Hermes Apollo voice — officially just Voice Mode in the Hermes docs — turns Hermes Agent into something you talk to: press-and-speak from the CLI, live voice channels on Discord, voice-bubble replies on Telegram, eleven text-to-speech engines to answer you back, and a full-duplex desktop mode when you want zero buttons.
📺 Watch: GPT 6 Astra + Hermes Voice Agent is a GAMER CHANGER!
🔥 Get the Agent OS as a free bonus: AI Profit Boardroom members get the full Agent OS zip, prompt libraries, daily tutorials and weekly live coaching calls. → Get inside
Two names, one capability, and plenty of confusion in between. We covered the launch story in our earlier Hermes voice agent piece — this page is the documentation deep-dive. Every command, every engine, and every honest limit below comes from Nous Research's official Hermes docs (the Voice Mode and Voice & TTS pages, fetched 26 September 2026), plus where the feature sits in Julian's own stack.
Apollo Or Voice Mode? The Name, Settled
Search the official Hermes documentation and you will find a Voice Mode page and a Voice & TTS page. The word Apollo appears on neither. It is not the name Nous Research uses.
Apollo is what the community calls this capability — the label you hear in YouTube coverage and around Julian's Agent OS bundle, where voice control ships as part of the package. Same feature, two labels: Voice Mode is what is printed on the box, Apollo is the street name it picked up along the way.
So, settled before we go deeper: everything below describes Voice Mode exactly as the official Hermes docs define it. Apollo is a nickname, not a product name — treat any page claiming otherwise as guesswork.
Hermes Apollo Voice: How A Conversation Actually Works
Inside the Hermes CLI, /voice toggles voice mode. Want it explicit? /voice on and /voice off do exactly what they say, /voice tts toggles spoken replies, and /voice status shows the current state. Five commands, zero mystery.
The listening model is push-to-talk with auto-detection. Press Ctrl+B and talk. You never press anything to finish — the docs describe a two-stage silence detection algorithm that stops recording after 3.0 seconds of continuous silence, then Hermes responds. If Ctrl+B clashes with your terminal habits, the record key is configurable through the voice record key option in the Hermes config file at ~/.hermes/config.yaml.
Two details the docs make refreshingly clear. A wake word is not the default — wake-word listening exists as a separate feature, but out of the box Hermes listens only while you drive it with the key. And ending a session is as human as it gets: say stop on its own — case-insensitive, punctuation ignored — and voice chat ends.
You can also interrupt. Hermes accepts barge-in at any point in its turn — cut the agent off mid-sentence and it yields, which we unpacked in our Hermes barge-in coverage. Echo is handled the sensible way: the bot automatically pauses its own audio listener while playing TTS replies, so it never hears itself and answers its own voice.
Voice-first agent workflows like this get traded daily inside AI Profit Boardroom — 3,000+ members running Hermes, Agent OS, and automations that talk back. If you want working setups instead of trial and error, start there.
📺 Watch: Hermes Agent: New Voice Activation Update is INSANE!
The No-Buttons Option: Full-Duplex GPT-Live
Push-to-talk still too much friction? The docs describe a Desktop GPT-Live mode — full-duplex voice with no keys at all, both audio directions open like a phone call. The honest print: it is billed at $0.05 per minute of session time. A hands-busy twenty-minute session costs a dollar; an all-day open line compounds fast, which is exactly why the free push-to-talk default exists.
Voice On Your Platforms: CLI, Telegram, Discord, WhatsApp
Hermes voice is not a terminal trick. Voice Mode follows your agent onto every platform it lives on, and the official docs tune the audio delivery per platform.
- CLI — push-to-talk with Ctrl+B, with reply audio saved to a local cache so nothing vanishes after playback.
- Telegram — replies arrive as native voice bubbles in Opus format. With /voice on, Hermes speaks only when you send a voice message; with /voice tts, every reply comes back spoken.
- Discord — the same /voice on and /voice tts logic, plus live voice channels: /voice join pulls the bot into your current channel (the docs list a channel alias for the same command) and /voice leave disconnects it. The bot needs Connect and Speak permissions, and audio ships as Opus with an MP3 fallback.
- WhatsApp — voice replies arrive as MP3 attachments, no drama.
📺 Watch: Hermes Voice Agents: Automate ANYTHING!
Who Does The Hearing: Seven STT Providers
Speech-to-text is where your words become the agent's input, and the Hermes docs list seven STT options. The default costs nothing: local Whisper runs on your own machine, with model sizes from tiny at roughly 75MB up to large-v3 near 3GB. Base is the default, and a sensible middle ground.
