Video: "Hermes Apollo: The AI voice agent that actually listens." by Julian Goldie on YouTube.

What Hermes Apollo actually is

Apollo is a voice interface bolted onto Hermes Agent — the open-source local AI agent built by Nous Research that Julian Goldie covers regularly. It is not a separate product or a cloud service. It is a mode you activate inside the same Hermes installation you may already have running.

The design is deliberately simple. There is no wake word — you just start talking. There is no button to hold. When you stop speaking, the agent replies immediately; it does not wait for a pause or make you indicate you have finished. And it does not just respond with text. Apollo hands the instruction to the rest of the Hermes stack, which then acts: opening browsers, building files, calling APIs, or whatever the task requires.

What Julian Goldie demonstrated

The demo in the video is a website build. He described a landing page out loud — the structure, the copy angle, a rough visual direction — and Apollo relayed the brief to Hermes, which built and rendered the page in roughly 90 seconds. He did not type a single character.

Worth knowing: the demo is favourable conditions. The page was simple, the model was loaded, and Julian already knew what he wanted. That said, the underlying mechanism is real. Apollo is a voice-to-Hermes bridge and the agent execution underneath it is the same Hermes stack that has been handling written prompts for months. Voice does not make Hermes smarter or more capable — it changes how you interact with it, not what it does.

Three components under the hood

Julian broke the system into three parts. The voice layer handles transcription — it picks up what you say and converts it to text that Hermes can process. The brain layer is Hermes itself — whichever model you have configured (Kimi K3, Sonnet 5, Opus, or anything else in the Hermes model list). The hands layer is the Hermes skill and tool system — the part that actually opens a browser, runs a terminal command, writes a file, or sends a message.

Apollo sits at the top. It feeds the voice input to the brain, the brain plans a response and issues tool calls, the hands execute them. It is the same chain that runs on typed input; you have simply removed the keyboard from the front end.

What is overhyped and what is genuinely useful

The 90-second website claim is real in the narrow sense that Julian did it. But building websites is something Hermes could already do from a typed prompt. Apollo does not make Hermes better at the task — it makes the task less friction to start. That is a meaningful distinction.

Where it does seem genuinely useful: hands-on work. If you are at a desk and want to issue a quick instruction while your hands are busy, or if you find writing long structured prompts tedious, speaking is faster. It also lowers the barrier for people who know what they want but struggle to phrase it as a prompt. Talking is easier than typing a brief for most people.

What it does not solve: the usual Hermes requirements. You still need a model configured, skills set up for the tasks you want to run, and a machine capable of running it. Apollo is a front end. The same constraints apply underneath.

What this means as a direction

Hermes has been adding components steadily this year — Cloud for always-on execution, Oracle for news monitoring, Astros for competitor tracking, Obsidian for memory. Apollo is the voice interface to the same stack. The pattern is an Agent OS: a system where different specialised pieces handle different jobs and you interact with the whole thing however suits the moment — typed, voiced, or via a dashboard.

It is worth tracking because the direction matters even if individual releases are incremental. The question for a business owner is not whether this demo was impressive; it is whether an AI agent that you can brief out loud, that runs tasks in the background, and that costs roughly what a software subscription does changes what a small team can produce. Based on what Hermes can already do, the answer is probably yes.

Where this connects to NordSys

If you want an AI agent running inside your business — briefed with your goals, handling tasks on a schedule, and reporting back without needing constant instruction — that is what our AI Agents service provides. We configure it, name it, and manage it for you, without the setup overhead. From £6 a day.

See our AI Agents →