The Hermes desktop browser is the update I had been waiting for without realising it: your agent can now open the actual live web page in a rail beside your chat, and it shipped in v0.20 on 3 August. I have been running it all week on real work, and the shift is simple but massive. Instead of the agent telling me to go and look at something, it goes and looks at it itself, while I watch in the same window.
📺 Watch: NEW Hermes Desktop Browser Update Is INSANE!
🔥 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
Below: how the new browsing works, the two demos I ran on camera, how to install it, and the three tips that stop you making the mistakes I am already seeing. Everything here comes from my own testing inside Agent OS, where I run every tool I review.
What actually changed: the agent browses, you watch
Before this release, Hermes could search the web and summarise what it found, but the loop always ended with me opening a tab to check things myself. Now the agent opens a live page beside our conversation, moves around inside it, and tells me what is on it. One window the whole time. No tab hopping. No pasting URLs into a chat box and hoping the model actually reads them. That is the whole story in four sentences; the rest of this page is the detail.
Quick primer: what Hermes actually is
If you are new to this, Hermes is an open-source AI agent built by Nous Research, and I mean agent, not chatbot. It reads files, writes files, runs commands, searches the web, and remembers what it learned. When it works out something hard, it saves the solution as a skill file in plain markdown, so it never has to solve the same problem twice.
It also lives in more than one place: a command-line version, a desktop app for Mac, Windows and Linux, and a gateway connecting Telegram, Discord, Slack, WhatsApp and email. It is the same agent everywhere, with the same settings, same memory, same sessions. Start in the desktop app, pick it up in the terminal, and nothing resets. I have a full Hermes desktop app guide if you want the complete picture; this page owns the browsing angle.
The Herald release, briefly
The desktop app first appeared in June this year as a preview. Version 0.20, the Herald release, landed on 3 August, and the numbers are genuinely mad: around 3,650 commits, roughly 1,400 merged pull requests, about 1,200 issues closed, and over 650 contributors in one release window. I have broken the whole thing down in my Hermes v0.20 overview and my Herald release deep-dive, so I will not repeat it all here. This page is about one thing: the browser.
Inside the Hermes desktop browser: how the preview rail works
The headline feature is the preview rail, a right-hand panel beside your chat that renders web pages, files and tool outputs side by side while you keep talking. Paired with Hermes's browser tools, it means the agent can open a live page, move around inside it, and tell you what is on it, all without you leaving the conversation.
Think about what that removes. The old workflow was: agent suggests a link, you open a new tab, you read, you come back, you paste bits into the chat, you explain what you saw. The Hermes desktop browser collapses all of that into one window. You talk on the left; the web happens on the right.
People are calling it the app you never have to leave, and I understand why. I will keep it balanced, though: some users are reporting rough edges too, which does not surprise me on a feature this new. My honest take is that both camps are right. It is the most useful thing in the release, and it is still v0.20 software. Go in with that expectation and you will be delighted rather than annoyed.
If you want your agent doing the browsing while you do the thinking, the AI Profit Boardroom has the Agent OS walkthroughs for updates exactly like this one. → Get the Hermes walkthroughs
📺 Watch: Run Hermes Desktop With Ollama for FREE
My demos: browsing Goldie Bench and building a landing page
First test on camera: I asked Hermes to look up goldiebench.com. That is Goldie Bench, my own hands-on model-testing site, so I know exactly what should be on it. The agent pulled the site up in the preview rail. Then I pushed it: go to the medal section, then the compare section. It navigated to both, and I never touched a browser tab once.
Second test: I asked it to create a simple HTML landing page for a free SEO audit offer, with a headline, benefit section, form, testimonials, FAQ and call-to-action buttons. It built the page and showed it to me in the rail. That is the pattern I want you to steal: describe what you want in plain English, let the agent build it, then look at the result in the preview rail instead of reading a description of what it supposedly did.
