Hermes Agent browser is the browser automation toolset built into Hermes Agent, and it lets the agent open web pages, read them, click, type, scroll and take screenshots.
Hermes Agent is the free open-source agent from Nous Research, so the browser toolset costs nothing to switch on.
This page is the reference version, so each feature gets a plain explanation, a score out of 10 and the limit that comes with it.
The facts come from the official Hermes browser docs, the toolsets reference and the security docs, which I re-read on 12 October 2026.
The scores are my own judgement of what the docs describe and how I use Hermes day to day, so please treat them as an opinion rather than a lab measurement.
Hermes Agent Browser At A Glance
| Question | Short answer |
|---|---|
| What is it? | It is the built-in browser automation toolset in Hermes Agent. |
| Who makes it? | Nous Research makes Hermes Agent, and the toolset ships inside it. |
| What does it cost? | The toolset is free, and cloud browser providers charge on their own plans. |
| How does the agent see a page? | It reads a text snapshot where each element has a reference such as @e1. |
| How do I turn it on? | You run hermes tools enable browser. |
| How do I turn it off? | You run hermes tools disable browser. |
| Where does the browser run? | It runs locally, in your own Chrome, or in a cloud browser. |
| What is the biggest limit? | The docs say it can't download files and it works from text, not pixels. |
| My overall score | I give it 8 out of 10. |
What The Hermes Agent Browser Is And Why It Exists
A language model can't use a website by itself.
It needs a tool that turns a page into something it can read and a second tool that turns its decision into a click.
The Hermes browser toolset does both jobs.
The docs say pages are represented as accessibility trees, which are text-based snapshots.
Interactive elements get reference IDs, and the agent uses those IDs for clicking and typing.
That design keeps the browser usable by ordinary text models, with screenshots held back for pages that need a visual check.
It exists because web search can only fetch information, and plenty of real work needs a form filled in or a button pressed.
How I Scored The Hermes Agent Browser
I scored eight areas from 1 to 10.
I looked at how much each area can do according to the docs, how hard it is to set up and how honest the limits are.
I didn't run a timed benchmark across every backend, and I won't pretend that I did.
My hands-on evidence is narrower, and I describe it in the examples section near the end.
Every Hermes Agent Browser Feature, Scored
| Feature | Score | One-line reason |
|---|---|---|
| Core page tools | 9 | Navigate, snapshot, click, type, scroll, press and back cover most web tasks. |
| Vision and screenshots | 8 | A screenshot plus AI analysis handles pages that text can't describe. |
| Developer tools | 8 | Console reading, JavaScript evaluation, raw CDP and dialog handling go deep. |
| Backend choice | 9 | The docs list local, your own browser and several cloud options. |
| Logged-in sessions | 8 | Real profile browsing and the credential vault are well thought out. |
| Safety controls | 8 | Blocklists, URL checks and origin-bound password fills are built in. |
| Setup and switching | 8 | One command enables it, although the backend menu takes some reading. |
| Limits and reliability | 6 | No file downloads, truncated snapshots and bot detection are real constraints. |
🔥 Want my exact Hermes Agent Browser Explained setup, step by step? Inside the AI Profit Boardroom you get step-by-step video trainings, my installable Agent OS and four coaching calls a week with 3,400+ members building real automations. → Join the AI Profit Boardroom
Core Page Tools: 9 Out Of 10
These are the tools the docs list for everyday browsing.
browser_navigateopens a URL and has to be called before any other browser tool.browser_snapshotreturns the text snapshot, either compact or full.browser_clickclicks an element by its reference.browser_typeclears a field and types new text.browser_scrollmoves the page up or down.browser_presspresses keys such as Enter, Tab and Escape.browser_backreturns to the previous page.browser_get_imageslists the images on a page with their URLs and alt text.
The compact snapshot shows only interactive elements, and the full snapshot shows the complete page.
I score this high because the set is small and each tool does one clear thing.
Vision And Screenshots: 8 Out Of 10
browser_vision takes a screenshot and analyses it with a vision model.
The docs suggest it for complex layouts, charts and visual verification challenges.
The screenshot is saved and the file path comes back with the analysis.
On Telegram, Discord, Slack and WhatsApp, the agent can send that screenshot to you as a photo.
Screenshots are kept in a cache folder and cleaned up after 24 hours.
I took two points off because the Lightpanda engine has no graphical renderer, so screenshots aren't available there in the default mode.
Developer Tools: 8 Out Of 10
browser_console returns console messages and uncaught JavaScript errors.
It can also evaluate a JavaScript expression and return the result.
browser_cdp is a raw Chrome DevTools Protocol passthrough, which the docs call the escape hatch.
It only exists when a CDP endpoint is reachable at session start.
browser_dialog answers native alert, confirm and prompt boxes, which used to freeze a page.
There is also a session recording setting that saves each session as a WebM video and cleans up recordings older than 72 hours.
These are strong tools, but most business users will never need them.
