01
Analyze
QAHub combines your PRD with live UI signals — visible inputs, buttons, console errors — to build a structured feature map.
PRD→feature map
AI-powered QA orchestration
Turn requirements into verified software.
QAHub reads your PRD, explores your running app, drafts a reviewed test plan, generates safe Playwright specs, and runs them in isolation — then hands you back a report with evidence and a first-pass failure diagnosis.
Project input
Target URL — https://app.acme.test
PRD — Users can log in, search, and check out…
Analyzing live app · 82%
Latest run
Login with valid credentials
Search returns relevant records
Checkout submits order
How it works
From PRD to evidence-backed test runs in three steps.
A typed pipeline you can trust — every stage produces structured output you can review, tweak, and replay.
02
Plan & review
A risk-ranked test plan is drafted. Cases without real selectors are marked as needing review so nothing silently passes.
plan→selected cases
03
Run & report
Selected cases are templated into safe Playwright specs, run locally or in a Docker sandbox, and returned with screenshots and a diagnosis.
run→report
Features
Built for engineers reviewing tests,
not marketing dashboards.
LLM + deterministic analyzer
Stories, archetype detection, and scenarios are produced by an LLM you configure — Claude, GPT, Gemini, OpenRouter, Ollama, MiniMax. Falls back to a deterministic pattern matcher when offline.
Agentic explorer
Playwright logs in with your test account, BFS-crawls authed routes, probes forms for validation, and captures XHR — so the plan reflects the real UI surface.
Review gate
Per-case verdicts (pass / needs revision / blocked) before anything runs.
Safe generator
Specs come from typed actions. Missing selectors fail loudly, never silently.
Isolated runners
local, dry-run, or Docker per task — multi-user safe.
Evidence-aware reports
Step screenshots, traces, stderr, and a first-pass failure diagnosis are aggregated into a structured report you can audit.
MCP server included
Coding agents (Claude Code, Cursor, Windsurf, Cline) connect via stdio or HTTPS — analyze and run tests from inside your editor.
Plays well with
Standards-based — no proprietary lock-in.
Playwright
Next.js
Postgres
Docker
MCP
OpenAPI
Ready to put your QA on a structured loop?
Spin up a workspace, point QAHub at your app, paste a PRD, and watch a reviewed test plan build itself in minutes.