Hermes Agent Browser Use is an official, built-in integration, and my verdict is that it is the best-documented browser setup I've read for an open-source agent, with real safety features and a few honest gaps.
Browser Use is the default browser driver in Hermes Agent, and Browser Use Cloud is a supported cloud browser provider.
That is the verdict in two lines.
The rest of this page is the reference version.
I'll go through what it is, who makes it, how to set it up, every feature, how to use it for real browser tasks, the costs, the limits, and a scores table near the end.
I'll also be clear about my evidence.
I read the official Hermes docs and Browser Use's docs, and I ran read-only commands on my own install on 12 October 2026.
I have not run a timed test of browser tasks, so the scores are my judgement of the design and the documentation, and they are not benchmark results.
Quick Reference: Hermes Agent Browser Use
| Item | Detail |
|---|---|
| What it is | It is Browser Use built into Hermes Agent as a browser driver and a cloud browser provider. |
| Who makes Hermes Agent | Nous Research makes it, under the MIT licence. |
| Who makes Browser Use | The Browser Use team makes it, and the main repo is MIT licensed. |
| The tool the model gets | The model gets one tool called browser_exec. |
| The engine | The engine is Browser Harness, wrapped by the Browser Use CLI 3.0. |
| Default status | It is the default when browser.backend is unset. |
| Free option | A local browser costs nothing. |
| Paid option | Browser Use Cloud is listed at $0.02 per browser-hour plus network usage. |
| Plugin or MCP needed | Neither is needed. |
I checked the last row myself.
The command hermes plugins search "browser-use" returned no catalog entries.
The command hermes mcp catalog printed its list of approved servers, and Browser Use was not on it.
That is not a gap, because the integration lives in the core of Hermes.
What It Is And The Problem It Solves
A browser agent has two jobs.
It has to read a page, and it has to act on it.
The older Hermes approach used about a dozen small tools, with one call for each move.
Every call sends the conversation back to the model, so long tasks get expensive.
Browser Use mode gives the model one tool instead.
The docs say the agent writes and executes Python in the browser to click, type, drag, scrape and interact with pages.
A single call can therefore complete a whole chain of steps.
Nous Research published its own test of this in the Hermes repo.
On their hard task set, the mean tokens per task fell by about 60% with Claude Opus 4.8 and about 66% with Kimi K3, at the same success rate.
Those are their numbers and their caveats, and I list the caveats in the limits section.
This page stays on Browser Use, and my Hermes Agent browser explainer covers the wider toolset.
🔥 Want my exact Hermes Agent Browser Use 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
How To Get It And Set It Up
- Install Hermes Agent and run
hermes update. - Run
hermes toolsand open Browser Automation. - Choose Local Browser for free local browsing.
- Leave
browser.backendunset, because Browser Use mode is the default. - Add
BROWSER_USE_API_KEYand choose Browser Use only if you want their cloud browsers.
The config for the cloud provider is short.
# ~/.hermes/config.yaml
browser:
cloud_provider: browser-use
One version note is worth your attention.
The current docs say the engine is a regular Python dependency of Hermes, so every install ships it.
My own install is a late-September build of v0.21.4, and its tools menu still lists Browser Use as a row that installs a separate CLI.
If Hermes tells you the Browser Use CLI was not found, it has fallen back to the built-in tools, and hermes update is the fix.
Hermes Agent Browser Use Features, One By One
Browser Use Mode And browser_exec
This is the headline feature.
It is the default driver, and it composes with whichever browser you have chosen.
The docs list Hermes's own headless Chromium, a Nous Subscription cloud browser, Browserbase, Firecrawl and Browser Use cloud browsers.
Camofox is the exception, because it has no endpoint for the harness to attach to.
The tool is only offered to sessions that have terminal access, since it runs model-written Python on your machine.
Verdict: It is the right default for strong models, and the terminal rule is a sensible limit.
Named Sessions
The browser_exec tool accepts a session name.
Each name gets its own background process and its own browser, so parallel subagents don't fight over one tab.
Verdict: It is a quiet feature that matters as soon as you run two browser jobs at once.
Browser Use Cloud
The docs say Browser Use Cloud runs managed Chromium with stealth and residential proxies on by default.
It includes CAPTCHA solving and persistent profiles for cookies, local storage and saved passwords.
Verdict: It is the option for sites that block a normal automated browser, and the pricing is simple.
Nous Subscription Tool Gateway
A paid Nous Portal subscription includes a Tool Gateway.
The docs say it routes cloud browser automation through Browser Use, with no separate Browser Use account.
Verdict: It is the least effort if you already pay Nous, and it is one more bill to watch if you don't.
