Hermes AI's New Browser Agents are Here

Hermes Agent just changed browser agents forever. A brand new free update just dropped and it makes your AI browser agent way more powerful. Your agent can now browse websites, click buttons and fill out forms twice as fast and it can run way longer before hitting limits. That means bigger jobs, more work done all whilst you do something else. In this video, I'll show you exactly how to turn it on. It's basically one line and I'll show you the one setup where this doesn't work at all. And that part catches most people out. So stick with me to the end, let's get into it. So until today, Hermes had 12 different browser tools, one tool to navigate to a page, one tool to click, one tool to type, one tool to scroll, one tool to press a key, one to go back and one to take a screenshot and so on. So you had 12 different tools. And here's why that was a problem. Every single one of those tools have to be described to the AI in every single request. So even if your agent just wanted to click a button, it was carrying the instruction manual for all 12 tools alongside it. Every time that's a lot of wasted tokens before the agent even does anything, but it gets even worse. So every action was its own round trip. If the agent clicks, then it stops. Then it reports back to the model. Then the model thinks, then it sends the next click. Then it stops again. So click report, think type report, think scroll report. Think think about how you'd feel if you had to phone a friend for permission before every single step you took in a supermarket. So can I walk to aisle three? Yes. Can I pick up the milk? Yes. Can I put it in the trolley? Yes. You'd lose your mind and you'd spend all your day on phone calls. That's how AI browser agents have worked. Not just Hermes, but most of them. Now here's a new way. It's called browser use mode and it runs on something called browser use CLI 3.0. So instead of 12 tools, the agent gets one. It's called browser use mode and it runs on something called browser use CLI 3.0. So instead of 12 tools, the agent gets one. And instead of asking permission for every click, the agent writes one script that handles the whole job. So navigate, click, type done. One plan written run once, run to start from finish. I call this the one script engine. And once you understand the one script engine, you understand where all AI agents are heading when it comes to browser use. Instead of the AI doing one tiny action at a time and checking in constantly, the AI writes a full set of instructions up front and hands them over. It's like your friend just writing you the full shopping list with directions and you just go and do the shop. It's one trip with no phone calls. The numbers actually back this up as well. So news research posted the results this morning. Token use dropped 48 to 66% across their tests with Opus 4.8 driving the browser. Tasks that used to burn 50 to 64,000 tokens dropped into the low 20,000 range. That's the same job, but it's done for roughly a third of what it used to cost. And they measured no drop in accuracy. So where does this come from? Well, first you got the tool schema, the instruction manual, the model carries around. That shrank from a whole stack of tools down to one. And second, the agent now drives the browser with code instead of individual actions. So the total conversation between the model and the browser gets way, way shorter. Now, why should you even care about token counts? Because tokens are the meter running in the background of everything your AI does. When the meter runs 60% slower, two things happen. Your agent gets cheaper to run and your agents can run for longer before hitting limits. Longer tasks, bigger jobs, more runs per day. The token cost was the ceiling on what browser agents could actually do for you. News research just raised that ceiling. So quick example of what that means day to day. Let's say, for example, your agent can now check five different competitor websites every morning, tell you what changed. And now that daily habit would cost about half of what it did last week. Now, if you want to actually put Hermes to work in your business, this is exactly what we do inside the Aeroprofibro.com. So Hermes is wired directly into our agent operating system. The same system where you can plug in your claw, your open claw, any other agent, or sharing one memory. We've got tutorials showing you how to set up Hermes browser automation step by step. So your agent can do research, fill out forms, and watch competitor sites whilst you go and do something else. You get the agent operating system zip file, a 30-day roadmap for setting it all up, and daily updates as we improve it. Plus, four weekly coaching calls every week where you can ask questions about your own Hermes setup live. And a community of over 3,700 business owners, plenty of them running Hermes right now. Link in the comments description or go to the AI Profitbomb.com. Okay, let's talk about how to actually turn this on because it's pretty simple. And then I'll show you what most people get wrong. So if you have Hermes installed and the browser use CLI is available on your machine, this mode is now the default. You don't have to do anything. Hermes checks if the CLI can run, and if it can, your agent gets the single browser tool automatically. If the CLI can't run, Hermes quietly falls back to the old 12 tools, so nothing breaks. If you want to force it, you add one line to your Hermes config, browser.backend, set to browser use, and that's it, one line. And if you'd rather turn it off and keep the old tools, you can type slash browser use off. One command either way. Now, here's the first belief I want to knock down because I can already hear it. Some people want to say, this sounds like coding. I'm not a coder. This isn't for me. If you listen carefully to what's actually happening here, the agent writes the code, not you. So you just type a plain sentence like, go to this website, fill out this form with these details. And the AI writes CPython itself, runs itself and reports back to