The hermes agent browser is a full browser automation toolset built into Hermes Agent. Per the Hermes docs, it lets your agent navigate websites, click and type on page elements, fill forms and pull out information, using a local browser or a cloud one. It's installed by default, so most people can use it on day one.
📺 Watch: New Hermes Browser Use Agents are 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
This guide covers what the browser tools do, how to set them up, which providers you can use, how Browser Use fits in, the safety settings worth knowing, and fixes for the common problems. Everything here comes from the official Hermes docs.
I run Hermes daily inside my own Agentic Operating System, with a separate profile per model. If you want the courses, the Agent OS zip and 4 coaching calls a week, join the AI Profit Boardroom ($69/mo).
What the Hermes Agent browser does
Hermes represents pages as accessibility trees, which are text snapshots a model can read easily. Interactive elements get ref IDs like @e1 and @e2, and the agent uses those to click and type. The docs list these built-in browser tools:
- browser_navigate opens a URL. It must be called before the other tools.
- browser_snapshot returns the accessibility tree, compact by default or the full page with full=true.
- browser_click and browser_type act on elements by ref ID. Typing clears the field first.
- browser_scroll, browser_press and browser_back handle scrolling, keys like Enter and Tab, and going back.
- browser_get_images lists images with URLs and alt text.
- browser_vision takes a screenshot and analyses it with vision AI, useful for complex layouts.
- browser_console reads console output and JavaScript errors, and can evaluate JavaScript.
- browser_cdp is raw DevTools access, only when a CDP endpoint is reachable.
- browser_dialog answers alerts, confirms and prompts.
Each task gets its own browser session, and idle sessions close automatically after a timeout. Want to know which browser backend to pick? I ranked them in the best browser for Hermes.
How to set up the Hermes Agent browser
Per the docs, the installers and hermes update install agent-browser and its pinned Chromium by default. The Browser Use engine is a regular Python dependency, so every install has it. Here are the steps:
- Check the toolset is on. The browser toolset must be in your config's toolsets list. You can enable it with hermes config set toolsets and include browser alongside hermes-cli.
- Pick a browser source. Run hermes tools and open Browser Automation. That writes your browser.cloud_provider choice.
- Add credentials if you chose cloud. Keys go in ~/.hermes/.env. The keys are credentials only, the selection in hermes tools decides which provider runs.
- Tune config.yaml if needed. The browser section sets inactivity_timeout (120 seconds by default), command_timeout (30), and record_sessions (off by default).
If you installed with --skip-browser, or the download failed, run hermes pm install agent-browser. Paid Nous Portal subscribers can skip keys entirely: new installs run hermes setup --portal, existing ones pick Nous Subscription in hermes model or hermes tools.
📺 Watch: Hermes AI Browser Agent: Build ANYTHING!
Hermes Agent browser providers
The docs list these browser sources:
| Provider | Type | Setup |
|---|---|---|
| Packaged Chromium (default) | Local | Installed with Hermes |
| Browser Use Cloud | Cloud | BROWSER_USE_API_KEY |
| Browserbase | Cloud | BROWSERBASE_API_KEY and BROWSERBASE_PROJECT_ID |
| Firecrawl | Cloud | FIRECRAWL_API_KEY |
| Nous Subscription | Cloud | Paid Nous Portal plan |
| Camofox | Local, self-hosted | CAMOFOX_URL |
| Lightpanda | Local engine | Binary on PATH |
| Your Chrome, Brave, Chromium or Edge | Local via CDP | /browser connect |
One handy feature: with a cloud provider set, Hermes auto-spawns a local Chromium sidecar for private addresses like localhost. So the agent can screenshot your local dashboard and scrape public sites in the same chat. The cloud provider never sees the private URL.
To use your own logins, the docs describe two routes. /browser connect attaches to your running browser from the interactive CLI. Or set browser.use_real_profile to true, and Hermes copies your active profile into a managed snapshot and drives that copy headless.
📺 Watch: Hermes Agent + Browser AI Agent is INSANE
Hermes Agent Browser Use
Yes, Hermes uses Browser Use, in two separate ways, and it's worth knowing the difference.
