Architectural Differences

A technical deep-dive comparing the architecture of OpenClaw (the base) and Steadybase (the fork).

Side-by-Side Comparison

Dimension	OpenClaw	Steadybase
Architecture	Monolith chat app	Multi-service platform with Temporal orchestration
Execution Model	Request/response	Durable workflows (survive crashes, retry failures)
State Management	In-memory (session)	Temporal Cloud durable state + hierarchical memory
AI Agents	1 (chat model)	8 specialized workers + Drew coordinator
LLM Providers	Single provider	Claude + GPT-4o + Gemini (routed by task)
Memory	Session-scoped	4-scope hierarchy (org/team/worker/session)
Authentication	Basic	JWT + invite codes + role-based access
API Surface	2-3 endpoints	20+ endpoints across 5 routers
Real-time	None	WebSocket for live updates
Mobile	None	iOS app (Capacitor)
Integrations	LLM API only	Temporal Cloud, VAPI, Slack, Gong, Salesforce
Security	Minimal	Auth middleware, rate limiting, audit logging, TLS, Helmet
Deployment	Local/simple	EC2 + PM2 + nginx → ECS Fargate (planned)

Architecture Diagram Comparison

OpenClaw Architecture

┌──────────────┐     ┌──────────────┐     ┌──────────────┐
│   Browser    │────►│   Express    │────►│   LLM API    │
│   (React)    │◄────│   Server     │◄────│   (single)   │
└──────────────┘     └──────────────┘     └──────────────┘

Steadybase Architecture

┌─────────────┐  ┌──────────┐  ┌────────┐
│  Dashboard  │  │  iOS App │  │  Slack │
└──────┬──────┘  └────┬─────┘  └───┬────┘
       │              │            │
       ▼              ▼            ▼
┌─────────────────────────────────────────┐
│  Express + WebSocket + Auth + Rate Limit│
│  ┌──────┐┌──────┐┌─────┐┌─────┐┌─────┐ │
│  │Brain ││Worker││Deals││Mem  ││Metr.│ │
│  │API   ││API   ││API  ││API  ││API  │ │
│  └──┬───┘└──┬───┘└──┬──┘└──┬──┘└──┬──┘ │
└─────┼───────┼───────┼──────┼──────┼─────┘
      │       │       │      │      │
      ▼       ▼       ▼      ▼      ▼
┌─────────────────────────────────────────┐
│         Temporal Cloud (mTLS)           │
│  6 Workflows + Memory Store             │
└─────────────────────────────────────────┘
      │              │             │
      ▼              ▼             ▼
  ┌────────┐   ┌────────┐   ┌────────┐
  │ Claude │   │ GPT-4o │   │ Gemini │
  └────────┘   └────────┘   └────────┘

Code Structure Comparison

OpenClaw

src/
  server.ts          # Express server + routes
  components/        # React chat UI

Steadybase

src/
  server.ts              # Express server with middleware
  api/
    brain.ts             # Brain chat API
    workers.ts           # Workers management API
    deals.ts             # Deal pipeline API
    memory.ts            # Memory hierarchy API
    metrics.ts           # Metrics and analytics API
  temporal/
    workflows/
      client-onboarding.ts
      lead-qualification.ts
      content-generation.ts
      ticket-resolution.ts
      memory-store.ts
      drew-coordinator.ts
    activities/
      llm-activities.ts
      data-activities.ts
      memory-activities.ts
  types.ts               # Shared TypeScript types

Dependency Comparison

Category	OpenClaw	Steadybase Additions
Runtime	Express, React	+ @temporalio/* (4 packages), ws, cors, helmet
AI	Single LLM SDK	+ @anthropic-ai/sdk, Claude CLI, GPT-4o, Gemini
Auth	Basic	+ aws-jwt-verify
Mobile	—	+ @capacitor/* (iOS)
DevOps	—	+ PM2, nginx configs

What This Means

The fork point represents a fundamental architectural shift:

OpenClaw is a chat application — it facilitates conversation with an AI
Steadybase is an agentic platform — it deploys AI workers to autonomously execute business processes

The codebase grew from a focused chat app (~2-3 core files) to a multi-service platform (~20+ modules) while maintaining the original TypeScript/React/Express foundation.

Architectural Differences

On this page