Commands - Bring AI Home

Run from redact/. Defined in package.json on course/redact-qvac.

App

Command	Runs	Writes
`npm run dev -- --port 3101`	Next.js dev server	`.next/`
`npm run build`	`next build --webpack`	`.next/`
`npm run start -- --port 3101`	Production server for the last build	—

Local AI

Command	Runs	Writes	Network
`npm run qvac:service`	`qvac-service/index.ts` on `127.0.0.1:4317`	`.qvac/`	Needed until both models are cached
`npm run qvac:ocr-smoke [image]`	One OCR pass on an image, prints JSON	`.qvac/`	Needed until the OCR model is cached
`npm run qvac:evaluate`	Labelled evaluation over four test images, prints JSON	—	Needed until both models are cached
`npm run qvac:offline`	One full analysis with remote TCP/UDP denied	—	Denied (macOS only)
`npm run qvac:service:offline`	The service with remote TCP/UDP denied	—	Denied (macOS only)

The service, qvac:evaluate and qvac:offline use Laya for suggestions. qvac:ocr-smoke defaults to public/test-files/redact/chat-private.png. qvac:offline and qvac:service:offline use macOS sandbox-exec. They fail on Linux and Windows.

Checks

Command	Runs
`npm run lint`	ESLint
`npm run typecheck`	`next typegen`, then `tsc --noEmit`
`npm test`	Vitest unit tests (87 at the final checkpoint)
`npm run test:watch`	Vitest in watch mode
`npm run test:e2e`	Playwright, desktop and mobile projects

npm run test:e2e needs a running app (REDACT_BASE_URL, default http://localhost:3101) and a Playwright browser. The live QVAC test runs only with REDACT_LIVE_QVAC=1 and the service running. Run typecheck and build one after the other, not in parallel: both write .next/.

Test files

Command	Writes
`npm run redact:test-files`	Regenerates `public/test-files/redact/`

The committed test files are already in place. Regenerating can change PDF bytes.