Skip to main content

​
Kamikaze Pods

Kamikaze Pods are Jataka’s intelligent UI testing engine. They run in isolated Kubernetes environments and use Vision AI to automatically locate UI elements, making tests self-healing and maintenance-free.

​
Architecture Overview

Isolated Execution

Each test runs in a fresh Kubernetes pod with no data leakage between tests.

Vision AI

Computer vision automatically locates UI elements without brittle selectors.

Self-Healing

Tests automatically adapt to UI changes without manual updates.

​
How Kamikaze Pods Work

​
1. Pod Spin-up

When a test is triggered, Jataka spins up a fresh Kubernetes pod: 
apiVersion: v1
kind: Pod
metadata:
  name: jataka-test-pod-abc123
spec:
  containers:
  - name: playwright
    image: jataka/playwright:v2.1.0
    resources:
      limits:
        cpu: "2"
        memory: "4Gi"
    env:
      - name: SF_LOGIN_URL
        value: "https://test.salesforce.com"

​
2. Vision AI Element Location

Instead of brittle CSS selectors, Kamikaze uses Vision AI: 
// Traditional brittle approach
await page.click('[data-testid="save-button"]');

// Jataka Vision AI approach
await page.click({
  vision: "red button with 'Save' text in top right"
});

​
3. Self-Healing Logic

When UI changes occur, Kamikaze automatically adapts:
1
Vision AI scans the page for the target element using multiple criteria
2
If primary selector fails, fallback strategies are tried automatically
3
Test continues without human intervention, and the new selector is learned

​
Vision AI Capabilities

​
Element Recognition

Kamikaze can identify elements by:
  • Visual appearance: Color, size, position
  • Text content: Labels, placeholders, values
  • Context: Surrounding elements and layout
  • Functionality: Buttons, inputs, links, dropdowns

​
Example: Complex Element Location

// Find the "Create Opportunity" button
await page.click({
  vision: {
    text: "Create Opportunity",
    type: "button",
    color: "blue",
    position: "top-right"
  }
});

​
Multi-language Support

Vision AI works across languages: 
// Works in any language
await page.click({
  vision: "primary action button with positive text"  // "Create", "Crear", "Créer", etc.
});

​
Test Configuration

​
Basic Test Definition

# jataka-tests/ui/create-opportunity.yml
name: "Create Opportunity Test"
url: "/lightning/o/Opportunity/list"
steps:
  - action: click
    target:
      vision: "New button"
  - action: fill
    target:
      vision: "Opportunity Name input"
    value: "Test Opportunity {{timestamp}}"
  - action: click
    target:
      vision: "Save button"
  - action: verify
    target:
      vision: "Success message"

​
Advanced Test Scenarios

name: "Complex Sales Process"
flows:
  - name: "Create Account and Opportunity"
    steps:
      - action: navigate
        url: "/lightning/o/Account/list"
      - action: click
        target: { vision: "New Account button" }
      - action: fill
        target: { vision: "Account Name input" }
        value: "Test Account {{uuid}}"
      - action: click
        target: { vision: "Save button" }
      - action: wait
        condition: { vision: "Related tab" }
      - action: click
        target: { vision: "Opportunities related list" }
      - action: click
        target: { vision: "New Opportunity button" }

​
Performance Optimization

​
Parallel Execution

Kamikaze can run multiple tests in parallel: 
jataka test --engine kamikaze --parallel 5

​
Smart Caching

Test data and screenshots are cached between runs: 
cache:
  screenshots: true
  test_data: true
  session_state: true

​
Resource Management

Pods are automatically scaled based on test load: 
scaling:
  min_pods: 2
  max_pods: 10
  target_cpu_utilization: 70

​
Integration Examples

​
GitHub Actions

- name: Run Kamikaze UI Tests
  run: |
    jataka test \
      --engine kamikaze \
      --parallel 3 \
      --fail-on-critical \
      --report-format junit \
      --output-file test-results.xml

​
Slack Integration

Get test results directly in Slack: 
integrations:
  slack:
    webhook: https://hooks.slack.com/...
    notifications:
      test_failure: "#qa-alerts"
      flaky_tests: "#dev-team"
      performance_issues: "#performance"

​
Best Practices

​
Troubleshooting

Common Issue: “Vision AI cannot locate element”Solution:
  1. Check if the element is visible in the viewport
  2. Try more specific vision descriptors
  3. Use the --debug-vision flag to see what Vision AI is detecting
Performance Tip: Use --headless mode for faster test execution in CI/CD environments.

​
Advanced Features

​
Custom Vision Models

Train custom vision models for your specific UI: 
jataka vision train --dataset ./ui-screenshots --model my-company-ui

​
A/B Testing Support

Test UI variations automatically: 
ab_test:
  variants:
    - name: "control"
      url: "/ui-version-a"
    - name: "variant"
      url: "/ui-version-b"
  success_metric: "conversion_rate"

​
Accessibility Testing

Kamikaze includes accessibility checks: 
accessibility:
  wcag_level: "AA"
  check_color_contrast: true
  check_screen_reader: true
  check_keyboard_navigation: true

​
What’s Next?

Try it now: Run jataka test --engine kamikaze --demo to see Vision AI in action!