Context Voice widget, commands and “Hey Context”

Context Voice is a small orb that floats always on top of other windows. It lets you keep talking to Context Code while you work on something else: the main window can be minimized and the conversation stays alive.

The app itself describes it like this in Settings → Voice and reading → “Context Voice as a widget”:

A small orb, always on top of other windows, so you can keep talking to the app even with the main window minimized. Drag it anywhere; double-click returns to the main window.

Context Voice: the floating orb with the conversation status, always on top.

Context Voice: the floating orb with the conversation status, always on top.

Settings

Setting	Options
Show widget	Never · When the main window is minimized (default) · Whenever voice is active
Screen position	It remembers where you leave it. Reset position returns it to the center-right.
Character	The look of the widget: Orb, Ring or Pixel.

“When minimized” also counts closing the window to the system tray, which is when you need it most. And if the saved position ends up outside the connected screens — you unplugged the monitor where you’d left it — the widget returns to the center-right on its own instead of showing up where you can’t see it.

With “When the main window is minimized” the widget appears even when no conversation is running, with a “Talk to Context Code” button to start one right from there. It’s the difference between a shortcut and an ornament: you minimize the app, the orb stays, and you talk when you need to.

It works both with the real-time conversation and with the classic voice mode, and it uses the theme you have set in the application.

The three characters

Character	What it looks like
Orb	A sphere that breathes with the voice. The usual one.
Ring	Three thin rings spinning.
Pixel	A 7×7 grid that fills up with your voice.

The selector shows them live: you see them move before choosing.

“Try widget”

If you’re not sure it will show up when you need it, in Settings → Voice and reading there’s the Try widget button: it shows it for five seconds with a sample status, without opening a conversation, and writes the diagnosis underneath — whether the window was created, whether the system considers it visible to the system, what position it ended up in, whether that position falls inside the screen and how many monitors it detects.

It’s the answer to “I turned it on and I see nothing”: it tells you where it breaks.

What the orb does

The orb is a mirror of the conversation in the main window: it reflects its status through its shape and color — Connecting…, Listening, Thinking, Speaking, Paused, No voice or Error — and shows the caption of what is being said, in two lines. If the main window stops responding for a few seconds, it warns you with “Waiting for the app…”.

And it has controls of its own:

• Pause / Resume — mutes the microphone without hanging up.
• Show the main window — also by double-clicking the orb.
• End the conversation.
• Drag it anywhere on the screen.

Voice commands for the app

With “Voice commands for the app” — enabled by default in Settings → Voice and reading — the agent doesn’t just answer: it drives the application for you while you talk. The app itself gives the examples:

“Create a new chat”, “open project X”, “switch to model Y”, “read me the last answer”, “pause”, “hang up”, “show the window”.

These are the commands it understands, with phrases that work:

Command	Say it like this	What it does
Create a chat	“create a new chat”, “open a new conversation”, “start a chat called Invoices”	Creates a new chat in the current project and selects it. The voice conversation moves to the new chat on its own once you finish speaking.
Open a chat	“open the chat about X”, “go back to the conversation about Y”	Looks up an existing chat by name across all projects and opens it.
Switch model	“switch to model X”, “use Claude Sonnet”, “move over to Gemini”	Changes the current chat’s model; if you name the provider (anthropic, openai, gemini, ollama…), it switches that too.
Open a project	“open project X”, “switch to project Y”	Opens a project by name among the recent ones, or by its path on disk.
Read the last answer	“read me the last answer”, “repeat the last thing you said”, “what did you answer me”	Reads the assistant’s last answer in the current chat out loud.
Pause voice	“pause”, “hold on a moment”, “mute the mic”	Mutes the microphone without closing the conversation; resume from the widget or the window.
End voice	“end the conversation”, “hang up”, “bye, close the voice”	Says goodbye and closes the conversation.
Show the window	“show the window”, “open the app”, “bring me the application”	Brings the main window to the front and restores it if it was minimized.

Note

The commands don’t ask for approval: they don’t touch your files or run commands, they only drive the window itself. Everything that does touch the disk still goes through the usual approval modes.

Caution

If the switch is off and you ask for one of these things, the agent will tell you instead of insisting: commands for the app are disabled in Settings → Voice and reading.

“Hey Context”: have it listen without touching anything

With always-on listening you don’t even need to press the voice button: you say “Hey Context” out loud and the conversation opens on its own, with its greeting (“I’m listening”) and the widget in front of you.

It’s turned on in Settings → Voice and reading → “Always-on listening (“Hey Context”)”, and the app itself explains the deal:

Keeps the microphone open and starts the conversation when it hears “Hey Context”. Everything is recognised on this machine: the audio never leaves the computer and only the short bursts where someone speaks are transcribed. While it is on you will see an indicator next to the voice button.

It ships off, and for good reason

It is the only feature of the app that opens the microphone without you asking. That’s why it’s off out of the box and has to be turned on by hand, and why while it listens there’s a visible indicator next to the voice button that pulses with the microphone level: if it’s listening, you can tell.

How you say it

“Hey Context”, “Oye Context” and plain “Context” all work. You don’t need to pronounce it perfectly: recognition tolerates what a small model really transcribes from a one-second burst (“hey contex”, “oye contexto”…).

In exchange, it gets demanding when it should:

• If there’s noise, it asks for two detections within 3 seconds before opening anything.
• Plain “Context” — a word that comes up on its own in a work conversation — needs three.
• After triggering there’s an 8-second cooldown: it doesn’t fire twice in a row.
• It never listens while a conversation is open or while the agent is speaking.

Better to have to repeat it than to have the application start listening on its own in the middle of a meeting.

What it costs to keep it on

Nothing noticeable at rest: an energy detector watches the microphone and discards silence without transcribing anything. Only when someone speaks is that short burst sent to Whisper (tiny or base model), and that’s about four tenths of a second of CPU per burst.

Everything happens on your machine: the audio never leaves the computer and isn’t stored.

How it combines with Conversation mode

The widget is the floating face of the real-time conversation: you start the conversation from the chat first, and the orb appears according to the Show widget setting. Set it to Whenever voice is active and you’ll have it in front of you from the first second; with When the main window is minimized it only shows up when it’s really needed.