Hermes Apollo voice is the setup everyone saw in the launch demos — someone tells their computer “build me a snake game,” sips coffee, and ninety seconds later it exists. No keyboard, no clicks. I went hands-on with the voice stack in the video above, and this page is the complete, verified picture: what Apollo actually is (including a naming surprise the hype posts skip), the free local engines that make it cost nothing to run, every command and config key from the official docs, the full-duplex barge-in behaviour that makes it feel like a phone call — and the one mode that quietly bills you by the minute.
Key takeaways
- What it is: full voice in and out for Hermes Agent — CLI, Telegram, Discord (including live voice channels) and desktop. Naming truth: “Apollo” is the launch-coverage name — the official docs simply call it voice mode plus wake word.
- The free stack is genuinely free: local Faster-Whisper STT (~150MB model, downloads on first use, no API key) plus Edge TTS (default, no key, ~1s latency) — zero-cost voice on your own machine.
- It’s full duplex: barge in at any point — the mic stays live while it thinks AND while it speaks — say “stop” alone to end the session hands-free, and 3 seconds of silence auto-submits each turn.
- The paid trap to know: desktop GPT-Live mode bills $0.05 per minute of session time — idle counts. The default chained mode costs nothing beyond your model.
What Hermes Apollo voice actually is
First, the naming honesty most coverage skips: I checked both official docs pages on 26 September 2026, and the word “Apollo” appears in neither — Nous Research documents this as voice mode and its companion wake word feature. “Apollo” is the name the launch-week coverage (mine included) attached to the experience, and it stuck. Search for Apollo, configure “voice” — same thing, and this guide uses the docs’ real setting names throughout.
The experience itself splits into three parts. The voice: press Ctrl+B and talk — or go fully hands-free with the wake word (“hey hermes” by default; the Porcupine engine ships listening for “jarvis,” which tells you who built this). The brain: whatever model your Hermes runs — local for privacy, or a flagship like Claude Opus 5.5 for the hard jobs. The hands: the same Hermes tools you already use — open apps, drive the browser, write files — now triggered by speech, with replies spoken back.
And it’s not desktop-only: voice replies work in Telegram and Discord messages, and the bot will physically join a Discord voice channel with /voice join and hold a live conversation there. The wake word is the one piece that needs a local microphone — it runs in the CLI, TUI and desktop app, not the messaging gateway.
Hermes Apollo voice setup, engines and commands
Setup is one installer and two choices. Run hermes tools and add the voice extra (plus messaging for Discord/Telegram and tts-premium if you want ElevenLabs); system-wise you need PortAudio and ffmpeg, plus the Opus codec for Discord channels. Then pick your two engines:
| Engine choice | Options (per the official docs) | Cost / latency |
|---|---|---|
| Speech-to-text | local Faster-Whisper (~150MB, auto-downloads) · Groq Whisper · OpenAI Whisper · Mistral voxtral-mini-latest · xAI grok-stt | Local: free, no key. Groq: free tier with key. Others: paid |
| Text-to-speech | edge (default) · ElevenLabs · OpenAI TTS · NeuTTS (local) | Edge: free, ~1s. OpenAI: ~1.5s. ElevenLabs: ~2s, best quality. NeuTTS: free, hardware-dependent |
One platform footnote from the docs: local Faster-Whisper is excluded on native Windows ARM64 and Intel Macs — those targets use a cloud STT provider instead. Everyone else gets the fully-local, zero-key stack. The commands you’ll actually use:
- Ctrl+B — start/stop recording in the CLI (remappable via
voice.record_key) - /voice on · /voice off · /voice tts · /voice status — enable voice, disable it, speak ALL replies, check state
- /wake on — hands-free session start by wake phrase (
wake_word.enabled: truein config; phrase, sensitivity and provider all configurable) - /voice join · /voice leave — pull the bot into your current Discord voice channel, or out of it
Want this working in your business, not just bookmarked? Getting a voice-driven agent doing real, useful work in your business is exactly the kind of thing we build together inside the AI Profit Boardroom — 3,700+ members, four live calls a week, daily tutorials, plug-and-play templates and a 30-day roadmap so you ship instead of watch.
Prefer it mapped 1-on-1 first? Book a free strategy session and we’ll plan it for your exact situation.
The behaviour that makes it feel alive
The loop is what the demos are really showing off. Press Ctrl+B: an 880Hz beep confirms recording, a live level bar tracks you, and three seconds of silence auto-submits — two lower beeps, transcription, and the reply starts streaming sentence by sentence as the model generates, then the mic re-opens by itself. You press exactly one key per session, not one per message.
