Give AI Eyes to See and Fix Your UI
Write YAML test suites, capture screenshots with Playwright, and let AI diagnose visual regressions — then tell you exactly how to fix them. Works with websites, web apps, Chrome extensions, WordPress plugins, and anything with a UI.
The Test-Diagnose-Fix Loop
No test frameworks to learn. Write YAML, run it, AI tells you what broke and how to fix it. Repeat until green.
Write YAML
Describe what your UI should look like in plain YAML. Navigate pages, click buttons, fill forms, take screenshots. Works for any URL — your website, admin panel, app, or localhost.
name: checkout-flow
type: web-app
base_url: https://mysite.com
flows:
- name: cart-page
steps:
- action: navigate
url: /cart
- action: screenshot
name: cart-view
- action: assert_element
selector: "#checkout-btn"AI Sees Your UI
Playwright captures pixel-perfect screenshots and diffs them against baselines. When something changes, AI vision analyzes the screenshot — not just pixels, but what the user actually sees.
npx vision-test-harness run checkout-flow
# AI sees: button shifted, text truncated,
# image missing, layout broken on mobile...
# Things pixel diff alone can't explain.AI Tells You the Fix
Claude reads your screenshot, console errors, and source code together. It doesn't just say what changed — it tells you WHY it broke and exactly which file and line to fix.
❌ cart-page: checkout button shifted 2px
Root cause: margin-top override in checkout.module.css:47
Fix: Remove line 47 (added in commit abc123)
→ You fix it → Re-run → ✅ PassingWhat You Get
A complete visual testing toolkit — free and open source. Add AI diagnosis when you want a second pair of eyes.
YAML-First Test Suites
Define test flows in readable YAML — no code to write. 16 built-in step types: navigate, click, type, wait, assert, screenshot, compare, evaluate, and platform-specific steps for WordPress and Chrome extensions.
Playwright Screenshot Engine
Headless Chromium captures pixel-perfect screenshots at any viewport size. Pixel-diff comparison via pixelmatch with configurable thresholds. Catches visual regressions between deployments.
AI Visual Diagnosis (BYOK)
When a test fails, AI analyzes the screenshot alongside console logs and your source code. Gets you from 'something looks wrong' to 'here's the fix' in one step. Works with any AI key — Anthropic, OpenAI, or Gemini.
MCP Server for AI Agents
Runs as a Model Context Protocol server — any MCP-compatible AI assistant (Claude Code, Cursor, etc.) can trigger test runs, read results, and iterate on fixes without leaving the coding flow.
Privacy Overlay for Screenshots
Capture marketing screenshots from real browser sessions. Built-in presets replace real user data (names, emails, subjects) with demo content while preserving your app's UI elements.
Platform-Specific Testing
Built-in step types for WordPress (wp_login, wp_activate_plugin, wp_navigate_admin) and Chrome extensions (open_side_panel, wait_for_content_script). More platforms coming.
This page was tested by the tool it describes.
Every screenshot in our README was captured by Vision Test Harness running against its own product page — 5 flows, 31 steps, all passing. The HTML report you see in the docs? That's the tool's output from testing itself. A visual testing tool that tests and documents itself is the best proof that it works.
If It Has a UI, You Can Test It
Any URL that Playwright can load — public sites, localhost, staging environments, admin panels, authenticated flows.
Websites & Landing Pages
Catch layout shifts, broken images, and responsive bugs across viewport sizes before they go live.
Web Applications
Test multi-step flows — login, navigate, interact, assert. Verify that your SPA renders correctly after every deploy.
Chrome Extensions
Test sidebar panels, popup UIs, and content script injections. Built-in steps for side panel and content script lifecycle.
WordPress Plugins
Log in to wp-admin, activate plugins, navigate settings pages, and assert notices. Purpose-built step types for WordPress.
E-Commerce & Shopify
Navigate admin iframes, verify storefront rendering, test product pages and checkout flows at multiple breakpoints.
Marketing & QA Screenshots
Capture polished screenshots from real browser sessions with privacy overlays that replace user data with demo content.
Enable AI Diagnosis
AI-powered screenshot analysis works with any LLM key you already have. Set an environment variable and it just works — no account, no subscription.
# Option 1: environment variable (recommended)
export ANTHROPIC_API_KEY=sk-ant-...
export OPENAI_API_KEY=sk-...
export GEMINI_API_KEY=AIza...
# Option 2: save key locally
vision-test-harness byok anthropic sk-ant-...
Open Source Core
The full CLI — Playwright screenshot engine, pixel diff comparisons, AI visual diagnosis, HTML reports, MCP server — is open source under MIT. No account. No subscription. Bring your own AI key (Anthropic, OpenAI, or Gemini) and get AI diagnosis for free.
Your UI deserves a second pair of eyes.
The free CLI catches visual regressions. AI diagnosis tells you why they happened and how to fix them. Open source core — no lock-in.
The Up Product Family
One platform. Your AI agents share context.
UpEdit
AI content agent — research, write, optimize, and publish across WordPress and any CMS
UpHelm
Operations command center — business health scores, anomaly detection, and CEO digests
UpInbox
AI email assistant — smart categorization, auto-drafts, and bulk cleanup that never leaves your browser
UpLens
Cross-product analytics — unified dashboards connecting all Up products, Stripe, Google Ads, and GA4
UpLift
Multi-platform ad intelligence — automated audits, waste detection, self-learning optimization
UpLook
AI visual testing — YAML-driven screenshot capture with AI diagnostics for websites, extensions, and plugins
UpRegulate
Regulatory intelligence and submission compliance checking
UpRoute
Voice & SMS triage and dispatch — maintenance requests, field service, and 24/7 bilingual coverage
UpSchedule
AI booking pipeline — scheduling, pre-meeting briefs, and post-meeting follow-ups
UpSeek
Pre-meeting intelligence — AI prospect research, CRM enrichment, and 90-second prep briefs
UpVault
Investor relations, data rooms, pipeline management, and fundraise readiness scoring
Every interaction makes your AI smarter. No manual retraining required.