Hermes voice activation is the update that finally let me stop typing at my own agent. You say a wake word, Hermes opens a fresh session, listens for your command, and answers — completely hands-free. It works in the CLI, in the TUI and in the desktop app, and it works inside my Agent OS too. This is what it does, how to switch it on, and where it genuinely helps.
What Hermes Voice Activation Actually Does
The idea is simple enough that it is easy to underrate. Instead of opening a terminal, starting a session and typing a prompt, you say a phrase out loud. Hermes hears it, spins up a fresh session, records what you say next, and replies.
The default phrase is "Hey Hermes". You are not stuck with it. Change it in settings to whatever you like — "wake up Neo", "computer", or something nobody else in the room is likely to say by accident. The wake phrase is yours to set.
Then it just sits there, quietly, waiting. And when you want something, you ask for it the way you would ask a person in the next chair.
The Loop, In Plain English
Here is what is happening under the bonnet, because it matters for the trust question more than the tech question.
A lightweight detector watches your microphone. It is small, and critically, it runs on your device. When it hears your wake word, it pauses, opens a fresh session, records the command you speak, and hands that to Hermes. Hermes answers. Then the loop resets and it is ready for the next one.
Two details are worth pulling out. First, the whole thing runs entirely on your machine — the listening part is not being shipped somewhere else. Second, it is off by default. Nothing is happening on your mic until you deliberately switch it on.
That second point is the one people react to, so I will come back to it properly later.
📺 Watch: Hermes Agent: New Voice Activation Update is INSANE!
Turning It On: Three Slash Commands
Inside a Hermes session there are three commands, and that is the entire interface.
| Command | What it does |
|---|---|
| /wakeon | Switches voice activation on and starts listening for the wake phrase |
| /wakestatus | Tells you the current wake phrase, so you know what to say |
| /wakeoff | Switches it back off |
Behind the scenes these edit your config file — the setting is called wake word enabled. You do not have to touch the file yourself, but it is useful to know that is all that is changing. This is a flag, not a service you have signed up to.
If you do not have the feature yet, run hermes update. That is the whole install step.
Three Wake Word Engines
Hermes supports three engines for the detection itself: openWakeWord, Whisper and Porcupine. Different projects have different strengths — some are lighter, some are more forgiving of accents and background noise — and having the choice means you can swap if the one you start with is not catching your phrase reliably.
My advice: do not agonise over this on day one. Turn it on with the default, say the phrase a few times in your normal voice at your normal desk, and only go engine-shopping if it is genuinely mishearing you.
📺 Watch: Hermes Jarvis: New AI Voice Agent is INSANE!
The Toggle Inside Agent OS
Slash commands are fine, but there is friction in opening a terminal to tell your computer you want to talk to it. So inside my Agent OS I added a simple on and off toggle button.
Click it, Hermes starts listening. Click it again, it stops. No terminal, no command to remember, no config file. It is the same feature with the ceremony removed.
In the demo I did exactly what you would do on day one — the daft test questions. "Hey Hermes, what time is it?" Instant answer. "How are you?" Instant answer. Nothing clever, but that is the point: the gap between thinking of something and getting an answer collapses to about a second, and you are not touching a keyboard.
Once that gap closes, you start asking for things you would not have bothered to type.
Hermes Apollo: The One I Actually Reach For
Voice activation is the doorway. Hermes Apollo is the room I spend my time in.
Apollo is a custom real-time voice agent, available inside my Agent OS, and it is a different experience from wake-word-then-command. It is properly interactive. It replies with a voice, in real time, so it feels like a conversation rather than a query and a response. It shows you the full transcript as you go, so you can see exactly what it heard and what it said. And it can build things — you can preview what it has built in a gallery.
That last part is what shifts it from novelty to tool. Talking to a computer is fun for a day. Talking to a computer that produces something you can then look at is a working method.
If you only try one of the two, try Apollo.
📺 Watch: Hermes Voice Agents: Automate ANYTHING!
Which Model You Run Behind The Voice
Voice activation is the interface, not the intelligence. What actually decides whether the answers are any good is the model sitting behind it — and that is a choice, not a fixed part of the feature.
This is where my Goldie Bench testing comes in. That is my own hands-on comparison work, run on my machine on my kind of tasks, not an objective league table — but it is how I decide which model I put behind a voice agent I am going to be talking to all day. A model that reads well in a long written answer is not automatically the one you want speaking to you in short bursts. Test it the way you will use it.
If you want hands-free AI agents making you money, check out the AI Profit Boardroom — Agent OS + Hermes Apollo inside. → Get the hands-free agent setup
The Three Objections, Answered Honestly
Every time I show this, the same three things come up. They deserve straight answers.
"So it is always listening." No. You toggle it on, and when it is on, the on-device detector is watching for one phrase — not transcribing your life. It is off by default, it runs on your machine, and switching it off is one click or one command. If you do not want it on during a call, turn it off during the call.
"This will burn through tokens." Fair worry, especially if you imagine yourself chatting to it constantly. There are token-reduction options, and there are free models you can run behind it. The cost of voice activation is not fixed — it depends entirely on what you point it at.
"I do not have an afternoon to set this up." You do not need one. It is one slash command, or one toggle in Agent OS. That genuinely is the setup.
Frequently Asked Questions
Is Hermes voice activation on by default?
No. It is off by default. You turn it on with /wakeon in a Hermes session, or with the toggle in Agent OS.
Can I change the wake phrase?
Yes. The default is "Hey Hermes", but you can set it to anything in settings — "wake up Neo", "computer", whatever suits you.
Where does the listening happen?
Entirely on your machine. The detector that watches for the wake word is lightweight and on-device.
Which wake word engines are supported?
Three: openWakeWord, Whisper and Porcupine.
How do I check what my wake phrase currently is?
Run /wakestatus inside a Hermes session and it will tell you.
Does it work outside the terminal?
Yes. It works in the CLI, the TUI and the desktop app, and inside Agent OS with a toggle button.
How do I get the feature?
Run hermes update.
What is the difference between this and Hermes Apollo?
Voice activation wakes a fresh session on a phrase and answers your command. Apollo is a real-time voice agent that talks back in real time, shows the full transcript, and can build things you preview in a gallery. Apollo is the more interactive of the two, and the one I prefer.
The Bottom Line
Hermes voice activation removes the last bit of friction between having a thought and having an agent act on it. Say the phrase, ask the thing, get the answer — on-device, off by default, and switchable with one command or one click.
Run hermes update, type /wakeon, and say "Hey Hermes" once. If you like the feel of it, go straight to Hermes Apollo inside Agent OS, where the conversation is real-time and the agent can actually build what you are describing. That is the version that changed how I work, not just how I type.