🔥 Want to pick the right tools and setup for your own agent? Inside the AI Profit Boardroom, you get step-by-step Hermes tutorials, my installable Agent OS and four coaching calls a week. You'll be learning with 3,400+ members. → Get access here
The Browser Use Default Mode
The docs say Browser Use mode is now the default.
When the browser-use command-line tool can run, the agent gets a single browser_exec tool, and it writes short Python to drive the page.
If that tool can't run, Hermes falls back to the built-in tools above.
browser_exec is only offered to sessions that also have terminal access, because it runs model-written code on your machine.
You can opt out with /browser use off.
I haven't scored this as a separate feature, because I cover it properly in my Hermes Agent browser use guide.
Backend Choice: 9 Out Of 10
| Backend | Type | What the docs say |
|---|---|---|
| Local browser | Local | Hermes runs its own packaged Chromium through agent-browser. |
/browser connect |
Local | It attaches to your running Chrome, Brave, Chromium or Edge over CDP. |
| Real profile browsing | Local | It drives a copy of your active profile with your logins. |
| Browser Use cloud | Cloud | It offers managed Chromium with stealth, proxies, CAPTCHA solving and saved profiles. |
| Browserbase | Cloud | It is an alternative provider with anti-bot tooling. |
| Firecrawl | Cloud | It offers cloud browsers with built-in scraping. |
| Camofox | Self-hosted | It is a Firefox-based anti-detection browser you run in Docker. |
| Lightpanda | Local engine | It is a lightweight headless engine. |
| Nous subscription | Cloud | Paid Nous Portal subscribers get cloud browsing through the Tool Gateway. |
The docs also describe hybrid routing, where a cloud browser handles public sites and a local sidecar handles localhost and LAN addresses.
That range is why I score it 9.
I haven't ranked the backends here, because that is the whole job of my best browser for Hermes list.
Logged-In Sessions And Cookies: 8 Out Of 10
The default local browser is a clean throwaway profile with no logins.
Real profile browsing is the setting browser.use_real_profile, and it is off by default.
When it is on, Hermes copies your active browser profile into a managed snapshot and launches your real browser on that copy.
The docs say your live profile is never opened directly.
It supports Chrome, Edge, Brave and Chromium, and it fails with a clear message on Firefox.
The credential vault is the second route.
It asks for a login the first time the agent reaches a sign-in page, encrypts it on your machine and fills it on later visits.
It can also generate two-factor codes when you save the authenticator key.
I took two points off for the friction, because you usually have to quit your browser before a real-profile session can copy the login databases.
Safety Controls: 8 Out Of 10
| Control | What it does |
|---|---|
| Origin-bound fills | A saved password is refused unless the page origin exactly matches the saved one. |
| Password redaction | The docs say the model never sees the password in tool results or logs. |
| Website blocklist | Listed domains are blocked across web search, web extract and browser navigation. |
| Private URL checks | Internal addresses are blocked on cloud backends and routed to a local browser instead. |
browser.restrict_evaluate |
It adds a denylist over sensitive JavaScript such as cookies, storage and network calls. |
| Dialog policy | You choose whether pop-up dialogs wait for the agent, auto-dismiss or auto-accept. |
| Ask before closing | Hermes never closes your browser to unlock a profile without your approval. |
The docs call real profile browsing a consent-gated convenience, not an isolation boundary.
They also admit that once a password is typed into a site, the site has it.
I respect that honesty, and it is part of why I trust the rest of the page.
Setup And Switching: 8 Out Of 10
The commands are simple, and I confirmed them against the help output on my install.
hermes tools list
hermes tools enable browser
hermes tools disable browser
hermes tools enable browser --platform telegram
Running hermes tools with no subcommand opens the interactive menu, and Browser Automation is where you choose a backend.
Inside a chat you can type /tools disable browser.
The desktop app has a Use My Real Browser Profile switch under Capabilities, Tools, Browser, according to the docs.
If the local browser is missing, hermes pm install agent-browser installs it.
When I ran the list command on 12 October 2026, the browser toolset was already enabled for the terminal on my machine.
I took two points off because the backend rules take a careful read.
A provider key only supplies credentials, and the active provider is whatever you selected in the menu.
Browser Tools vs Web Search And Extract
The web toolset has web_search and web_extract, and it fetches information.
The browser toolset drives a live page.
The docs tell you to prefer the web tools for simple retrieval, because they are faster and cheaper.
They tell you to use the browser when you need to interact with a page.
The toolsets reference adds that web_search is deliberately not part of the browser toolset, so disabling the browser keeps quick lookups working.
🔥 Want the Hermes setup I actually run every day? The AI Profit Boardroom includes my Agent OS, Hermes walkthroughs and four live coaching calls a week with 3,400+ members. → Join the AI Profit Boardroom
Limits And Reliability: 6 Out Of 10
The docs list these limits themselves.
- Interaction is text-based and relies on the accessibility tree rather than pixel coordinates.
- Snapshots over 15,000 characters are truncated, with the full version saved to a file.
- Cloud sessions expire based on your provider's plan.
