FAQ
Do I need to know how to code?
No. You describe what you want in plain language and your AI assistant builds it inside the guardrails. It helps to understand the Concepts, but you don't need to read or write the code yourself.
What does the first command do?
npx @rystad/create-rystad-dashboard my-dash --dashboard-id 300
It scaffolds a complete dashboard — React + Vite + TypeScript frontend and a
Rystad Flask backend — wired together with one shared dashboard id and the
gateway route, plus the guardrail files the AI needs. This umbrella CLI is the
default; @rystad/create-rystad-app (React frontend-only) and
@rystad/create-rystad-api (backend-only) scaffold just one half. Prefer
Angular? Use @rystad/create-rystad-angular-app (see below).
How do I run it — the first time and after?
On the first run, create-rystad-dashboard installs both sides and
auto-starts the whole stack with docker compose up once the local
prerequisites are ready (Docker, the registry images pulled via az acr login,
an aspose.license.key file, and a filled .env). If they aren't ready it
prints the one-time steps — see the generated RUN_LOCAL.md.
Every run after that is one command from the project folder:
cd my-dash
docker compose up
For a fast inner loop while building, run just the frontend dev server with
cd my-dash/frontend && pnpm dev.
Can I build in Angular instead of React?
Yes. Scaffold with the Angular CLI:
npx @rystad/create-rystad-angular-app my-dash --dashboard-id 300
You get a modern Angular dashboard (standalone components, signals, TanStack
Angular Query) with the same guardrails and gateway wiring — guided by the
guide-angular-mcp server. It's frontend-only today; pair it with
@rystad/create-rystad-api if you also need a backend. Run it with
npm install && npm start, and check it with npm run verify.
What is a dashboard id?
A single number that ties your frontend and backend together and identifies your gateway route. Your team assigns it; you pass it once at scaffold time.
What stops the AI from inventing an API?
The api-mcp server exposes the real gateway endpoints and OpenAPI specs. The AI queries them live, so it can only use endpoints that actually exist. See the three guardrail layers on the Why page.
What does "verify" mean?
Each Task Recipe ends in a verify gate — a real check such as a build or a
contract test. Work is only "done" when that check passes, not when the AI
claims success.
How do I know a build is safe?
Run it and confirm it's green. For React:
npm run build
For Angular, run the full gate with npm run verify (ng lint && ng build && ng test). A passing build means the code compiles and the contracts hold.
Where do the reference docs for endpoints and components live?
They are auto-generated from the MCP servers and appear under the
Generated reference section of this site (endpoints, recipes, guardrails for
guide-mcp, api-mcp, and backend-mcp).
Is there a way to stop the AI skipping the guardrails?
Yes — register the optional orchestrator (rails-only mode). Instead of trusting the AI to follow the recipe and run the checks, the orchestrator drives the steps in order and won't let a feature be marked done until the verify gate passes. See Concepts → The orchestrator.
Something looks wrong — what do I do?
Ask your AI assistant to re-check against the MCP servers and to run the verify
step again. If it wants to disable a guardrail to "move faster," don't — that's
a red flag (see Safe prompting). For setup and
tooling errors — 401, 404, a missing pnpm/uv, or Docker not starting —
see the Debugging page.