The hermes herald ai agent is Hermes Agent running as the interface of Herald OS: you talk to it or type to it, and it opens apps, plans missions, remembers your preferences and runs your automations for you. Herald OS is the shell. Hermes is the brain doing the work. Here is how that agent actually behaves inside Herald, how it keeps you safe, and whether you should pick Herald OS or Hermes Desktop.
📺 Watch: Herald OS DESTROYS Hermes Agent Desktop?
🔥 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
Herald OS is an independent, MIT-licensed project by Luke The Dev, per the README at iamlukethedev/Herald-OS. It is not made by Nous Research, and it is not affiliated with or endorsed by Nous. It builds on Hermes Agent, which is also MIT. It is alpha (0.1), and the first public release, v0.1.0-alpha.1, came out on 6 October 2026.
If you want to get properly good at Hermes, the AI Profit Boardroom has Hermes Agent courses, daily Hermes updates and a course on building an agentic operating system. Join the AI Profit Boardroom for $69/mo.
How the Hermes Herald AI agent works
The whole idea, per the README, is "an agent-native operating system, with Hermes Agent as the interface." Instead of clicking through menus, you tell Hermes what you want.
When Hermes first came out around March or April, everyone ran it in the terminal. It worked, but it was messy. Putting the agent behind a proper UI makes it much easier to navigate, control and command. That is what Herald does.
One thing to know up front: Herald OS does not include Hermes. You install Hermes Agent separately, using the official install script from hermes-agent.nousresearch.com, then run hermes setup. A source build of Herald needs Hermes Agent 0.21.3 or newer. Honestly, I just went into Claude and said, can you set this up? About two minutes later it was running.
Talking to Hermes: chat and voice
You can type to Hermes on its chat page, which has a preview panel and sources, so you can add files for Hermes to read. Or you can talk.
Per the README, you press the voice key or say "hey Hermes". The free voice engine works with any speech provider, and there is an opt-in realtime engine if you want faster back-and-forth. You can even operate the UI by voice, for example "open missions". In Settings I could choose a local or live engine, test it and see the list of voice commands. For more on the voice side, see my Hermes voice agent breakdown.
Missions
Missions are where Hermes acts like an agent, not a chatbot. You give it a mission, something like "build out a new website for me", and it plans it step by step. You can start missions from the command bar too, which also opens pages, adds memories and runs automations.
Memory
Hermes remembers your preferences, and the Memory page shows you everything it knows. That transparency is good. It also means you need to be careful. My Memory page showed my Telegram details on screen. If you screen-share or record, check that page before you hit record.
Automations
Hermes runs scheduled routines for you, and you can switch each one on or off. My automations synced automatically across Herald OS, Hermes Desktop and my own Agentic Operating System. I run a daily 6am to-do recap, daily Japanese learning and morning inbox triage summaries.
Per the README, automations can also fire on login, after a break, on low battery or after a crash.
📺 Watch: Hermes3D Is INSANE: AI Agents Now Have a 3D Office
Permission tiers, refused locations and the audit log
An agent that can touch your whole computer needs guardrails. Per the README, every system tool in Herald OS has a tier:
- Read: runs directly and is logged.
- Act: runs directly and is logged.
- Mutate: asks you first.
- Destructive: asks you every single time.
Some places are always refused, no matter what you tell Hermes: your ~/.ssh folder, your keychains and Hermes's own credentials. Everything gets written to an audit log saved on disk, so you can check what the agent did.
On top of that, Settings has an auto-approve switch, for example "ask before sending messages", and an activity history. I find activity history really useful, because stuff gets set up in the background and you forget about it.
The README still says: "Expect rough edges, and keep backups of anything you let an agent touch." Do that.
The pick-up-where-you-left-off privacy detail
The dashboard can suggest what to pick back up. Per the README, this is off by default. When you turn it on, it suggests up to three threads of work using names and dates, not file contents. Those names and dates are sent to your model provider. Small detail, but you should know where your data goes before you flip the switch.
