The SlyPilot Assistant: Say It Your Way

When you say (or type) something SlyPilot has no command for, the SlyPilot Assistant asks your own AI model what you meant and works out the steps: voice commands, window actions on your open programs, modes and automations. Anything more than one simple command is shown to you before it runs.

It is part of the free SlyPilot. It uses the model you chose in AI Actions Setup: one on your PC, your own server, Ollama or OpenRouter. It is off until you switch it on (tray > Voice Commands > SlyPilot Assistant (AI)).

  "Put Chrome and Code side by side on desktop 2 and start a 25 minute timer"
                                   │   your words + the names of commands, programs, desktops, modes
                                   ▼
                          your AI model plans
                                   │
   ┌──────────────────── Run this plan? ─────────────────────┐
   │ ☑ 1. move chrome to desktop 2     ☑ 4. snap code to right │
   │ ☑ 2. snap chrome to left          ☑ 5. "start a 25 minute │
   │ ☑ 3. move code to desktop 2               timer"           │
   │                               [ Cancel ]  [ Run ]          │
   └────────────────────────────────────────────────────────────┘
       or say "run the plan" / "cancel the plan"

✨ At a Glance

🎙️ Talk to It

Press Ctrl+Alt+Win+Space (or say "hey assistant"), hear the chime, and ask in your own words. SlyPilot records until you stop talking, turns it into text with the speech engine on your PC, and the assistant answers out loud (the voice you chose for reading aloud) or does what you asked. When it has finished speaking it listens a few seconds more, so "and in Spanish?" works; stay quiet or say "never mind" to end. While it listens, your words cannot start a voice command. It needs the assistant switched on, an AI model and the AI speech engine.

What goes to your model

• Always: your words, the names of SlyPilot's commands, and the names of the programs that have windows open (such as chrome.exe), of your desktops, modes, automations and workspaces. Never window titles, never the screen. Protected programs (password managers, private browser windows, your own list) are left out.
• Only when you ask about the screen: the text of the window you were using (read on your PC with Windows OCR, at most 6000 characters). If your model is outside your PC, a card asks first: No, Only this time or Always for this service.
• The words and the window text are never written to SlyPilot's log.

What it can and cannot do

• Can: any of SlyPilot's 550+ voice commands, window actions on a program you name (focus, minimize, maximize, restore, snap to a zone, move to a desktop or monitor, keep on top, pin), start or end a mode, run or make your automations, make a mode.
• Cannot: press keys, type, click, close windows, or start programs that are not in your apps list. Window actions follow the same rules as AI agents: protected programs and SlyPilot's own windows are refused.
• Instant for the everyday sentences: "put Chrome on the left", "send Spotify to desktop 3", "keep Notepad on top", "switch to Outlook", "meeting mode for half an hour", "I'm done with the meeting", and several of them joined by "and", are worked out on your PC without asking the model: they happen at once and nothing is sent. Anything else goes to your model.
• Shows the plan when it has more than one step, touches a window, a mode or an automation, or includes a command that asks for a confirm (pressing Run is that confirm). Untick a step to leave it out. One plain command runs at once, and so does one instant window or mode step (it is exactly what you said). ([Assistant] ShowPlan = steps | always | never.)
• All or nothing: every step is checked before anything runs; if one step cannot be done, nothing runs and you are told why.

Next Steps

• Set up a model: AI Actions → (guide)
• What the assistant can trigger: Voice Commands →, Automations & Modes →
• Privacy: what is sent, and when