Skip to content

OSTwin

The zero-agent operating system for composable AI engineering teams.

How It Works

OSTwin turns a markdown plan into a fully-orchestrated multi-agent execution. No custom agent code required.

PLAN.md ──► Engine ──► DAG ──► War-Rooms ──► Agents
│ │ │ │
│ "Build a game" │ Waves │ Channels │ Skills
│ │ of │ Memory │ MCP Tools
▼ ▼ epics ▼ Lifecycle ▼
Goal + Epics Dependency Isolated Role-based
+ Constraints resolution execution specialists

Three Axes of Agent Identity

Every agent in OSTwin is defined by three orthogonal axes. This is what makes agents composable rather than monolithic.

Identity (WHO)

Defined by role.json + ROLE.md. Sets personality, constraints, and communication style. An architect thinks differently than an engineer.

Expertise (WHAT)

Loaded from SKILL.md files on demand. Domain knowledge injected into context only when needed. Skills are portable across roles.

Execution (HOW)

MCP servers provide isolated tool access. Each war-room gets its own tool boundary. No agent can reach outside its sandbox.

Five Architectural Pillars

Quick Start

Write a plan, run it, watch agents collaborate:

Terminal window
# 1. Install OSTwin
git clone https://github.com/igot-ai/os-twin.git
cd os-twin && pwsh Install.ps1
# 2. Set your API key
export ANTHROPIC_API_KEY="sk-ant-..."
# 3. Write a plan
cat > my-plan/PLAN.md << 'EOF'
# My First Plan
## Config
provider: anthropic
model: claude-sonnet-4-20250514
## Goal
Build a REST API with authentication and tests.
## Epic 1: Auth System
### DoD
- JWT auth working with login/register endpoints
### Tasks
- TASK-001: Implement user model and auth routes
EOF
# 4. Run it
pwsh Engine.ps1 -PlanPath ./my-plan/PLAN.md

Explore the Docs