First Run and Onboarding

Onboarding Overview

Onboarding has four steps:

1. Welcome
2. Permissions
3. Local transcription setup
4. Feature tour

If you installed from the DMG, you can complete onboarding without entering paths or downloading extra files. The bundled whisper.cpp runtime and Small English model are detected automatically.

Permissions Step

Steno shows the status of three macOS permissions:

• Microphone — required to record

• Accessibility — required for direct typing and accessibility insert

• Input Monitoring — required for global hotkeys to work while other apps are focused

• Continue requires Microphone permission, but Set up later lets you proceed without it. Recording remains unavailable until access is granted.

• Accessibility and Input Monitoring can be skipped, but related features will be limited until you grant them later.

• When you switch back to Steno after toggling a permission in System Settings, the on-screen status refreshes.

Engine Step

What you see depends on how you installed Steno.

From the DMG (bundled runtime detected)

• The step shows a "Based on your Mac" card with your detected hardware tier and a recommended model.
• The Small model is already included, so you can continue with it.
• If your Mac qualifies for Medium or Large V3 Turbo, you can download it inline before continuing. Steno chooses the recommendation from a built-in table of Apple silicon chip classes and unified memory.
• Each download is checked against the published file before it is installed. If a download fails, the reason appears next to the Download button.

From a source build (no bundled runtime)

• You confirm two paths:
  • whisper-cli binary path
  • whisper model file path (for example ggml-small.en.bin)
• Steno checks that both files exist. Continue requires valid paths; Set up later lets you finish onboarding and configure them later.

For better silence and background-noise suppression, the engine step also offers the optional Silero VAD model. VAD is enabled by default in the downloadable build because the model is bundled. Source builds need to download the VAD model and either point Steno at it directly or place it next to the selected Whisper model.

Hardware-Aware Recommendation

Steno recommends a model based on your chip and memory. A Validated label means that exact hardware and model combination has a recorded test result. Your recordings may give different results, so start with Small and compare optional models using speech typical of your work.

Finish and Verify

When you click Open Steno, Steno saves:

• updated whisper/model paths (if you changed them)
• the onboarding completion flag

After onboarding:

1. Hold Option and dictate a sentence into a supported text field.
2. Toggle hands-free with your configured function key.
3. Open the History tab and confirm a transcript appears with Inserted status.
4. Open Settings → Speech model and run Test Setup. It transcribes a short test clip and reports each step.

If anything fails, see Troubleshooting.