An open-source, multi-agent AI framework for Playwright test automation — featuring a 2D pixel-art office, live SSE telemetry, and Gemini / Ollama LLM routing.
Open-QA is a multi-agent AI workspace where specialist AI agents collaborate to write, run, debug, and heal your Playwright test suite — all inside a real-time animated pixel-art office.
Type a request in the chat. Edi M, the Team Manager, analyses your intent and delegates to the right specialist automatically. Meanwhile the Playwright Dashboard streams live test output, and when a run finishes Edi M delivers a structured AI post-mortem with root-cause analysis and ready-to-paste fix code.
| Feature | Description | |
|---|---|---|
| 🏢 | 2D Pixel-Art Office | 7 specialist agents at animated desks — agents walk to their desk and type live while responding to chat, celebrate with checkmark bubbles on passing runs, show red ✗ bubbles on failures, and display handoff bubbles when passing work between agents in multi-turn conversations |
| 🤖 | Edi M — Team Manager | Orchestrator agent that analyses your request and routes it to the right specialist |
| 🤝 | Multi-Agent Collaboration | Tag two or more specialists (or @Edi M) and they iterate: primary drafts → critic reviews → primary refines → manager delivers the polished synthesis. Each turn streams into its own chat bubble with a role pill (primary / critic / synthesis). Chat history and image attachments are forwarded to the initial primary turn so vision prompts and follow-up context are preserved across collaborative rounds |
| ✨🦙 | Gemini + Ollama | Switch between cloud Gemini and fully-local Ollama models with a single click — no restart required |
| 🎭 | Playwright Dashboard | Run your test suite, stream live terminal output, and view pass/fail telemetry in real time. Five KPI cards (Total Executions, Passed, Failed, Flaky, Success Rate) each show a sparkline from the last 8 runs and a trend delta. A three-panel insights row surfaces Executions Over Time, Top Failed Tests, and a Failure Reasons donut chart (UI Text Change / Element Not Found / Timeout / Other). Cards pulse live during a run, updating counts from the SSE stdout stream. Every run also lands an Edi M post-mortem in chat. |
| 🧠 | Post-Run AI Summaries | After a test run Edi M reads the JSON results, classifies each failure (Timeout / Assertion / Locator / Network), and streams a structured executive summary with TypeScript fix snippets |
| 📜 | Persistent Run History + Logs | Each run is archived in a better-sqlite3 database (test-results/runs.db) with its full stdout — click any past run in the history table to replay results AND the full original log (no line-count truncation) |
| 🔧 | Self-Healing Locators | Broken selectors are ranked by resilience: getByTestId → getByRole → getByLabel → CSS/XPath with confidence scores and caveats |
| 🔬 | 6 CLI Agents | Standalone Node.js agents: Self-Healing, Auto-POM, Bug Triage, Visual Regression, A11y Scanner, Data Generator |
| 📚 | Built-in QA Course | 12-chapter automation curriculum with animated, interactive lessons built directly into the UI |
┌──────────────────────────────────────────────────────────────────────┐
│ Browser · React 18 + Vite 6 (port 5173) │
│ │
│ User types in chat │
│ │ │
│ ▼ │
│ ┌──────────┐ @-mention / route ┌─────────────────────────────┐ │
│ │ Edi M │ ──────────────────► │ Specialist Agent │ │
│ │ Manager │ │ E2E Tester · POM Architect │ │
│ └──────────┘ │ Bug Triager · A11y Expert │ │
│ │ CI Engineer · Locator Healer│ │
│ └────────────────┬────────────┘ │
│ SSE stream ◄──┘ │
└──────────────────────────────────────────────────────────────────────┘
│ POST /api/qa-agent
▼
┌──────────────────────────────────────────────────────────────────────┐
│ Express (port 3001) │
│ │
│ /api/qa-agent → Gemini API or Ollama (local) │
│ streams token chunks back via SSE │
│ │
│ /api/run-dynamic-test │
│ Phase 1 → spawn npx playwright test → stream stdout │
│ Phase 2 → read pw-results.json → Edi M AI summary │
│ classify failures → stream summary_chunk events │
└──────────────────────────────────────────────────────────────────────┘
- Node.js 20+
- npm 10+
- A Google Gemini API key or Ollama running locally
git clone https://github.com/MyNameIsEdi/open-qa.git
cd open-qa
npm install
cd ui && npm install && cd ..npx playwright install --with-deps chromiumOption A — Gemini (cloud):
cp .env.example .env
# Edit .env and paste your GEMINI_API_KEYOption B — Ollama (fully local, no key needed):
# Install Ollama from https://ollama.ai, then:
ollama pull qwen2.5-coder # or llama3.2, codellama, mistral, etc.
