Hey Barside voice commands: talk to your computer, hands free

With Hey Barside voice commands, you talk to your computer the way you would talk to a colleague. Say “Hey Barside”, pause, then say what you need: “Add a task: call Jon on Friday,” or “Pin this note to the left edge of monitor 2.” Barside hears the greeting offline on your computer, turns your command into text with Whisper on your computer, and hands it to the Barside assistant, which does the work. Voice wake-up is off until you turn it on.

Barside is a free desktop app for Windows, Linux and Mac. Its AI assistant uses the AI you connect in Settings → AI, so a spoken command can reach your notes, tasks, browser tabs and apps.

How Hey Barside voice commands work

A voice command goes through three steps.

1. The greeting. While Barside runs, even in the tray, a listener waits for “Hi Barside”, “Hey Barside” or “Hello Barside” said on its own. On Windows, Windows’ offline speech recognition finds the greeting and Barside’s local Whisper checks it a second time. On Linux, a hidden Barside window cuts speech into short pieces, and Whisper decides whether one of them is the greeting. Barside for Mac listens the same way as on Linux.
2. Your command. Barside opens the assistant and shows Listening to your command. Speak normally. About two seconds of silence ends the command, ten seconds without speech cancels it, and a command stops after 45 seconds. Say “cancel”, “stop” or “never mind” to drop it, press Cancel, or press Use command to send it at once.
3. The action. Your words appear in the conversation, and the assistant plans the steps with your AI. Changes inside Barside, such as a new note or task, happen right away and show in the activity log. Anything that would type or click in another app shows first, with the window, the control and the change, and waits for Approve this action.

How to turn on voice wake-up

1. Open Settings. Under AI, find Vision, memory & voice wake-up and press Assistant settings. The gear inside the assistant opens the same page.
2. Press Install speech if you have not yet. Whisper is a one-time download of about 150 MB and needs no AI key.
3. On Windows, make sure the English (US) Windows speech pack is installed, because Windows’ recognizer listens for the greeting. Linux needs no speech pack, and neither does Barside for Mac.
4. Under Voice wake-up, tick Listen for Barside.
5. Choose a Wake phrase: Hi, hey, or hello Barside (recommended), or Also allow “Barside” on its own.
6. Choose a Recognition threshold: Strict for fewer accidental activations (recommended), or Balanced to wake more easily.
7. Press Test wake-up for 30 seconds. Wait until the status says listening, say “Hi Barside” and pause. The test uses the options on screen and never runs a command.
8. Press Save assistant settings.

Connect an AI in Settings → AI too: one click for Claude, ChatGPT, Gemini or OpenRouter. Without one, Barside still hears you, but the assistant can only tell you how to connect an AI.

Things you can say

Anything you can type to the assistant, you can say. These examples work well.

Notes and your desktop

• “Create a note called Project ideas under Personal and add these points…”
• “Rename this note to Launch plan,” or “Move it to the Studio workspace.”
• “Pin this note to the left edge of monitor 2.”
• “Minimize all Barside tabs.”

Tasks and reminders

• “Add a task: call Jon on Friday.” It lands in the workspace’s Tasks note and on the task board.
• “Remind me to send the Northwind invoice tomorrow at 9.” The assistant makes a note with a reminder.
• “Mark the task about the demo room as done.”

Browser and reading

• “Bookmark all pages in this browser window in a new note, grouped as Research.” The links land in the note’s bookmark bar.
• “Bookmark the website I’m on under a new workspace called Suppliers.”
• “Save this page to Reading.” Barside keeps a readable copy in your read-it-later inbox.

Calls and screenshots

• “Transcribe this call into a new note in my Clients workspace.” See call transcription.
• “Create a note and insert a screenshot of my browser.”

Your day and your library

• “What needs my attention today?” brings up your daily briefing.
• “Find the tool I saved that handles LinkedIn outreach” asks AI search and shows the sources.

Other apps

• “Find my open email app and prepare a draft of the recap.” Every keystroke or click in the email app waits for your approval, you type passwords yourself, and nothing is sent without you.

On Windows the assistant reads and presses the controls of accessible apps. On Linux it can list, focus and minimize windows and press keys, but it cannot read the controls inside other apps, so commands about Barside itself work best there. In Barside for Mac, it reads and presses controls through System Events, once you give Barside the Accessibility permission.

Tune the greeting and the threshold

Strict means fewer accidental activations and is the recommended choice. Balanced is easier to wake, which can help when a laptop microphone hears you faintly. The threshold applies to Windows’ first check; on Linux, and in Barside for Mac, Whisper alone decides. Allowing “Barside” on its own is quicker to say, and on Windows Barside asks for more certainty before it accepts the single word.

The test in Assistant settings tells you what went wrong:

• The level bar should move when you speak, next to your microphone’s name. If it stays still, the microphone is off, muted or not allowed.
• “The microphone received sound, but no greeting was verified” means the sound arrived. Try Balanced, say “Hi Barside” and pause.
• “Possible greeting was too uncertain” means Barside nearly heard you. Try Balanced, or move closer to the microphone.

A headset keeps nearby voices and speakers away from the microphone, which means fewer accidental wakes.

If wake-up shows an error

The status line under Voice wake-up names the problem, and each message says what to do:

• “Install local speech in Assistant settings first.” Press Install speech.
• “The microphone listener did not start.” Check that Barside may use the microphone, in Windows privacy settings, your Linux system settings or, in Barside for Mac, System Settings → Privacy & Security → Microphone, then turn wake-up off and on.
• “Barside can’t use the microphone.” On Linux, allow the microphone in your system’s privacy settings, then turn wake-up off and on.
• “Wake listener stopped. Toggle wake-up off and on to retry.” Untick Listen for Barside, save, tick it again and save.

Every error also has a Send to support button. One click sends Barside’s support a report that never contains your notes, their titles or your files.

When it listens, and when it does not

• It is off until you turn it on. Restoring a backup keeps it off on that computer until you turn it on again.
• It listens while Barside runs, including in the tray. Closing or minimizing the window keeps it listening; quitting Barside stops it.
• It pauses during assistant tasks and recordings, while the screen is locked, and while the computer sleeps. After a command it takes a short break before it listens again.
• The tray menu shows the wake-up state and has Turn off voice wake-up. The assistant’s footer shows the state too.

Windows, Linux and Mac

On Linux, and in Barside for Mac, listening uses more power while people talk nearby, because Whisper checks each short piece of speech.

Speak in your language

The greeting is always English. Your command can be in another language. In Assistant settings, Speech language has English built in, and Install more languages · 141 MB adds automatic detection and 34 more languages, all turned into text on your computer. The same setting is used for call transcripts.

Privacy

• The listener works offline. Background audio is discarded. Only a short piece that might be the greeting is turned into text on your computer, and that audio is deleted.
• Your spoken command is turned into text on your computer. Its audio is not uploaded, and it is not kept.
• The command text, and the results of the tools the assistant uses, go to the AI company you connected, whose data policy applies.
• Wake-up is not speaker identification. It cannot tell your voice from someone else’s, and it cannot promise zero false wakes, especially when someone says the exact greeting.

Read how recordings and voice wake-up work and how the AI you connect uses your data.

How it compares

Prefer the keyboard? The search bar (Alt+Shift+Space in any app) does the same through Quick AI: type a request, press Tab, and the assistant answers there and asks before it acts. Ctrl+Shift+A opens the assistant, where the microphone button takes a spoken command without the greeting. If you are replacing a launcher, read the Raycast alternative for Windows. For dictation and meetings, see the best transcription apps.

Works well with