Browser automation resource

The browser: resource drives a real browser (Chromium, Firefox, or WebKit) via Playwright. Use it to navigate pages, fill forms, run JavaScript, capture screenshots, and maintain authenticated sessions across resources.

Native browser: is compiled in. A registry component of the same name (kdeps registry install browser) adds stealth, persistent sessions, and file upload. They are not the same thing.

Where it runs

Both workflow mode and agent mode. In workflow mode it executes as a DAG step. In agent mode, the workflow containing this resource runs as a single callable tool.

Basic usage

# resources/browse.yaml
actionId: captureTitle
name: Capture Page Title
browser:
  engine: chromium
  url: "https://example.com"
  actions:
    - action: evaluate
      script: "document.title"
apiResponse:
  success: true
  response:
    title: "{{ get('captureTitle') }}"

Configuration options

Option	Description	Default
engine	Browser engine: chromium, firefox, or webkit	chromium
url	Initial URL to navigate to	-
waitFor	CSS selector to wait for before running actions	-
headless	Run browser in headless mode	true
sessionId	Reuse a named persistent browser session	-
timeout	Global action timeout (e.g. "30s")	30s
viewport.width	Browser viewport width in pixels	1280
viewport.height	Browser viewport height in pixels	720
userAgent	Custom User-Agent string for the browser	(default Mozilla/5.0)
stealthMode	Enable anti-bot detection features	false
actions	Ordered list of browser actions	[]

Action types

15 action types are available: navigate, click, fill, type, upload, select, check, uncheck, hover, scroll, press, clear, evaluate, screenshot, wait. See the Browser Action Types Reference for the complete field-level documentation of each action.

Stealth mode

Enable stealthMode: true to evade bot detection on websites like LinkedIn that block headless browsers. Stealth mode configures the browser with anti-detection settings including:

• Disables AutomationControlled blink features
• Adds realistic viewport, locale (en-US), and timezone (America/New_York)
• Removes automation flags (--disable-blink-features=AutomationControlled)
• Sets a realistic User-Agent string

# resources/example.yaml
browser:
  engine: chromium
  headless: true
  stealthMode: true
  url: "https://www.linkedin.com/login"
  actions:
    - action: fill
      selector: "#username"
      value: "{{ get('email') }}"

For sites with sophisticated bot detection, also consider:

• Setting headless: false to run with a visible browser window
• Using a custom userAgent that matches your OS/browser
• Reusing sessionId with pre-authenticated sessions instead of logging in fresh

Persistent sessions

By default each resource runs in a fresh, ephemeral browser context. Set sessionId to share a browser context across multiple resources or API calls - cookies, local storage, and page state are preserved.

# Resource 1 - log in and save the session
browser:
  engine: chromium
  url: "https://app.example.com/login"
  sessionId: "user-session"
  actions:
    - action: fill
      selector: "#username"
      value: "{{ get('username') }}"
    - action: fill
      selector: "#password"
      value: "{{ get('password') }}"
    - action: click
      selector: "#login-button"

# Resource 2 - reuse the authenticated session
browser:
  engine: chromium
  sessionId: "user-session"
  actions:
    - action: navigate
      url: "https://app.example.com/dashboard"
    - action: evaluate
      script: "document.querySelector('.user-greeting').textContent"

Examples

Stealth mode login (bot detection evasion)

For websites that block headless browsers, enable stealthMode and consider using non-headless mode to appear more human-like:

# resources/linkedin-login.yaml
actionId: linkedinLogin
browser:
    engine: chromium
    headless: true
    stealthMode: true
    sessionId: "linkedin-session"
    url: "https://www.linkedin.com/login"
    waitFor: "#username"
    actions:
      - action: fill
        selector: "#username"
        value: "{{ get('linkedin_email') }}"
      - action: fill
        selector: "#password"
        value: "{{ get('linkedin_password') }}"
      - action: click
        selector: "button[type='submit']"
      - action: wait
        wait: "3000ms"

Form fill and submit

# resources/submit-form.yaml
actionId: submitForm
browser:
    engine: chromium
    url: "https://forms.example.com/contact"
    waitFor: "#name"
    actions:
      - action: fill
        selector: "#name"
        value: "{{ get('contact_name') }}"
      - action: fill
        selector: "#email"
        value: "{{ get('contact_email') }}"
      - action: fill
        selector: "#message"
        value: "{{ get('message') }}"
      - action: click
        selector: "#submit"
      - action: wait
        selector: ".confirmation-message"
      - action: evaluate
        script: "document.querySelector('.confirmation-message').textContent"
  apiResponse:
    success: true
    response:
      confirmation: "{{ get('submitForm') }}"

Screenshot of a dynamic dashboard

# resources/dashboard-shot.yaml
actionId: dashboardShot
browser:
    engine: chromium
    url: "https://dashboard.example.com"
    viewport:
      width: 1920
      height: 1080
    actions:
      - action: wait
        selector: ".chart-loaded"
      - action: screenshot
        outputFile: /tmp/dashboard.png
        fullPage: true
  apiResponse:
    success: true
    response:
      screenshot: "/tmp/dashboard.png"

Extract JavaScript-rendered data

# resources/extract-data.yaml
actionId: extractData
browser:
    engine: chromium
    url: "https://spa.example.com/products"
    waitFor: ".product-list"
    actions:
      - action: evaluate
        script: |
          Array.from(document.querySelectorAll('.product-item')).map(el => ({
            name: el.querySelector('.name').textContent.trim(),
            price: el.querySelector('.price').textContent.trim()
          }))
  apiResponse:
    success: true
    response:
      products: "{{ get('extractData') }}"

Multi-step login with persistent session

# resources/login-step.yaml
actionId: loginStep
browser:
    engine: firefox
    url: "https://secure.example.com/login"
    sessionId: "{{ get('session_id') }}"
    actions:
      - action: fill
        selector: "[name=username]"
        value: "{{ get('user') }}"
      - action: fill
        selector: "[name=password]"
        value: "{{ get('pass') }}"
      - action: click
        selector: "[type=submit]"
      - action: wait
        selector: ".dashboard"
      - action: evaluate
        script: "document.cookie"
  apiResponse:
    success: true
    response:
      cookies: "{{ get('loginStep') }}"

File upload

# resources/upload-document.yaml
actionId: uploadDocument
browser:
    engine: chromium
    url: "https://docs.example.com/upload"
    actions:
      - action: upload
        selector: "#document-input"
        files:
          - /tmp/document.pdf
      - action: click
        selector: "#upload-button"
      - action: wait
        selector: ".upload-success"
  apiResponse:
    success: true
    response:
      status: "uploaded"

Output

After execution the resource output contains the result of the last evaluate action (or the final page URL when no evaluate action is present). Access it with get('actionId').

# Capture and use document.title
actions:
  - action: evaluate
    script: "document.title"

# In a subsequent resource before block:
before:
  - "set('title', get('captureTitle'))"

Supported engines

Engine	Notes
chromium	Default. Chromium-based (Chrome/Edge compatible).
firefox	Gecko-based. Good for Firefox-specific testing.
webkit	WebKit-based. Useful for Safari compatibility testing.

Installation

Playwright browsers must be installed on the host running kdeps:

npx playwright install chromium
# or install all browsers:
npx playwright install

See also

• HTTP client - simple HTTP requests without a browser
• Scraper - text extraction from already-fetched pages
• Python - Custom Python automation scripts