🧠 Quality Intelligence Platform

An AI platform that crawls a web app, understands and remembers it (the App Brain), and generates accurate test cases, test data, and test scripts — with enterprise connectors and strict tenant isolation.

This repository is the Phase 1 foundation. It's organized so an AI coding agent (or a developer) can continue building phase by phase using your prompt packs.

🧩 What's here

quality-intelligence-platform/
├── README.md                  ← you are here
├── package.json               ← monorepo (npm workspaces)
├── docker-compose.yml         ← local Postgres
├── .env.example               ← copy to .env and fill in
├── docs/
│   ├── CONSTITUTION.md         ← the rules every agent must follow
│   └── ARCHITECTURE.md         ← how the pieces fit (plain English)
├── scripts/setup.sh           ← one-command local setup
└── packages/
    ├── core/                  ← shared types, config, tenant context
    ├── database/              ← data model + Zeta Core + tenant isolation (RLS)
    ├── app-brain/             ← knowledge store + GROUNDING (direct load)
    ├── crawler/               ← crawl + LOGIN + screen reader + screenshots
    ├── agents/                ← AI generation agents (test cases)
    ├── connectors/            ← GitHub + JIRA (read-only)
    ├── api/                   ← the backend (Fastify)
    └── web/                   ← the frontend (to build)

🛠️ The stack (and why)

• TypeScript everywhere — one language across crawler, backend, and (later) UI.
• PostgreSQL + Prisma — reliable database; Postgres Row-Level Security enforces tenant isolation at the lowest layer (the "sacred" rule).
• Playwright — the standard browser-automation "robot hands" for crawling. (Add Stagehand or Browser Use for smarter AI-driven exploration.)
• Fastify — a fast, simple backend.
• Anthropic SDK — the AI that generates tests (grounded in the Zeta Core).
• Direct loading, NOT RAG — because the workspace is closed/bounded, we load a project's knowledge directly. Vector search is added later only for oversized projects. (See packages/app-brain/src/grounding.ts.)

🚀 Getting started (local)

# 1. Prerequisites: Node 20+, Docker
cp .env.example .env          # then fill in ANTHROPIC_API_KEY etc.
npm install
npm run db:up                 # start Postgres
npm run db:generate           # build the database client
npm run db:migrate            # create the tables
# then apply Row-Level Security:
#   psql "$DATABASE_URL" -f packages/database/prisma/migrations/0001_tenant_isolation/migration.sql
npm run dev:api               # start the backend
curl localhost:4000/health    # -> {"ok":true,...}

Or run npm run setup to do most of the above in one step.

✅ How to keep building

Use your Phase 1 Prompt Pack. Each prompt maps to a part of this scaffold:

• Prompt 1 (data model) → packages/database
• Prompt 3 (crawler + login) → packages/crawler ⭐
• Prompt 6 (grounded test cases) → packages/app-brain + packages/agents
• Prompt 7 (connectors) → packages/connectors

When Phase 1 passes its Final Acceptance, move to the Phase 2–6 Prompt Packs.

🧭 The golden rules (full list in docs/CONSTITUTION.md)

1. Correctness first. 2. Start simple (direct loading, not RAG).
2. The Zeta Core is the moat. 4. Tenant isolation is sacred.
3. Ground, don't guess. 6. Human approves before tests are trusted.

🖥️ Running the frontend (the web app)

The @qip/web package is a Next.js app covering the Phase 1 loop: sign in → create project → paste URL & crawl → view screen map → review tests.

# in one terminal — the backend:
npm run dev:api                  # http://localhost:4000

# in another terminal — the frontend:
cd packages/web
cp .env.local.example .env.local
npm install
npm run dev                      # http://localhost:3000

Open http://localhost:3000, "sign in" with any user id + tenant id (Phase 1 identity stub), create a project, paste a URL, and watch the screen map fill in.