📺 Watch: Hermes AI Agents Just Went Portable
Choosing the model behind Hermes
Hermes needs a model provider. Per the README, that can be a Nous Portal account or an API key from OpenRouter, OpenAI, Anthropic, a local model and others. You change it with hermes model in the terminal, or in Settings under Hermes and agents.
Settings also let you choose where agents complete tasks, including Docker, which people ask me about a lot, and how many background agents run. You can tell it what software it can use, such as Claude Code, Codex and LM Studio. If you are not sure which model to run, check Goldie Bench.
Want help picking the right model and setup for your business? Book a free AI strategy session.
A quick look at the rest of the UI
The dashboard greets you ("Good morning, Julian. Your day already in motion."), shows what is scheduled today, what needs attention and your missions. Spaces let you split ideas, work and personal, or set up one space per client, which is perfect for people running several businesses. Files covers recent, favourites, projects, downloads, shared and deleted. The Connections page did not work or sync for me, and the top of the window cuts off slightly. It is open source, so who am I to be picky?
That is the short version. For every page and setting, read my full Herald OS review and UI tour.
Herald OS vs Hermes Desktop
Both put a UI on top of Hermes Agent. I think Herald OS looks much nicer. Hermes Desktop is still good, and it has plugins. Here is how they line up based on what I tested and what the Herald README says.
| What matters | Herald OS | Hermes Desktop |
|---|---|---|
| UI | Full agent-native OS: dashboard, spaces, missions, memory, files, automations, command bar, twelve themes | Desktop app for Hermes; good, but not as polished-looking in my test |
| Plugins | Paste a GitHub link and click add | Has plugins |
| Platforms | macOS 13+ on Apple silicon only, plus Linux (ISO, Arch, Omarchy, tarball) | Mac, Windows and Linux (per Nous Research) |
| Status | Alpha 0.1, first release 6 October 2026, independent project | Established option I still rate as good |
| Automations | Synced with Hermes Desktop and my Agentic Operating System | Synced with Herald OS |
| Who it suits | People who want a clean, full-screen agent workspace and are fine with alpha bugs | People who want the steadier pick with plugins |
My full take on the other side is in my Hermes Desktop app review.
Which one should you use?
Use Herald OS if you want the nicer UI, you run multiple clients or companies and want separate spaces, and you are on an Apple silicon Mac or Linux. Use Hermes Desktop if you want something more settled, you rely on its plugins, or you are on an Intel Mac, since Herald has no Intel build.
Because automations synced across both for me, you do not have to pick forever. Try Herald OS, keep Hermes Desktop installed, and see which one you actually open each morning.
If you want to build your own system around all of this, my Agent OS guide shows how I run mine, including agents beyond Hermes.
Ready to go further? Get the Hermes courses inside the AI Profit Boardroom, or grab a free AI strategy session and we will map out your setup.
FAQ
What is the Hermes Herald AI agent?
It is Hermes Agent acting as the interface of Herald OS. You talk or type to Hermes, and it opens apps, organises files, plans missions, runs automations and remembers your preferences, with permission tiers controlling what it can change.
Do I need Hermes installed for Herald OS?
Yes. Hermes Agent is installed separately. Per the README, you install it with the official script from hermes-agent.nousresearch.com, run hermes setup, then install Herald OS. You also need a model provider such as a Nous Portal account or an API key.
Herald OS or Hermes Desktop?
Herald OS if you want the cleaner, fuller UI and can live with alpha bugs. Hermes Desktop if you want the steadier option with plugins or you are on an Intel Mac. Automations synced across both for me, so you can run them side by side.
Can Hermes delete my files in Herald OS?
Not without asking. Per the README, destructive tools ask every time and mutate tools ask first. Your ~/.ssh folder, keychains and Hermes's credentials are always refused. Keep backups anyway, because it is alpha.
