For the complete documentation index, see llms.txt. This page is also available as Markdown.

Platform

Overview of the platform’s main building blocks (projects, modules, entities, UI, dashboards, rules and permissions).

Madrix is a platform for designing, generating and operating enterprise apps with AI.

Use this page as a mind map. It helps you navigate the blocks and flows.

Shortcuts:

End-to-end flow (the “happy path”)

  1. Login to the platform.

  2. Open the Home.

  3. Enter Projects and create/select a project.

  4. Generate and evolve via AI Chat (with approval checkpoints).

  5. Test in the Preview with Run.

  6. Deploy with Publish (when you want Cloud).

If you want a step-by-step with ready prompts: Start here.

Interface map (where things are)

Login

Platform entry. After login, you land on the Home.

Home

Starting point. Usually you do:

  • Select a recent project.

  • Create a new project.

  • Open global settings (user/organization).

Projects

A Project is the “container” of your app. It stores:

  • Structure (modules, entities/components, fields, scripts).

  • Generated UI (screens, routes, menus).

  • Settings and environments.

Project Editor (top bar)

This is the area where you switch between generate and edit.

Common items:

  • Project name

  • Mode toggle: AI Chat ↔ Editor

  • Run: opens/refreshes the app Preview

  • Publish: deploys to Cloud environment (when configured)

  • Home: returns to Home

  • Notifications

  • User/Organization settings

User / Organization Settings

Where preferences and organization context live (e.g.: permissions, access, etc).

Selected language

Controls interface texts. Also interacts with translations like pt-BR, en, etc.

Essential concepts (short definitions)

  • Project: the entire app (code, data, settings and deploy).

  • Module: a functional domain (e.g.: Sales, Tasks, Finance).

  • Entity / Component: your main “object” (becomes a table in the database).

  • Field: entity attribute (type, required, unique, picklist, etc).

  • Relationship: connection between entities (1-N, N-N).

  • Script: backend business rule (e.g.: validation and automation).

  • UI: generated screens, routes, menus and forms.

  • Dashboard / Report: managerial view and analyses.

  • Permissions (RBAC): who can see/do what.

How AI generates systems in Madrix (mental model)

The flow tends to be this:

  1. You describe objective, entities and rules in the prompt.

  2. The platform generates a specification for you to approve.

  3. After approval, agents generate backend + UI + project resources.

  4. You test in the Preview and request incremental changes.

Practical tip: treat the specification as a “contract”. It reduces rework.

Common flows (when to use each mode)

AI Chat (large and iterative changes)

Use when you want to:

  • Create an app from scratch.

  • Add entire modules and entities.

  • Refactor rules or flows.

  • Request changes in “business language”.

Open: AI Chat Mode

Editor (“surgical” adjustments and fine control)

Use when you want to:

  • Adjust fields, relationships, menus and event scripts.

  • Understand the generated frontend scaffold.

  • Resolve details the chat did not handle well.

Open: Editor Mode

AWUs (Agentic Work Units)

An AWU is the consumption unit for agentic work. In practice, it grows with:

  • Scope of the request (number of modules/entities/screens).

  • Number of iterations (changes that rewrite a lot).

  • Complexity of rules, integrations and permissions.

Best practices to consume less:

  • Start with a smaller prompt and evolve in steps.

  • Approve the specification before generating.

  • Make changes in small, testable “batches”.

Limits (limits, best practices and anti-patterns)

Use as a guardrail to have more predictability.

Best practices:

  • Write prompts with objective + entities + rules.

  • Prefer consistent names. Avoid synonyms for the same thing.

  • Request changes in checklist format. This reduces ambiguity.

  • Test in the Preview right after each batch of changes.

Anti-patterns:

  • Requesting “a complete ERP” in a single prompt.

  • Mixing many domains in the same first step.

  • Changing names and structures “in bulk” without validating impact.

  • Deploying without validating the Preview and basic permissions.

Next steps

Last updated

Was this helpful?