# No .env change needed — switch to Ollama in the app Settings panelnpm run devOpens at http://localhost:5173 — the Express API runs on http://localhost:3001.
The provider toggle is built into the UI — no server restart needed.
- Open the QA Office page (
/office) - Click ⚙ Settings in the top bar, or use the ✨ Gemini / 🦙 Ollama inline toggle in the chat header
- For Ollama: set the base URL (
http://localhost:11434) and model name, then save
Your choice is persisted to localStorage — it survives page refreshes.
open-qa/
├── server/
│ └── index.ts # Express API — /api/qa-agent, /api/run-dynamic-test, SSE
│
├── ui/
│ └── src/
│ ├── pages/ # React pages (OfficePage, PlaywrightDashboard, …)
│ ├── components/ # Shared UI components + animated lesson components
│ ├── context/ # SettingsContext — global agent config + chat state
│ ├── office/ # Pixel-art office engine (game loop, renderer, sprites)
│ └── App.tsx # Router + layout shell
│
├── src/
│ ├── agents/ # Standalone CLI agents
│ │ ├── self-healing/ # Self-healing locator agent
│ │ ├── auto-pom/ # Page Object Model generator
│ │ ├── visual-regression/# Pixel-diff visual regression
│ │ ├── visual-a11y/ # WCAG 2.1 AA accessibility scanner
│ │ └── explore/ # URL crawler / QA audit reporter
│ ├── skills/ # Reusable skill modules
│ │ ├── data-gen/ # Edge-case test data generator
│ │ └── log-analyzer/ # Automated bug triage
│ └── core/
│ └── llm-client.ts # Anthropic SDK wrapper (CLI agents)
│
├── tests/ # Playwright E2E test specs
├── examples/ # Runnable workflow examples
└── docs/ # Additional documentation
The six standalone agents in src/agents/ run independently from the UI:
| Command | Agent | What it does |
|---|---|---|
npm run heal |
Self-Healing | Recovers broken Playwright selectors — outputs ranked repair candidates |
npm run run:auto-pom |
Auto-POM | Generates a typed Page Object Model class from a live URL |
npm run run:bugreport |
Bug Triage | Turns raw error logs into a structured P0–P3 Jira-ready bug report |
npm run run:visual-regression |
Visual Regression | Compares screenshots and reports pixel-level diffs |
npm run run:visual-a11y |
A11y Scanner | Audits a URL for WCAG 2.1 AA violations with fix recommendations |
npm run run:datagen |
Data Generator | Creates edge-case payloads for API fuzzing and boundary testing |
All agents run in deterministic MOCK mode if no ANTHROPIC_API_KEY is present — perfect for CI and first-time exploration.
| QA Office — Pixel-Art Map | Playwright Dashboard |
 |
 |
| Home — Agent Marketplace | Autonomous Agents |
 |
 |
# Run the full Playwright suite (headless)
npm test
# Run with the HTML reporter open
npx playwright test --reporter=html
# Type-check everything
npm run typecheck # server + root
cd ui && npx tsc --noEmit # UIContributions are welcome! Please read CONTRIBUTING.md before opening a pull request.
Key guidelines:
- All PRs must pass
npm run typecheckandnpm run lint - Follow the existing TypeScript strict-mode patterns
- Agent additions go in
src/agents/<name>/index.ts - UI changes live in
ui/src/
See docs/INTEGRATION_TESTING.md for patterns on combining multiple agents into workflows.
MIT © Open-QA Contributors
Built with ❤️ using React · Vite · Express · Playwright · Gemini · Ollama