Need speed over privacy? Groq's Whisper API transcribes in roughly half a second and offers a free tier — the fastest hearing Hermes can get without paying. OpenAI's Whisper API is the paid classic, with Mistral Voxtral, xAI, ElevenLabs, and DeepInfra rounding out the seven. Language hints are configurable too — if you dictate in one language all day, tell the STT engine once and accuracy climbs.
Who Does The Talking: Eleven TTS Engines
Text-to-speech is where the Hermes docs get genuinely fun — eleven TTS providers, three of them fully local and free.
| TTS engine | Cost | What stands out |
|---|---|---|
| Edge TTS | Free (default) | Good quality, roughly one-second latency, 322 voices across 74 languages |
| ElevenLabs | Paid | Excellent quality |
| OpenAI TTS | Paid | Voices include alloy, echo, fable, onyx, nova, shimmer |
| MiniMax | Paid | Excellent, with global and CN endpoints |
| Mistral Voxtral TTS | API | Mistral's voice engine |
| Google Gemini TTS | Free tier | Excellent — 30 prebuilt voices such as Kore and Puck |
| xAI TTS | Paid | Excellent, custom voice cloning via the xAI console |
| DeepInfra | API | Listed as a supported provider |
| NeuTTS | Free, local | Runs entirely on your machine |
| KittenTTS | Free, local | Voices like Jasper, Bella, Luna — unavailable on Intel macOS and Windows ARM64 |
| Piper | Free, local | 44 languages, models auto-download at roughly 20-90MB |
The details worth savoring. Gemini TTS supports persona prompt files and expressive audio tags — whispers and excited delivery, switched on via a setting — which turns a flat reply into a performance. xAI lets you clone a custom voice through its console, so Hermes can answer in a voice you built. And Edge TTS, the free default, quietly ships 322 voices; most people never change it and still sound great.
Limits And Setup Reality
Honest constraints, straight from the docs. Every TTS provider caps characters per request — Edge takes 5,000, OpenAI 4,096, Gemini a roomy 32,000 — and longer replies split automatically into sentence-aware chunks, so you hear natural breaks instead of mid-word cuts.
Setup needs real system dependencies: PortAudio for the microphone, ffmpeg for audio conversion and Opus on Discord, and espeak-ng if you run NeuTTS. On macOS the big one is a single command — brew install ffmpeg. Voice configuration lives in the Hermes config file at ~/.hermes/config.yaml, under its tts and stt sections, though you rarely touch it by hand: run hermes tools to pick voice engines and hermes model to pick the brain.
Two shortcuts. Nous Portal subscribers get OpenAI TTS through the managed Tool Gateway with no separate key. And Julian's standing AIPB support advice applies here: do not hand-install dependencies — ask your agent to set them up and report back. Want a human to map this onto your business first? Book a free strategy session and leave with a voice-agent plan.
Where Voice Fits Julian's Agent OS
In Julian's stack, voice is the input layer for Hermes as an operating system. The Apollo name travels with his Agent OS bundle, where voice control ships alongside the agent's own screen — you speak, Hermes acts on its own desktop — and the Hermes memory system that carries context between conversations. His latest coverage (his video, 15 September 2026) pairs the Hermes voice agent with GPT-6 Astra as the brain: Astra thinks, Hermes listens and speaks.
Start with the Agent OS guide for the full architecture, then check Goldie Bench to see how the models behind the voice actually score. And if you want the bundle plus the people running it daily, AI Profit Boardroom is where it lives — $69/month locked in (normally $110), with 3,000+ members inside.
Hermes Voice FAQ
Is Apollo Official?
No. The official Hermes docs call the feature Voice Mode, with a companion Voice & TTS page — Apollo never appears in them. It is community shorthand that travels with Julian's Agent OS bundle. Same capability, unofficial label.
Does Hermes Voice Need A Wake Word?
No. The default is push-to-talk: press Ctrl+B, speak, and let the silence detection close your turn. Wake-word listening exists as a separate feature, and saying stop by itself ends the voice chat entirely.
Is Hermes Voice Free?
It can be, end to end. Local Whisper hears you for free, and Edge TTS — or the local NeuTTS, KittenTTS, and Piper engines — talks back for free. The metered exception is the full-duplex Desktop GPT-Live mode at $0.05 per minute; ElevenLabs, xAI, and the other paid voices are optional upgrades.
Say Something To Your Agent Today
Voice is the difference between using Hermes and working with it. Type /voice on, press Ctrl+B, and give your agent one real task out loud — you will not go back to silent typing.
Two ways to go faster: join AI Profit Boardroom for the Agent OS bundle and 3,000+ members already running voice-first, or grab a free strategy session and map Hermes voice onto your business one-on-one.
