kazuph/agent-browser

Browser Automation with agent-browser

Positioning

agent-browser has multiple engines. Treat Lightpanda as one option, not the default answer.

• Use --engine lightpanda when an AI should quickly probe behavior first:
  • page reachability
  • DOM structure
  • interactive elements
  • console/errors
  • network requests
  • simple multi-session or batch checks
• Do not use Lightpanda for:
  • final E2E judgment
  • visual verification
  • trustworthy screenshots as evidence
  • layout regressions
  • media generation/download completion checks
  • flows where the user needs "it definitely worked" proof

If the task needs rendered output, screenshots, downloads, or user-facing proof, switch to a real browser engine or another real-browser tool such as playwright-cli.

Engine selection

Real browser / Chrome family

Use this when fidelity matters.

# Set environment variable before running commands
export AGENT_BROWSER_EXECUTABLE_PATH="/Applications/Google Chrome Canary.app/Contents/MacOS/Google Chrome Canary"

Or specify per-command:

agent-browser --executable-path "/Applications/Google Chrome Canary.app/Contents/MacOS/Google Chrome Canary" open <url>

Note: If you change --executable-path, run agent-browser close first to restart the daemon with the new browser.

Lightpanda

Use this only for a quick first pass driven by the AI.

agent-browser --engine lightpanda open <url>
agent-browser --engine lightpanda snapshot -i -c
agent-browser --engine lightpanda console
agent-browser --engine lightpanda errors
agent-browser --engine lightpanda network requests

Important:

• A Lightpanda screenshot may be a placeholder rather than a rendered page.
• Success in Lightpanda means "the AI could inspect DOM/state/network", not "the user-visible workflow is verified".

Quick start

agent-browser open <url>        # Navigate to page
agent-browser snapshot -i       # Get interactive elements with refs
agent-browser click @e1         # Click element by ref
agent-browser fill @e2 "text"   # Fill input by ref
agent-browser close             # Close browser

Core workflow

1. Navigate: agent-browser open <url>
2. Snapshot: agent-browser snapshot -i (returns elements with refs like @e1, @e2)
3. Interact using refs from the snapshot
4. Re-snapshot after navigation or significant DOM changes

Recommended first-pass workflow with Lightpanda

Use this only when the goal is "let the AI try the flow once and inspect what happens".

1. Start with agent-browser --engine lightpanda open <url>
2. Inspect snapshot -i -c
3. Check get title, get url, console, errors, and network requests
4. Try the minimum interaction needed to see whether the flow advances
5. Stop early and switch tools if the page needs rendered proof

Use this workflow for:

• quick smoke checks on internal apps
• confirming that a route loads
• checking that a button exists and triggers requests
• comparing local vs preview in separate sessions
• AI-led exploratory probing before deeper verification

Do not keep pushing Lightpanda once the task becomes visual or proof-oriented.

When using Lightpanda through agent-browser

Prefer this decision rule:

1. Try Lightpanda first only if you want a fast AI-only behavior check
2. Stay in Lightpanda only while DOM, text, and network are enough to answer the question
3. Switch immediately if any of these happen:
   • screenshot output looks like a placeholder
   • the page depends on rich rendering or embedded app shells
   • the UI shows repeated Error without a clear DOM-level explanation
   • the task requires download confirmation
   • the task requires generated media confirmation
   • the task requires evidence to show another human

If a wrapper page blocks useful inspection, open the direct app URL instead of staying on the shell page.

Commands

Navigation

agent-browser open <url>      # Navigate to URL
agent-browser back            # Go back
agent-browser forward         # Go forward  
agent-browser reload          # Reload page
agent-browser close           # Close browser

Snapshot (page analysis)

agent-browser snapshot        # Full accessibility tree
agent-browser snapshot -i     # Interactive elements only (recommended)
agent-browser snapshot -c     # Compact output
agent-browser snapshot -d 3   # Limit depth to 3

Interactions (use @refs from snapshot)

agent-browser click @e1           # Click
agent-browser dblclick @e1        # Double-click
agent-browser fill @e2 "text"     # Clear and type
agent-browser type @e2 "text"     # Type without clearing
agent-browser press Enter         # Press key
agent-browser press Control+a     # Key combination
agent-browser hover @e1           # Hover
agent-browser check @e1           # Check checkbox
agent-browser uncheck @e1         # Uncheck checkbox
agent-browser select @e1 "value"  # Select dropdown
agent-browser scroll down 500     # Scroll page
agent-browser scrollintoview @e1  # Scroll element into view

Get information

agent-browser get text @e1        # Get element text
agent-browser get value @e1       # Get input value
agent-browser get title           # Get page title
agent-browser get url             # Get current URL

Screenshots

agent-browser screenshot          # Screenshot to stdout
agent-browser screenshot path.png # Save to file
agent-browser screenshot --full   # Full page

For Lightpanda, treat screenshots as diagnostic output only. Do not use them as visual proof.

Wait

agent-browser wait @e1                     # Wait for element
agent-browser wait 2000                    # Wait milliseconds
agent-browser wait --text "Success"        # Wait for text
agent-browser wait --load networkidle      # Wait for network idle

Semantic locators (alternative to refs)

agent-browser find role button click --name "Submit"
agent-browser find text "Sign In" click
agent-browser find label "Email" fill "[email protected]"

Example: Form submission

agent-browser open https://example.com/form
agent-browser snapshot -i
# Output shows: textbox "Email" [ref=e1], textbox "Password" [ref=e2], button "Submit" [ref=e3]

agent-browser fill @e1 "[email protected]"
agent-browser fill @e2 "password123"
agent-browser click @e3
agent-browser wait --load networkidle
agent-browser snapshot -i  # Check result

Example: Authentication with saved state

# Login once
agent-browser open https://app.example.com/login
agent-browser snapshot -i
agent-browser fill @e1 "username"
agent-browser fill @e2 "password"
agent-browser click @e3
agent-browser wait --url "**/dashboard"
agent-browser state save auth.json

# Later sessions: load saved state
agent-browser state load auth.json
agent-browser open https://app.example.com/dashboard

Sessions (parallel browsers)

agent-browser --session test1 open site-a.com
agent-browser --session test2 open site-b.com
agent-browser session list

With Lightpanda, sessions are especially useful for fast local-vs-preview comparisons:

agent-browser --engine lightpanda --session local open http://localhost:3000
agent-browser --engine lightpanda --session preview open https://preview.example.com
agent-browser --engine lightpanda --session local snapshot -i -c
agent-browser --engine lightpanda --session preview snapshot -i -c

JSON output (for parsing)

Add --json for machine-readable output:

agent-browser snapshot -i --json
agent-browser get text @e1 --json

Debugging

agent-browser open example.com --headed  # Show browser window
agent-browser console                    # View console messages
agent-browser errors                     # View page errors

For Lightpanda-first investigation, also check:

agent-browser --engine lightpanda network requests
agent-browser --engine lightpanda network response <id>
agent-browser --engine lightpanda dashboard