- Cloud sessions consume provider credits.
- The browser cannot download files.
There are a few more worth knowing.
/browser connectis a terminal-only command and does nothing inside Telegram, Discord or a web chat.- Since Chrome 136, the debug port is ignored on your default profile, so you need a dedicated profile folder.
- Idle browser sessions are closed after 120 seconds by default.
CAPTCHAs and bot detection are the limit people ask about most.
The docs say Browser Use cloud includes CAPTCHA solving, and Browserbase lists CAPTCHA solving under basic stealth.
Headed mode keeps a visible window open so you can solve a challenge by hand.
Nothing in the docs guarantees access to a site that doesn't want automated visitors.
Troubleshooting Quick Reference
| Symptom | Likely cause | Fix |
|---|---|---|
Connection refused on /browser connect |
Chrome ignored the debug flag on the default profile. | Launch Chrome with a dedicated --user-data-dir. |
/browser connect is ignored in a chat app |
The command is terminal-only. | Run it inside hermes in the terminal. |
| Real profile opens signed out | The browser was running and held the login databases. | Quit the browser fully and retry. |
| Missing key error | The selected provider has no credentials. | Re-select a provider in hermes tools. |
| No browser tools listed | The local browser didn't install. | Run hermes pm install agent-browser. |
Real Examples From My Own Content
My clearest browser example is a Hermes skill for a daily AI news scan, which is published in my Hermes AI skills guide.
That skill falls back to browser_navigate and a full browser_snapshot when web search and web extract return nothing.
My note in it says the browser returned full content every time in March 2026, when the web tools returned none.
The same skill notes that Product Hunt showed a bot check, and the instruction is to skip it.
The video at the top of this page covers Hermes Computer Use, which controls desktop apps in the background.
The Hermes docs say browser work goes through the browser toolset rather than Computer Use, so treat that video as the neighbouring feature.
If you'd like help applying agents to your own SEO, you can book a free SEO strategy session and we'll map out a plan together. If you're already experienced and want a higher-level room, have a look at the SEO Elite Circle.
My Verdict On The Hermes Agent Browser
I give the Hermes Agent browser 8 out of 10.
It is free, it works with text models, it offers more backends than most people need and its password handling is careful.
It loses points for no file downloads, for the setup reading around backends and for the bot detection that no agent fully escapes.
You should use it if your agent needs to click, type or work inside a logged-in page.
You should skip it for simple lookups, because web search and extract are faster and cheaper.
Related Reading
- My best browser for Hermes list ranks the backends in this review.
- My Hermes Agent browser use guide explains the Browser Use project and step-by-step tasks.
- My best Hermes Agent tools verdict scores the browser next to the other built-in tools.
- My best Hermes Agent setup list helps you choose where to run Hermes in the first place.
FAQ: Hermes Agent Browser
What is the Hermes Agent browser?
The Hermes Agent browser is the built-in browser automation toolset in Hermes Agent from Nous Research.
It lets the agent navigate, read, click, type, scroll and take screenshots on real web pages.
Which tools are in the Hermes browser toolset?
The docs list browser_navigate, browser_snapshot, browser_click, browser_type, browser_scroll, browser_press, browser_back, browser_get_images, browser_vision, browser_console, browser_cdp and browser_dialog.
The toolsets reference also lists browser_exec and the browser_vault tools for saved logins.
How do I enable or disable the Hermes Agent browser?
Run hermes tools enable browser to switch it on and hermes tools disable browser to switch it off.
Add --platform with a platform name to change one platform only.
Which browsers does Hermes Agent support?
The docs list a local packaged Chromium, your own Chrome, Brave, Chromium or Edge over CDP, Browser Use cloud, Browserbase, Firecrawl, Camofox and Lightpanda.
Paid Nous Portal subscribers can also use cloud browsing through the Tool Gateway.
Does the Hermes Agent browser need a vision model?
Most browsing works from text snapshots, so a vision model isn't needed for clicking and typing.
The browser_vision tool uses vision AI when a page needs a visual check.
What are the limits of the Hermes Agent browser?
The docs say it cannot download files, long snapshots are truncated at 15,000 characters and cloud sessions cost provider credits.
Some sites will also block automated browsers.
Also On Our Network
- 🌐 The step-by-step Hermes Agent browser setup walkthrough
- 🌐 Seven business jobs for the Hermes Agent browser
- 🌐 The Hermes Agent browser quick-start for an Agent OS
- 🌐 Real model scores for the brains behind the Hermes browser
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
About Julian
I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
- I've built a 7-figure SEO and link-building agency, Goldie Agency, with 70+ people in Bangkok.
- I've grown a YouTube channel to 400,000+ subscribers.
- I wrote the book "Link Building Mastery", which covers nine link building strategies.
- My Udemy courses have taught 29,000+ students.
→ Get my best AI training inside the AI Profit Boardroom
Use web search for lookups, switch the browser on for clicks and logins, and you'll get the best out of the Hermes Agent browser.
