Getting started / Permissions · 3 min

Time to ship · about 3 minutes

Permissions.

Aura asks for microphone access, optional Accessibility for approved Mac control, and one selected project folder. Nothing else. Live voice frames go only through the selected provider route.

Each permission serves a specific function. Aura binds to a concrete project folder instead of broad personal folders so the working area stays small. What reaches the model is a digest of context, not raw source.

01.

Microphone — always required.

Live voice cannot function without the microphone. macOS will prompt on first launch. Grant it in the dialog, or navigate to System Settings → Privacy & Security → Microphone and enable Aura there. Local voice activity runs on your Mac; during live voice and streaming STT, encoded audio frames are sent through the selected provider route and are not stored by the Worker.

Local voice processing

The microphone captures audio for the active call. Project context is filtered before it leaves the device, while live voice/STT frames go to the model provider through the Worker. See the privacy policy for the full data flow.

02.

Accessibility — approved Mac control.

The Swift app does not currently register a system-wide shortcut. Accessibility is only needed when Aura performs approved direct Mac actions, such as pressing keys in another app. If you decline, voice and project setup still work; direct Mac control remains unavailable until you grant it at System Settings → Privacy & Security → Accessibility.

03.

File access — one project folder.

Aura reads project context from the folder you choose during setup or store in project_root. Broad personal folders are rejected by the project-selection path. The worker receives the current project boundary and still needs explicit approval before edit-capable work starts.

04.

What reaches the model.

Only a digest reaches the model proxy — not raw source, not file contents, not stack traces. Secrets, file paths, and line numbers are filtered before speech output and before any upstream call. For the complete breakdown of what is sent and what stays local, read the privacy policy.