Browser Use mode is the default browser driver. Per the docs, it uses the Browser Use CLI 3.0 instead of the built-in tools. When browser.backend is unset, the agent gets one tool, browser_exec, and writes and runs Python in the browser to click, type, drag and scrape. Its engine is browser-harness, and if that's ever missing, Hermes falls back to the built-in tools and hermes update restores it.
Browser Use mode is a driver, not a browser. It drives whichever source you picked: Hermes' own headless Chromium, the Nous subscription browser, Browserbase, Firecrawl or Browser Use Cloud. The one exception is Camofox, which has no CDP endpoint, so it keeps the built-in tools.
Locally it uses the packaged Chromium, not your Chrome, so there's no remote-debugging popup. browser_exec also accepts a session name, and each name gets its own daemon and browser so parallel subagents don't clash. To opt out, run /browser use off or set browser.backend to off.
Browser Use Cloud is the hosted service. Add BROWSER_USE_API_KEY to ~/.hermes/.env from browser-use.com. Per the docs, it runs managed Chromium with stealth and residential proxies on by default, includes CAPTCHA solving, and supports persistent profiles for cookies, localStorage and saved passwords.
For a wider look at what else to switch on, see the best Hermes Agent tools. If you'd like a hand planning this for your own business, book a free AI strategy session.
Safety notes from the docs
- Terminal access gate. Because Browser Use mode runs model-written Python on your machine, browser_exec is only offered to sessions that also have terminal access.
- SSRF protection. Browser and web tools block private networks, loopback, link-local and cloud metadata addresses by default. Redirects to internal addresses are blocked too.
- Website blocklist. security.website_blocklist lets you block domains for web and browser tools. It's off by default.
- Real profile is not isolation. The docs call use_real_profile a consent-gated convenience. A page the agent visits runs with your real logins, so only turn it on when you want the agent acting as you.
- JavaScript evaluation. browser_console evaluation is unrestricted by default. Set browser.restrict_evaluate to true for a denylist when browsing hostile pages with a logged-in profile.
- Dialogs. The default dialog_policy, must_respond, waits for the agent and auto-dismisses after 300 seconds so a page can't stall forever.
Troubleshooting the Hermes Agent browser
- /browser connect does nothing in Telegram or Discord. It's an interactive CLI command only. Start Hermes from the terminal and run it there.
- Port 9222 never opens. On Chrome 136+, the debug flag is ignored on the default profile. Launch the browser with a dedicated user-data-dir.
- Real profile opens signed out. A running browser can lock the login databases. The docs say to quit the browser fully before launching a real-profile session.
- Browser tools fail in Docker. Playwright needs shared memory. Add --shm-size=1g to your docker run command, and give it at least 2 GB of RAM with browser tools active.
- Camofox keeps losing logins. Set browser.camofox.managed_persistence at that nested path, not top level, then fully restart Hermes.
- Snapshots get cut off. Large pages truncate at 15,000 characters by default. Raise browser.snapshot_threshold and restart the session.
- Private URLs blocked. If you turned off auto_local_for_private_urls, localhost is rejected unless you allow private URLs.
The docs also note limits: the browser can't download files, and cloud sessions use provider credits. If you need control of desktop apps, see Hermes computer use. My Agent OS guide shows how the browser fits into a full agent setup, and Goldie Bench tracks how models perform.
FAQ
Does Hermes Agent use Browser Use?
Yes. Browser Use mode, built on the Browser Use CLI 3.0, is the default browser driver per the Hermes docs, and Browser Use Cloud is a supported cloud provider with its own API key.
Do I need an API key for the Hermes Agent browser?
No, not for local browsing. The packaged Chromium works out of the box. Cloud providers like Browser Use Cloud, Browserbase and Firecrawl need their own keys.
Can I watch the browser while Hermes works?
Yes. Set browser.headed to true for a visible window, or use /browser connect with your own browser. You can also set record_sessions to true to save WebM recordings.
How do I turn off Browser Use mode?
Run /browser use off, or set browser.backend to off in ~/.hermes/config.yaml. Hermes then uses the built-in browser tools.
Want agents that browse and work for you every day? Join the AI Profit Boardroom, or book a free AI strategy session and we'll plan your setup together.
