Docs

Launcher

The launcher is the main Agentastic interface for chat, tools, screenshots, and dictation.

Open The Launcher

Use either flow:

  • Global shortcut: ⌘Enter by default (ai_assist_shortcut)
  • Text selection popup: select text, then click the popup button

Text selection popup is controlled by show_agentastic_shortcut in config.

Window Modes

The launcher supports two modes:

  • Overlay mode: borderless floating overlay
  • Window mode: standard macOS window

Switch modes with ⌥⌘O.

Launch flags:

  • --window forces window mode
  • --overlay forces overlay mode

Default mode comes from default_launcher_window_mode ("window" fallback when unset).

Core In-Launcher Shortcuts

  • Enter: send message (when input is focused)
  • Esc: cancel in-flight request; also dismisses open selectors
  • ⌘Q: close launcher
  • ⌘M: minimize
  • ⌘N: new conversation
  • ⌘P: pin/unpin conversation
  • ⌘S: run save flow for current conversation
  • ⌘T: new session tab
  • ⌘W: close session tab
  • ⌥⌘W: close all session tabs
  • ⌥⌘↑ / ⌥⌘↓: previous/next session tab
  • ⌘F: toggle prompt-only minimized view
  • ⌘L: toggle embedded browser panel
  • ⌘1 (or ⌘I): agent mode
  • ⌘2: bundle mode
  • ⌘3: vision mode
  • ⌘5: search mode
  • ⌘6: imagine mode
  • ⌘7: kernel mode (agent mode only)

Notes:

  • ⌘4 is wired for thinking mode in the view, but default global screenshot shortcut is also ⌘4, so screenshot capture usually wins unless remapped.

Screenshot Selection

Screenshot selection can be triggered by:

  • Configured screen capture shortcut (screen_capture_shortcut, default ⌘4)
  • ⌘⇧S
  • Cmd+grave (`)

Behavior:

  • Requires screen recording permission
  • Requires macOS 14+
  • Drag-select region, minimum size is greater than 5x5
  • Esc cancels
  • Captured image is attached to the conversation
  • Captured image is copied to clipboard
  • Vision mode auto-enables after capture

Voice And Dictation

Global push-to-talk shortcuts:

  • Voice activation: hold Space (default)
  • Dictation app trigger: hold Fn (default)

Global hold threshold is 0.5s.

Inside the launcher, hold-to-talk starts after 1.0s hold.

Tab Context

Type /tab: to open the tab selector and include browser tabs in prompt context.

Current behavior:

  • Browsers: Chrome, Arc, Safari
  • Firefox is detected but tab enumeration is not supported
  • List refreshes every 5 seconds
  • Multi-select: ⌘-click, ⇧-click range, ⌘A, Space
  • Selected tabs are inserted as placeholders and content extraction runs asynchronously

Conversation Behavior

Conversations auto-save in the background. Pinning (⌘P) marks important conversations while still keeping normal auto-save behavior.