What else in v0.20 backs up the browsing
The browser is the headline, but several other Herald features make it far more useful in practice. My Hermes desktop features guide covers them properly; here is the short version.
| Feature | Why it matters here |
|---|---|
| Real conversational voice | It speaks as it generates, and you can interrupt it mid-sentence; it stops and listens. Works across CLI, desktop and messaging. |
| Custom wake words | On-device detection, so nothing is sent anywhere while it waits. Different wake words per profile, and saying "stop" ends the session. |
| Grounded citations | A research skill that matches quotes against the actual page text so it is not inventing them, citations that link to exact evidence, plus a fact-checking mode. An agent that browses and proves its quotes is a serious research tool. |
| The app as a platform | Generated HTML and small apps run in a sandboxed live preview, there is a plugin system with a Kanban board as the first plugin, a global hotkey to capture a thought from anywhere, and multiple windows. |
| Agent-to-agent | The A2A protocol v1.0 lets Hermes discover other agents, talk to them, and be driven by them. |
| CLI power commands | Shell mode plus /init, /context and /focus, and tools that recover from their own failures. |
📺 Watch: Hermes Desktop + Ollama is Insane (FREE!)
How to install it
- Head to the Nous Research Hermes site and download the app for Mac, Windows or Linux. There is also a one-line terminal install if you prefer.
- Already have Hermes in your terminal? Do not download anything. Run the hermes desktop command and the app launches with your existing config, keys, sessions and skills. Same agent, new window.
- On Windows, the installer is not code-signed yet, so SmartScreen shows a warning the first time. Click More info, then Run anyway. That is expected, and Nous Research says so in their docs.
- Open a chat, ask the agent to pull up a site you know well, and watch the rail render it.
If anything fights you, my Hermes desktop installation guide walks through the whole setup step by step.
Three pro tips, plus the /journey bonus
These come straight from my own sessions, including the mistakes.
- Use the preview rail deliberately. Do not let the agent describe what it did; ask it to show you. Say "open it in the preview" after every build or edit. That one habit cuts revision cycles right down, because you catch problems by looking rather than by trusting a summary.
- Watch the context meter. There is a meter in the status bar; click it and you get a breakdown of what is eating space. When it is filling up, start a fresh chat. Long, messy sessions are where quality quietly drops, and most people never work out why.
- Do not turn off approvals. This is the one people get wrong. There is a mode that skips confirmation prompts on risky commands, and it feels brilliantly fast, but this agent runs real commands on your real machine. Leave approvals on while you are learning, then turn them off later, on purpose.
Bonus: type /journey in the chat and you get an interactive map of everything Hermes has learned, every skill and memory on a timeline. Scrub through it, and edit or delete anything that is wrong. I check mine weekly. Clean memory means a better agent.
Hermes desktop browser FAQ
What is the preview rail exactly?
It is a right-hand panel in the Hermes desktop app that renders web pages, files and tool outputs beside your chat. The agent can open a live page in it, navigate around, and tell you what it found while you keep talking on the left.
Does the Hermes desktop browser replace Chrome or Safari?
Not yet, and I would not pitch it that way. It replaces the constant switching between your agent and your browser during agent work; for everyday browsing, keep your normal browser. And if you would rather run Hermes in a browser tab than a desktop app, see my Hermes agent web UI guide and my best Hermes web UI round-up.
Is it safe to let an agent browse and run commands?
It is as safe as your settings. Hermes asks for approval before risky commands, and as I said above, leave that on. An agent that can browse, build and execute is powerful precisely because it touches real things, so keep the confirmation prompts until you genuinely understand what you are approving.
Does it work on Mac, Windows and Linux?
Yes, downloads for all three are on the Nous Research site. Windows users will see a SmartScreen warning because the installer is not code-signed: click More info, then Run anyway, and you are through.
Do I need to reinstall to get the browser?
If you already use Hermes in the terminal, no. Run the hermes desktop command and the app launches with your existing config, keys, sessions and skills. New users just grab the normal download.
My verdict
The Hermes desktop browser is the clearest signal yet of where agents are heading: they do the looking, not just the talking. The preview rail has turned Hermes from a clever text window into something closer to a colleague sharing their screen, and once you have watched your agent open a page, click through it and report back, copy-and-paste research feels prehistoric. It is v0.20 software with the odd rough edge, but it is also the feature I now use every single day.
If you want an agent that browses, builds and shows its work while you stay in one window, check out the AI Profit Boardroom — inside you get Agent OS walkthroughs, weekly live coaching calls where you bring your screen and get unstuck, prompt libraries, 30-day roadmaps, and a room of people testing the same updates as you. → Join the Boardroom and put the preview rail to work