Real Profile Browsing
Setting browser.use_real_profile to true lets the agent browse with your own logins.
Hermes copies your active Chromium-family profile into a managed snapshot and drives that copy headless.
Your live profile is never opened directly, and turning the setting off deletes the snapshot.
Verdict: It is powerful and clearly explained, and the docs rightly call it a convenience and not an isolation boundary.
The /browser connect Command
This command attaches Hermes to a running Chrome, Brave, Chromium or Edge window over the Chrome DevTools Protocol.
It only works in the terminal CLI, and it does not work from Telegram, Discord or the web interface.
Verdict: It is the best way to watch the agent work in a window you can see.
Passwords And Logins
When the agent reaches a sign-in page with no saved login, Hermes asks you in a masked prompt.
The password is encrypted on your machine, bound to that site, and filled into the page without the model seeing it.
Two-factor codes from an authenticator key can be generated automatically, and it can read logins from 1Password or Bitwarden if their command-line tools are signed in.
Verdict: It is the strongest part of the whole design.
Card And Address Fills
Saved cards and addresses are filled on their bound checkout site only.
Every card fill asks you first, and sessions with nobody present are refused.
Verdict: It is well guarded, and I'd still do the final payment click myself.
Lightpanda Engine
Lightpanda is a lightweight local browser engine that Hermes can start for you in Browser Use mode.
It has no graphical renderer, so screenshots are unavailable.
Verdict: It is a good fit for small servers doing text-first work.
Headed Mode And Session Recording
Headed mode opens a visible Chromium window and keeps it open between turns.
Session recording saves browser sessions as WebM video files when you switch it on.
Verdict: They are both useful while you are learning what the agent actually does.
Using It For Real Browser Tasks
The second meaning of Hermes Agent Browser Use is the activity itself.
Here are three task shapes with the approval step I'd add to each.
| Task | Example instruction | Approval step |
|---|---|---|
| Research | "Open these three pricing pages and build a table with the page address next to each price." | You spot-check the addresses. |
| Form filling | "Fill in this form with the details below, do not press submit, and show me a screenshot." | You press submit yourself. |
| Logged-in check | "Open my dashboard and report yesterday's totals without changing any settings." | You keep it read-only. |
The docs make one point I agree with.
For simple information retrieval, web search and web extract are faster and cheaper than a browser.
The browser earns its place when a page needs clicks, a login or dynamic content.
Hermes's approval system is built around dangerous shell commands, with smart, manual and off modes.
As far as I can tell from the docs, there is no prompt before each individual click.
So the stop instruction in your prompt is the approval gate for browser work.
🔥 Want to see how I test tools like this before I trust them? Inside the AI Profit Boardroom, I share my agent builds and test notes as I go, plus four coaching calls a week where 3,400+ members compare what is actually working. → Get access here
Costs
| Item | Published cost | Source |
|---|---|---|
| Hermes Agent | It is free. | Hermes docs |
| Browser Use mode, local browser | It is free. | Hermes docs |
| Browser Use Cloud session | It is $0.02 per browser-hour, with a one-minute minimum. | Browser Use pricing page |
| Managed residential proxy | It is $5 per GB. | Browser Use pricing page |
| Direct or own-proxy traffic | It is $0.20 per GB. | Browser Use pricing page |
| Nous Subscription browser | It is pay as you use against a paid plan. | Hermes docs |
Browser Use says it has no subscriptions and that top-up credits never expire.
Two of its pages gave different figures for the free signup credit when I checked, so I've left that out.
Your model bill sits on top of all of this.
Limits And Safety
| Limit | Why it matters |
|---|---|
| Weak models gain little | Nous's notes say a small local coder burned the savings retrying its own code. |
| The benchmark was narrow | Nous ran it on simple practice sites with three or fewer runs per cell. |
| Text-based reading | The built-in tools rely on a text snapshot of the page and not on pixels. |
| Snapshot size | Snapshots are cut at 15,000 characters by default, with the full copy saved to a file. |
| No file downloads | The docs list browser file downloads as unsupported for the built-in tools. |
| CAPTCHAs | Browser Use says no browser setup guarantees every challenge is avoided or solved. |
| Profile sync | Browser Use says cloud profile sync moves cookies but not local storage or extensions. |
| Real logins | A page the agent visits in real profile mode runs with your real accounts. |
My safety rules are short.
- Use a clean profile by default.
- Keep
approvals.modeon smart or manual. - Write a stop before any submit, send, pay or delete.
- Use the website blocklist for anything the agent must never open.
- Follow each site's terms on automated access.
If you want a second opinion on your own setup and your search plan, you can book a free SEO strategy session with my team.
What About MCP, The Standalone CLI And Jev Ultrafast?
Browser Use documents two more routes.