you. So you never see the code unless you want to. The whole point of this update is that the machine handles the machine stuff. Your job is just to describe what you want in normal words. So if you can write a text message, you can run this. Here's what makes this update even better than it sounds. It's not actually locked to one type of browser. This is actually worse for everyone. So it could be your local browser on your own computer, browser base, cloud browsers, browsers, cloud browsers, whatever you're already using. The new mode sits on top and drives whichever browser you've already picked. There's exactly one exception. And this is the catch I mentioned at the start. So if you're using Camofox, that's the local privacy focused browser option. Some Hermes users run. The new mode doesn't work with it. Camofox setups automatically keep the old browser tools instead. So if you flip the setting and nothing changes, it might be because you're on Camofox. And there's a clever detail buried in the documents that solves something really annoying. So Hermes now does hybrid routing. If your agent needs to visit a normal public website, it uses your cloud browser. But if it needs to look at something private, something running on your own machine or your home network, it automatically spins up a local browser for that instead. So the cloud provider never even sees your private stuff. So both can happen in the same conversation without you touching a single setting. So what does this all actually let you do? Let's keep this real because the point of browser automation was never the technology. The point is like the boring jobs. It takes those off your plate. Here's an example. Let's say you run a small shop and you want to know when a supplier changes their prices. Your agent checks the page every morning and messages you only when something moves. That kind of job used to be exactly the type of task where the old system fell over. Lots of pages, lots of clicks, lots of tokens. It would work, but slowly and the costs added up if you ran it up daily. Under the OneScript engine system, the agent writes OneScript for the whole check and then runs it clean. Some people as well are going to say like browser agents aren't that great. They look good in the video, but they're not reliable enough for real work. And there's some truth in the history here. I'm not going to pretend otherwise. So web automation has been flaky, right? Pages change, butters move, agents get lost. But two things are different now. So first, News Research actually tested this new mode against their old system and measured no accuracy drop. So the savings didn't come at the cost of the results. And second, the whole approach got simpler. So there's fewer moving parts, one tool instead of 12. Fewer round trips means fewer chances for the agent to lose a thread halfway through a task. Should you still check your agent's work? Yes. Should you start with small tasks before trusting it with big ones? Absolutely. But this stuff never works is an outdated belief and it's getting more outdated every day. Now let's zoom out for a second because there's a bigger lesson in the whole update. It tells you where everything's going. The interesting part of this announcement isn't just that Hermes got more efficient, it's how they did it. They stopped making the AI do one tiny action at a time, started letting it write a plan and execute the plan. So one reply on the announcement actually summed it pretty well, which is a dozen schemas per click was never a capability. The agents were always capable. They were just being run in the most expensive way possible. And you're going to see this same shift everywhere. Agents that write their own scripts. Agents that plan a whole job upfront instead of asking permission every step. And every major agent tool is moving this direction because the savings are too big to ignore. So Hermes just got out the first time browsing in public with numbers attached. Which brings me to what you should actually do about this. Three simple steps. So step number one, if you already run Hermes, update it and check whether browser use mode is active. If the CLI is available, it probably already is. It's a default now. Run one small browser task and watch how much faster it feels. Step number two, if you don't run Hermes yet, this is a genuinely good week to start. It's free, it's open source from news research and browser automation just became one of its strongest features. You can set up, connect it to your browser and give it one boring job you do every week. And step number three, this is one that really matters is you can pick a real task from your business. Not a test, not a toy, just a real repeating annoying task that involves a website, checking something, filing something, hand it to the agent. Because the moment you watch an AR complete a task, that's pretty nice. This update handed everyone a browser agent that costs roughly half as much to run as it did before. Inside the AI Profit Boardroom, we've actually built the setup path for exactly this. Hermes plugs into the agent operating system alongside Claude, OpenClaw, all your other agents. And we've added workflows for browser tasks, research agents, SEO agents, even video and avatar tools inside one system with shared memory. So you get the agent OS zip file, the 30-day roadmap to get it running in your business, video tutorials and daily updates as new versions ship, including updates built around this new browser use mode. Four weekly coaching calls where you can bring your Hermes browser setup and get it fixed live. And daily step-by-step tutorials on putting agents to work for leads and customers. We've got over 3,700 business owners inside, many of them already running Hermes Daily. And a member of Maps, you can find people near you who are doing the same thing. There's always someone online too, so link in the comments description or go to the AI Profit Boardroom.com. To summarize, 12 tools became one, token costs dropped by half for browsing, accuracy stayed the same and browser use is getting better.