How to Set Up Hermes Agent Browser Automation

❌ `browser_exec` is missing or Hermes shows the built-in browser tools

✅ Run /browser use status in the interactive CLI. Browser Use mode requires a runnable Browser Use CLI and terminal access. A locked-down messaging profile intentionally retains the built-in tools.

❌ Parallel agents overwrite each other’s browser tabs or state

✅ Assign a unique session name to every browser_exec call. The unnamed default daemon is shared and is intended for one-at-a-time work.

❌ Browser tools are missing after setup

✅ Confirm Browser Automation is enabled for the active profile, then start a new Hermes session. Tool changes do not modify the current session's tool schema.

❌ Cloud browser credentials are rejected

✅ Check only the selected backend's key and project fields without printing them. Run a minimal public navigation before adding proxies, stealth, persistence, or scheduled jobs.

❌ `/browser connect` does nothing in Telegram, Discord, or Web UI

✅ That command is implemented only by the interactive CLI. Open hermes in a terminal, run /browser connect, and verify with /browser status.

❌ Chrome or Brave never opens port 9222

✅ Launch a separate browser process with --remote-debugging-port=9222 and a dedicated --user-data-dir. An existing normal browser process may otherwise open the new window without the debug port.

❌ A cloud browser cannot open localhost

✅ Install the local agent-browser sidecar and keep hybrid routing enabled. Hermes should route private and loopback URLs locally rather than sending them to the cloud provider.

❌ Camofox loses login state after every task

✅ Put managed_persistence: true under browser.camofox, fully restart Hermes, and verify the Camofox server version persists profiles by stable user ID. Hermes cannot force persistence if the server creates a fresh context every time.

❌ A page loads but the workflow still fails

✅ Inspect the final URL, accessibility snapshot, screenshot, native dialog state, and browser_console output. A transport-level page load does not prove the CTA, form, checkout, or app state worked.

❌ A CAPTCHA or protected flow blocks progress

✅ Use only authorized provider capabilities and access. If human interaction is required, capture the state and pause or report the blocker; do not guess success or bypass site rules.

❌ Browser Use or another cloud backend returns 502 after the session was idle

✅ Start a fresh browser task and test one public URL. Check provider status and quota, shorten work into checkpoints, and update Hermes if the install predates current expired-session cleanup.

❌ A local app or SearXNG works on the host but connection is refused in Docker

✅ Do not use container-local 127.0.0.1 for a host or sibling service. Use a shared-network service name or supported host alias and verify the URL from the Hermes runtime.

❌ The browser opens but clicks use missing or wrong elements

✅ Take a new accessibility snapshot after navigation, modals, or major DOM changes. Element references belong to the snapshot that created them.

❌ The visible page looks correct but the workflow still failed

✅ Inspect browser console errors and the final network/business result. Require a screenshot plus the expected created record, message, or confirmation before reporting success.

❌ Real-profile browsing opens signed out on macOS

✅ Update Hermes, start a new real-profile session, and test one harmless authenticated page. Current upstream launches the signed browser binary without mock-keychain flags and refreshes the snapshot automatically. Do not delete cookies or expose profile databases while diagnosing it.

Why does Hermes show browser_exec instead of separate browser tools?

Browser Use mode is the current default when Browser Use CLI 3.0 is runnable and terminal access is available. It combines browser work into one browser_exec tool. Use /browser use status to inspect the mode or /browser use off to force built-in tools.

How do I run two Hermes browser tasks at once?

Give every browser_exec call a different session name. Named sessions isolate the local harness daemon and cloud browser so parallel agents or chats do not share tabs and state.

What is Lightpanda used for?

Lightpanda is a fast, low-memory local browser engine designed for machine workloads. Hermes falls back to Chrome automatically when an action is not supported.

Do I need Browserbase for Hermes browser automation?

No. Hermes also supports Browser Use, Firecrawl, self-hosted Camofox, local Chromium through agent-browser, and your own Chrome, Brave, Chromium, or Edge over CDP. Paid Nous Portal subscribers can use Tool Gateway browser access without separate provider keys.

How do I use my logged-in Chrome or Brave session?

Use a dedicated browser profile, start Hermes in the terminal, and run /browser connect. Treat the profile as a credential and approve side effects explicitly.

Why does `/browser connect` not work in Web UI or Telegram?

It is an interactive-CLI slash command, not a gateway command. Gateway chats send that text to the model instead of changing the browser connection.

Can Hermes use a cloud browser and test localhost in one task?

Yes. Hybrid routing sends public URLs to the cloud provider and private or loopback URLs to a local Chromium sidecar by default.

How do I persist Camofox logins?

Set browser.camofox.managed_persistence: true, restart Hermes, and verify the Camofox server honors stable per-user profiles. The setting is profile-scoped and must be nested correctly.

Can Hermes replace Playwright tests?

Hermes can run Playwright-style natural-language QA and investigate failures, but deterministic CI suites should remain formal browser tests. Use Hermes where judgment, visual evidence, and cross-tool follow-up matter.

How do I keep browser automation safe?

Use read-only checks by default, isolate profiles, keep secrets out of prompts, gate side effects, respect access controls, and require URLs, screenshots, console output, or artifacts as proof.

What is the difference between /browser connect and browser.use_real_profile?

/browser connect attaches the interactive CLI to a dedicated live Chromium process over CDP. browser.use_real_profile creates and refreshes a Hermes-managed snapshot of an existing Chromium profile for local browser tasks without directly opening the everyday profile.