Architecture Overview

MDWR combines workflow orchestration with operational decision memory.

High-Level Architecture

┌─────────────────┐
│  Developer SDK  │
│  (TypeScript/   │
│   Python)       │
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│  MDWR Control   │
│     Plane       │
│  (Spec Compiler)│
└────────┬────────┘
         │
         ▼
┌─────────────────┐      ┌──────────────┐
│  Motia Runtime   │      │  Mem0/Redis  │
│  (Event-Driven) │◄────►│  (Memory)     │
└─────────────────┘      └──────────────┘

Components

1. Developer SDK

Language-specific SDKs (TypeScript, Python) that provide:

• Workflow definition API
• Step handlers
• Memory interface
• Execution management

2. MDWR Control Plane

Converts workflows to canonical specification and deploys to Motia:

• Workflow compilation
• Step generation
• Deployment orchestration

3. Motia Runtime

Event-driven backend runtime:

• Step execution
• State management
• Event system
• Observability

4. Memory Layer

Decision storage and similarity search:

• Mem0 (production)
• Redis (local development)
• Automatic decision storage
• Similarity queries

Data Flow

1. Define Workflow: Developer creates workflow using SDK
2. Deploy: SDK compiles to canonical spec and generates Motia steps
3. Execute: Workflow runs through Motia event system
4. Store Decisions: Each decision automatically stored in memory
5. Query Memory: Future executions query similar decisions
6. Learn: Workflows improve based on historical outcomes

Integration Points

Motia Integration

• Workflows converted to Motia Steps
• Steps placed in motia-atlas/src/{workflow}/
• Uses Motia’s state management
• Leverages Motia’s event system

Memory Integration

• Automatic decision storage
• Similarity search via Mem0
• Outcome linking
• Cross-workflow learning

Next Steps

• Design Principles - Core principles
• Motia Integration - Integration details