Skip to main content

Why Rystad AI-Safe Dashboards

You are an analyst. You know the data, the market, and the story you want to tell. You should not have to become a frontend engineer, a backend engineer, and a security reviewer just to ship a dashboard. This portal exists so you don't have to.

Your first project​

Scaffold a full-stack dashboard — the React frontend and the Flask backend, wired together under one shared dashboard id — with a single command:

npx @rystad/create-rystad-dashboard my-dash --dashboard-id 300

That's it. This umbrella command sets up both sides, the gateway route, and the guardrails an AI agent needs to help you build the rest safely — then you run the whole stack together with docker compose up.

First time on this machine? Do the one-time First-time setup first, so the command can reach the private Rystad feed (it uses the scoped @rystad/ name — the unscoped create-rystad-dashboard isn't on public npm).

The mission​

Let non-technical analysts build production Rystad dashboards with an AI assistant — without hallucinated APIs, invented UI components, or insecure backends. You describe what you want; the AI does the work; the guardrails keep it honest.

The problem​

AI coding assistants are powerful, but on their own they will happily:

  • Invent API endpoints that don't exist, so your dashboard calls nothing.
  • Hallucinate UI components and design tokens, so your dashboard looks off-brand and breaks in review.
  • Skip authentication and validation, so your backend leaks data or fails in production.
  • Confidently claim success without ever verifying the result.

For someone who can't read the generated code line-by-line, those failures are invisible until it's too late. That's the gap this system closes.

The three guardrail layers​

Every safe build rests on three layers working together. When all three agree, the AI stays inside the lines.

1. Static rules — instruction files​

Hand-written instruction files ship with every scaffolded project. They tell the AI the non-negotiables up front: house style, folder structure, what it is never allowed to invent. This is the constitution the agent reads before it writes a single line.

2. Live MCP queries — real endpoints and tokens​

The AI doesn't guess what exists — it asks. Model Context Protocol (MCP) servers expose the real gateway endpoints, OpenAPI specs, design tokens, and component templates. If an endpoint or token isn't returned by a live query, it doesn't exist, and the AI can't use it.

3. Task Recipes with a verify gate​

Complex work follows step-by-step Task Recipes. Each recipe ends in a verify gate that actually checks the result — the build compiles, the endpoint responds, the contract holds — before anything is called "done." No verify, no success.

The outcomes​

When the three layers hold, you get:

  • No hallucinated APIs. Every call maps to a real gateway endpoint.
  • On-brand UI. Components and tokens come from the design system, not the model's imagination.
  • Secure backends. Auth, validation, and logging are wired in by template, not by luck.
  • Verified done. "Done" means it built and passed the gate — not that the AI said so.
  • Analyst speed. You ship in plain language, and the guardrails do the code review you can't.

Where to next​

  • Quickstart — from zero to a running dashboard.
  • Concepts — the mental model behind the guardrails.
  • Safe prompting — how to talk to the AI so it stays safe.
  • FAQ — quick answers to common questions.
  • Gallery — dashboards built with the system.