Better: it’s full duplex. The docs are explicit that the microphone stays live from the moment you finish speaking until the reply has fully played — so you can interrupt mid-generation (your interjection becomes the next message) or mid-speech (playback cuts, your words submit). Say “stop” on its own and the session ends — customisable and multilingual via voice.stop_phrases. Barge-in has its own tuning keys (barge_in_threshold_multiplier, a half-second grace window) for noisy rooms. This is the “no awkward Siri pauses” thing the launch posts raved about, and it’s real.
The desktop app adds a second architecture: the default chained mode (STT → your LLM → TTS, costs nothing beyond your model) and GPT-Live — a real-time full-duplex voice model that delegates tools and memory back to Hermes. GPT-Live is desktop-only and bills $0.05 per minute of session time, idle included — leave it open through a cup of tea and you’ve paid for the tea. My hands-on in the video sticks with chained mode, and honestly, for agent work it’s all you need. Pair it with Bot Screen and you can watch the hands work while you talk.
Limitations, straight from the docs (26 September 2026): noisy environments can still produce garbage transcripts despite the built-in hallucination filter; local Whisper is unavailable on native Windows ARM64 and Intel macOS; the wake word only runs where a local mic exists (not Telegram/Discord messaging) and one microphone serves one surface at a time; GPT-Live mode isn’t available in the CLI/TUI. And “Apollo” itself isn’t an official product name — if you’re searching the docs, look for voice mode and wake word.
The bottom line on hermes apollo voice
Hermes apollo voice is the rare demo that survives contact with the docs: a genuinely hands-free, full-duplex, interruptible voice agent you can run entirely locally for free — ~150MB of Whisper, default Edge TTS, one keypress or a wake phrase, and your agent’s full toolbox on the other end of the conversation. Just know the two honest footnotes: the name “Apollo” lives in the coverage rather than the docs, and GPT-Live’s per-minute meter is the one part of a free stack that isn’t. Set it up with my voice activation guide, then stop typing for an afternoon — it recalibrates what “using a computer” means.
FAQ: hermes apollo voice
What is Hermes Apollo voice?
It’s the voice stack for Hermes Agent, Nous Research’s open agent platform: speak to your agent and it replies aloud and acts — in the CLI, Telegram, Discord (including live voice channels) and the desktop app. “Apollo” is the name the launch coverage gave it.
Is Apollo an official Nous Research product name?
Not in the documentation — as of 26 September 2026 the docs call it voice mode and wake word, and “Apollo” appears on neither page. The name comes from launch-week coverage and community posts, mine included.
Is Hermes voice free to use?
The core stack is: local Faster-Whisper speech-to-text (~150MB, no API key) plus the default Edge text-to-speech (no key, ~1s latency). Paid options exist — ElevenLabs or OpenAI voices, and the desktop GPT-Live mode at $0.05 per minute including idle time.
What are the main voice commands?
Ctrl+B starts and stops recording; /voice on, /voice off, /voice tts and /voice status manage the mode; /wake on enables the wake word (“hey hermes” by default); /voice join and /voice leave move the bot in and out of Discord voice channels.
Can I interrupt it while it’s answering?
Yes — it’s full duplex. The mic stays live during both generation and playback, so speaking cuts it off and your words become the next message. Saying “stop” alone ends the voice session entirely, and both behaviours are configurable.
Does the wake word work everywhere?
It runs wherever a local microphone exists — CLI, TUI and desktop — with “hey hermes” as the default phrase. It doesn’t run in Telegram or Discord messaging (no mic), and only one microphone can serve one surface at a time.
Two ways I can help from here. If you want the community, the templates and the weekly momentum, join the AI Profit Boardroom — it’s where your hands-free agent setup gets built with 3,700+ members doing the same.
If you want a personal plan first, grab a free strategy session and bring your questions — no pitch-fest, just the roadmap.
About Julian Goldie
I’m Julian Goldie — SEO agency founder, best-selling author, and one of the most-watched AI SEO educators on YouTube with 394K+ subscribers. I’ve spent 10+ years in SEO and link building, hold a 100% Job Success Score on Upwork, and run a community of 75K+ members learning AI-powered SEO. I test everything on my own sites first — what you read here comes from those tests. Join the AI Profit Boardroom for the daily builds, or book a free strategy session to talk through yours.
Related reading
- Hermes Voice Activation: Hands-Free Agent Setup (2026)
- Hermes Bot Screen: Your Agent’s Second Monitor
- Hermes Claude: The Complete Stack Guide
Last updated September 2026. This page is a living guide to hermes apollo voice — the facts here move fast and I update it as they do.
