your device.
Anarlog stores its canonical meeting record in local SQLite. You can keep transcription and summaries local too, with a supported setup. Choosing a hosted model or another connected feature changes that boundary.
Local storage and local processing are different
A cloud transcription service receives audio to turn speech into text. Saving its resulting transcript locally does not undo that upload. Likewise, a local speech model can create a transcript that a cloud language model later receives to generate a summary.
Anarlog keeps those choices separate in Settings → Transcription and Settings → Intelligence. Changing one does not change the other. The models and providers documentation explains their inputs and requirements.
| Setup | Audio processing | Summary processing | What leaves the device for AI processing |
|---|---|---|---|
| On-device transcription and local Intelligence | Local | Local | Meeting content does not need a hosted AI provider |
| On-device transcription and hosted Intelligence | Local | Hosted | Text and context needed for the summary request |
| Hosted transcription and local Intelligence | Hosted | Local | Audio needed for transcription |
| Hosted transcription and hosted Intelligence | Hosted | Hosted | Audio for transcription, then text and context for summaries |
Using your own API key gives you a choice of provider; it does not make that provider local. Retention and training policies depend on the service and agreement you use.
Check your hardware and language first
Anarlog's built-in on-device transcription options require a supported Apple Silicon Mac. Apple Speech has additional macOS 26 and system-language requirements. An Intel Mac, Windows installer, or Linux installer does not establish that these built-in models are available there.
Language coverage also depends on the selected speech model. The app's overall language support is not a promise that every local model handles the same languages. Check the current on-device model requirements before downloading one.
Local and cloud accuracy or speed depend on the model, hardware, language, and recording. A cloud service is not automatically the faster or more accurate choice. Try a representative sample before relying on either setup.
Prepare both AI stages before disconnecting
Choose and download your speech model, then configure a local Intelligence provider and download its language model. A local server must remain available to Anarlog during summaries and chat.
For the actual setup and a short disconnected test, follow the offline meeting transcription guide. It covers audio permissions, model preparation, and checking the transcript and summary separately. The official offline checklist explains what can work without a network.
Check features beyond the models
Local AI selections do not prevent other features from sending data elsewhere:
- CloudSync can replicate encrypted notes to your other devices.
- Sharing uploads the generated summary and any audio you explicitly include.
- Cloud API & Connectors uploads a separate server-readable copy when enabled.
- Chat's web search sends the search query to a hosted endpoint, even with local models.
Avoid those features for a meeting that must remain on one device. Review usage analytics and error-reporting preferences separately; they are different from meeting-content processing. The data and privacy guide describes these boundaries.
Can you transcribe without retaining the audio?
Transcription and permanent audio retention are separate choices. Anarlog offers Don't save under Settings → Meetings → Audio file retention, but capture still uses temporary audio for recovery. Once that audio is deleted, an incomplete transcript cannot be repaired from it. Choose retention deliberately rather than treating transcription as a guarantee that no audio file ever existed.
If you are comparing broader deployment choices, the local AI meeting notes comparison separates desktop local processing from hosted services. A configurable desktop app is not proof of a complete self-hosted storage service or automatic compliance.