The first is the standalone CLI, installed with uv tool install browser-use and registered with browser-use skill install.
Hermes then drives it through its terminal tool.
The second is MCP.
Browser Use runs a hosted MCP server at https://api.browser-use.com/v3/mcp, and a local one that starts with uvx --from 'browser-use[cli]' browser-use --mcp.
Hermes can add a custom MCP server with hermes mcp add.
I have not tested that pairing, and Browser Use's MCP page names Claude, Cursor and Windsurf but not Hermes.
Jev Ultrafast is a third thing again.
It is Browser Use's separate agent that lets TypeSafe's Jev model choose every click, and I reviewed it in my Jev Ultrafast GitHub verdict.
My Scores
These scores are my judgement from the docs, the code in my install and the published numbers.
They are not measured results.
| Category | Score | Why |
|---|---|---|
| Ease of setup | 9/10 | It is the default, and a local browser needs no key. |
| Features | 9/10 | It covers local, cloud, real profile, a credential vault and named sessions. |
| Safety design | 8/10 | The vault and card prompts are strong, but browser clicks are not individually approved. |
| Value | 9/10 | Local use is free, and the cloud price is low and simple. |
| Model flexibility | 6/10 | It rewards strong models and gives little to weak ones. |
| Documentation | 9/10 | Both companies document the integration, and they agree with each other. |
| Overall | 8.3/10 | It is the route I'd choose for browser work in Hermes. |
Who Should Use It
- You should use it if you already run Hermes and want browser tasks with the least setup.
- You should use it if you run long, multi-step browser jobs on a strong model.
- You should add Browser Use Cloud if a site blocks your local browser.
- You should stay on the built-in tools if you run a small local model or a chat bot without terminal access.
For help choosing the browser underneath, see my best browser for Hermes ranking.
🔥 Not sure which agent tools are worth your time? Bring your shortlist to one of the four weekly coaching calls inside the AI Profit Boardroom, where 3,400+ members are testing these tools on real work. → Join the AI Profit Boardroom
Related Reading
- Hermes Agent browser is the capability explainer for the whole browser toolset.
- Best browser for Hermes ranks the browsers and backends.
- Jev Ultrafast GitHub is my verdict on Browser Use's Jev-powered agent.
- Best Hermes Agent tools ranks the toolsets I rely on most.
Also On Our Network
- 🌐 Set up Hermes with Browser Use in five steps
- 🌐 Seven browser jobs Hermes can take off your plate
- 🌐 Where Browser Use sits in an Agent OS
- 🌐 Model picks for browser tasks, backed by GoldieBench scores
FAQ: Hermes Agent Browser Use
Is Hermes Agent Browser Use an official integration?
Yes, it is an official integration.
The Hermes docs describe Browser Use mode as the default browser driver, and Browser Use's own site says Hermes Agent ships Browser Use as its default browser driver.
It is built into Hermes and is not a plugin or an MCP server.
What is browser_exec in Hermes Agent?
It is the single browser tool the model gets in Browser Use mode.
The agent passes it Python code, and that code clicks, types, scrolls and scrapes in the browser.
It replaces the separate browser_navigate, browser_click and browser_type tools while the mode is active.
Do I need a Browser Use API key for Hermes?
You only need one for Browser Use's own cloud browsers.
Browser Use mode on a local browser needs no key.
A paid Nous Portal subscription can also run Browser Use cloud browsers without a separate Browser Use account.
How do I turn Browser Use mode off in Hermes?
Type /browser use off in a chat, or set browser.backend to off in the Hermes config file.
Hermes then goes back to the built-in browser tools.
Camofox setups use the built-in tools automatically.
Does the AI model see my passwords?
The Hermes docs say it does not.
Saved logins are encrypted on your machine, bound to one site, and filled into the page without the password entering the model's context.
The site you type it into still receives it, exactly as it would from you.
Is Jev Ultrafast part of Hermes Agent?
No, it is not part of Hermes Agent.
Jev Ultrafast is a separate open-source browser agent from Browser Use that uses TypeSafe's Jev model to choose each click.
It is related because the same team builds it, but it is not wired into Hermes.
📺 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 AI entrepreneur, SEO expert, and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation, and SEO.
- I have 400,000+ YouTube subscribers who watch my AI tool tests every week.
- I built a 7-figure agency from the ground up.
- I run daily AI training inside the Boardroom.
- I wrote "Link Building Mastery", my book on link building strategy.
→ Get my best AI training inside the AI Profit Boardroom
Final verdict: the integration is real, it is on by default, and it is safest when you add your own stop points.
Pair it with a strong model and a clean profile, and you'll get the best out of Hermes Agent Browser Use.
