Open Source · Apache License 2.0

How AppWithAI Builds Your Application

A sophisticated multi-agent AI pipeline, orchestrated by Mastra.ai, transforms your business description into production-ready software — with a mandatory human review gate before any code is generated.

For the reasoning behind this design — and why the engineering standards belong in the platform rather than in each project — see why AppWithAI exists.

The AI Pipeline

Seven specialized AI agents work in sequence to transform your description into a deployable application.

You describe your business process in plain English
Domain Analysis Agent (Mastra.ai + Claude / OpenAI / Ollama)
Understands your business domain; extracts core concepts, candidate entities, and their likely relationships
Entity and Relationship Agents
Refine each entity's attributes, data types, and validation rules; determine relationship cardinality
Human-in-the-Loop Review Gate
You review and approve the proposed data model in the visual ERD designer. No code is produced until you confirm.
Mermaid Agent (ERD Design)
Produces a complete Entity Relationship Diagram in Mermaid Crow's Foot notation, validated for correctness
Application Dictionary Generator
Generates the sys_ metadata tables that drive forms, validations, field layouts, and RBAC at runtime
Code Generation Engine (Handlebars Templates)
Produces production-ready TanStack Start frontend, NestJS backend, Kysely migrations, Playwright tests, and Docker configuration
Your Running Application — Ready to Deploy

Step-by-Step Process

From description to deployment in a guided, transparent process — every decision visible to you.

Step 1: Describe Your Business Process

Write your business process in plain English. No technical knowledge or coding required. Describe what your business does, what data you track, and what rules govern your processes.

Example Input

"We need a purchase order system. Vendors send us quotes, we create POs with line items for products and quantities. POs over $5,000 require manager approval; over $10,000 require finance director approval. We track delivery status and match invoices to POs before authorizing payment."

Step 2: AI Domain Analysis

The Domain Analysis Agent, powered by your choice of AI (Claude, OpenAI, or a local Ollama model), deeply analyzes your business domain. It extracts the core concepts and identifies the key entities relevant to your process.

Entities Detected

Vendor, PurchaseOrder, POLine, Product, Invoice, Delivery

Relationships Found

Vendor → PurchaseOrder → POLine → Product; Invoice → PurchaseOrder

Business Rules Identified

Approval thresholds, three-way matching, delivery tracking requirements

Step 3: Entity and Relationship Analysis

The Entity and Relationship Agents refine each detected entity — identifying all relevant attributes, data types, validation constraints, and the precise cardinality of each relationship.

Generated Entity Example: PurchaseOrder
Field Type Constraint
po_number String (auto-generated) Unique, required
vendor_id UUID Reference Foreign key → Vendor
order_date Date Required, default today
total_amount Decimal (currency) Calculated from line items
approval_status Enum Draft | Pending | Approved | Rejected

Step 4: Human-in-the-Loop Review

This is the critical differentiator of AppWithAI. Before generating a single line of code, the platform pauses and presents the complete data model for your review in the visual ERD designer. You can:

  • ✓ Confirm entities and attributes the AI identified correctly
  • ✓ Modify any field name, data type, or relationship directly in the Mermaid editor
  • ✓ Add entities or attributes the AI did not identify
  • ✓ Remove anything that does not apply to your requirements
  • ✓ See your changes reflected instantly in the React Flow graph visualization
Only after your explicit approval does the system proceed to generate code.

Step 5: ERD Generation

The Mermaid Agent produces a complete, validated Entity Relationship Diagram using Crow's Foot notation — showing all entities, attributes, primary keys, foreign keys, and relationships. This diagram serves as the authoritative blueprint for the generated application.

Step 6: Application Dictionary Generation

The Application Dictionary is generated — a metadata layer that controls how every form looks and behaves at runtime. This enables application administrators to reconfigure form layouts, field visibility, and validation rules without code changes.

sys_table

All data tables — purpose, type, access level, and configuration

sys_column

Every field in every table — type, constraints, validation, and display settings

sys_window / sys_tab

Application screens and their tab structure — what they show and how they're organized

sys_access_role / sys_access_field

Role-based access definitions — which roles access which tables and fields

Step 7: Code Generation

The Code Generation Engine processes the approved ERD and Application Dictionary through Handlebars templates, producing a complete, formatted, and deployable application. All generated code is standard TypeScript — no proprietary runtime required.

🎨
Frontend

TanStack Start with React 19, Shadcn UI, and TailwindCSS — server-rendered, type-safe

⚙️
Backend

NestJS 10+ with Fastify and Kysely — modular, DI-driven API server

🗄️
Database

PostgreSQL with full Kysely migration scripts and type-safe query builders

🧪
Tests

Playwright end-to-end tests auto-generated for every CRUD operation

🐳
Docker

Docker and Docker Compose configuration for all services

🔐
Security

Better-Auth integration with complete RBAC guards per entity

Step 8: Deploy and Iterate

Your application is ready to deploy on any cloud provider or on-premise environment. After initial deployment, administrators use the built-in admin interface to customize forms, rules, and workflows — no developer involvement required for ongoing configuration changes.

No Account · No Install · Ten Minutes

Try It Yourself

Describe your business to ChatGPT or Gemini Deep Research along with our machine-readable specification, bring back the one .mmd file it produces, and watch a complete application build and start in your browser. No account, no install, nothing sent anywhere.

Show Me the Three Steps → I Already Have a Model →

The Application Dictionary Advantage

A metadata-driven architecture — proven in enterprise ERP systems over decades — that makes every generated application configurable without developer involvement.

Why Metadata-Driven Architecture?

When application behavior is stored in database tables rather than hardcoded logic, administrators gain direct control over how the application looks and behaves — without waiting for development cycles.

Traditional vs AppWithAI

Requirement Traditional AppWithAI
Add a field to a form Hours of development Minutes in admin
Reorder form fields Code change + deploy Drag and drop
Change a validation rule Code + test + deploy Update config
Modify approval routing Change code + redeploy Update in admin
Create a new user role Code + configuration Create role + assign

Form-to-Form Relationships

The Application Dictionary defines not just individual forms but the relationships between them — master-detail views, linked lookups, and cross-document navigation.

sys_window: Purchase Order
├─ sys_tab: Header (Vendor, Date, PO Number)
├─ sys_tab: Line Items
│ └─ sys_field: Product, Quantity, Unit Price, Line Total
├─ sys_tab: Delivery Status
└─ sys_tab: Approval History

Resulting Behavior:
✓ Master-Detail relationships are rendered automatically
✓ Cross-document navigation is built in
✓ Lookup relationships are enforced throughout
✓ Administrators can change layout without code

Time-to-Value Comparison

AppWithAI dramatically compresses the timeline from requirements to production deployment.

Phase Traditional Development AppWithAI
Requirements and Design 4–8 weeks 1 day
Data Modeling 2–4 weeks Minutes (AI-generated, human-reviewed)
Form Development 8–16 weeks Hours
Business Rules Implementation 4–8 weeks Hours (visual decision flowcharts)
Workflow Configuration 4–8 weeks Hours
Testing and QA 4–8 weeks 1–2 weeks
Total Time-to-Value 26–52 weeks 1–3 weeks

AppWithAI delivers production-ready applications 90% faster than traditional development.

Ready to See It In Action?

Watch AppWithAI generate a complete business application from your description — live, in your meeting, in under 30 minutes.

Schedule a Live Demo