How Runner works

Complete Setup, choose the microphone input you want Runner to use, and grant Microphone, Accessibility, and Input Monitoring access when macOS asks. Accessibility is required before Runner can send keyboard or pointer input. Input Monitoring is required for capture.

Open Commands, choose the pack you want to edit, then add a command name, one or more spoken phrases, an optional shortcut, and at least one action. Enable the pack and command when they are ready.

Run a command by starting listening and saying its phrase, using its shortcut, or selecting Run. Commands that collect a spoken value must start by voice.

Use Start listening, the global listening shortcut, or Hold to listen. Reversible commands enter a bounded serial queue, so Runner can keep listening while earlier commands complete.

A wake phrase can require a short phrase before spoken commands. An optional spoken stop phrase ends an active listening session; the Abort shortcut stops listening, capture, confirmation, queued actions, input, and active child processes.

Runner uses an English-only, on-device Moonshine speech model. Speech recognition and trigger matching happen locally.

Packs organize commands for any workflow or application. Selecting a pack does not make it the only active pack.

Runner normalizes phrases, matches whole words, prefers the longest valid trigger, and refuses ambiguous enabled conflicts. The same phrase can exist in disabled material while it is being edited, but it cannot become an ambiguous active trigger.

A single recognized buffer can contain multiple commands. Runner consumes matching spans from left to right and places accepted commands in a bounded queue.

Available macOS actions include keyboard input, typed text, pointer movement and clicks, files, folders, websites, application activation, waits, and reviewed POSIX shell or AppleScript steps. A macro combines steps in order.

Choose a fixed application when the action should always go to one target. For reversible keyboard, text, or pointer work, a command may instead use the application that was in front when voice or a global shortcut started it.

Runner checks permissions, target identity, and dispatch conditions before sending input. A sent result proves dispatch, not that another application completed the intended task.

Choose A number to fill selected repeat counts within the limits configured on the command. Choose Text until… to collect text until a chosen end word and send it through selected text actions.

Value collection is a separate visual stage after the exact trigger matches. It is not wildcard trigger matching. Say a value command on its own so Runner can identify which command should receive the value.

Invalid input, timeout, Stop, or Abort runs nothing. A command's Require confirmation setting still applies after a valid value is collected.

Start capture, demonstrate the keyboard and pointer steps, then stop. Application focus changes may be included, which allows one captured macro to cross application boundaries.

Review the native steps, remove or adjust anything unintended, choose exact targets where needed, and save the result as an editable macro.

Capture stops rather than recording a focused sensitive field or an input context Runner cannot verify. It also stops at the bounded event limit.

Enable Require confirmation on any command that should be reviewed immediately before it runs. The confirmation shows the command Runner is about to execute.

Imported packs do not silently trust scripts. Script actions remain disabled until they are reviewed, and every script has a bounded timeout and output.

Use Abort when the active work should stop immediately. Runner also stops protected work when required licensing or permissions are no longer valid.

Activity keeps a bounded local history of generic command results. Sound cues have one volume control and can be previewed in Setup.

Spoken confirmations are optional and speak only after the microphone has stopped. System notifications are also optional, with separate categories for commands, errors, listening changes, and application launches.

Feedback failure does not change the outcome of the command itself. External notifications use generic results rather than command text or captured input.

A trial or license is activated for this Mac through Runner's license service. The desktop saves a device-bound, server-signed receipt rather than a provider credential.

Runner verifies the receipt locally and can continue through its bounded offline period. It refreshes when a connection is available; an expired, revoked, suspended, invalid, or out-of-version entitlement fails closed.

Use the web account portal for billing and device management. The portal login and the desktop activation credential are intentionally separate.

On macOS, mutable Runner data lives in ~/Library/Application Support/Runner/. Exported packs or support reports stay wherever you choose to save them.

Reset removes known packs, commands, activity, and settings after confirmation. It preserves exports, macOS permission records, the app bundle, login-item ownership, and files Runner does not recognize.

The in-app uninstall flow can move the application to Trash and optionally move the exact Runner application-support folder to Trash. macOS privacy permissions remain under System Settings.

If listening cannot start, check the microphone permission and selected audio input. If an action cannot send input, check Accessibility. If capture cannot record, check Input Monitoring.

If a phrase does not run, confirm that both its command and pack are enabled and that no active phrase or reserved wake/stop phrase conflicts with it.

A support report can export build and system-health details without commands, triggers, scripts, captured input, activity text, audio, or file paths.