Offline dictation for Mac

Before you start

• Use an Apple Silicon Mac running macOS 14.0 or later.
• Connect to the internet for the first model download and leave enough disk space for the model you choose.
• Allow microphone access for dictation, and Accessibility access when you want to insert text into other apps.

Set up local dictation

1. Open Settings → Speech Recognition. Choose a local ASR model and use Download and Enable. Wait for installation to finish and confirm that the local model is selected.
2. For a basic setup, turn off Timeout fallback under Fault tolerance settings to keep recognition on the selected local path.
3. If you only need dictation, keep AI enhancement off. If you want it, download a compatible local enhancement model and select Local MLX in AI Enhancement. Check that every other processing feature you enable also uses a local configuration.

With the required models downloaded and selected, you can dictate without an internet connection using your configured shortcut.

For prompts in an editor or terminal, see the developer dictation guide. If your work uses names or technical terms, learn about automatic memory.

Local AI enhancement is a separate choice

The ASR model turns speech into text. AI enhancement uses its own model to work on that text. Downloading an ASR model does not download an enhancement model or change a cloud enhancement setting.

Check the whole processing path, including fallback models. ASR has its own Timeout fallback setting. AI enhancement can try other configured, verified models, including cloud models; a local primary model alone is not a local-only guarantee. For a local-only setup, use local models throughout the enhancement configuration and configure memory separately. Cloud API and subscription models still need internet access.

Memory models and document import

If memory processing is enabled, open Memory settings and select an available local Memory model for offline use. Follow AI Enhancement requires AI Enhancement to be enabled and its selected model to be local for this setup. To keep AI enhancement off while using memory, select an available local memory model explicitly. Choosing local ASR does not change the memory model.

To add your own documents, open Memory in Settings and choose Import Memories. You choose the documents to import. This is separate from automatic dictation context and does not mean that attunz scans your project files automatically.

You can also choose cloud processing

You can use third-party cloud APIs with your own API key (BYOK), or supported third-party subscription accounts. Cloud processing requires internet access, and any provider fees are paid to that provider. These options are separate from Attunz Cloud, the optional Attunz service, which is available.

All Attunz software features are free forever with local models or BYOK (your own API key). Attunz charges no fees for these options. You pay any third-party model fees directly to the provider. The Attunz hosted cloud service uses manual usage-credit top-ups for optional cloud ASR and AI enhancement, with no automatic renewal at present. It is available. Pricing

What still needs internet access?

The first model download, downloading missing or replacement model files, update checks and downloads, and submitting feedback need internet access. Offline dictation does not mean that every networking feature in the app is disabled.

If local dictation is not ready

Check that the selected local model has finished installing and that its files are available. Missing or damaged files may need to be downloaded again while online. Also check the permissions needed for recording and text insertion. Model languages, resource use and recognition results vary by model and device; compatibility with every text field is not guaranteed.

Common questions