Multi-agent orchestrator for Claude Code. Spawn and coordinate hundreds of AI agents in parallel via a single CLI. No API key required, powered by your Claude Code subscription. Ideal for developers, teams, and anyone who wants to multiply their AI coding capacity.

---

Demo

# Spawn 4 parallel agents in seconds
oa run "Research React 19 new features" --name researcher --model claude/sonnet --direct
oa run "Write unit tests for auth module" --name tester --model claude/sonnet --direct
oa run "Generate OpenAPI docs from bridge.py" --name documenter --model claude/haiku --direct
oa run "Review PR #42 for security issues" --name reviewer --model claude/opus --direct

# Watch them work
oa status

┌─────────────┬────────┬─────────┬──────────────────────────────┐
│ NAME        │ MODEL  │ STATUS  │ TASK                         │
├─────────────┼────────┼─────────┼──────────────────────────────┤
│ researcher  │ sonnet │ running │ Research React 19 new...     │
│ tester      │ sonnet │ running │ Write unit tests for auth... │
│ documenter  │ haiku  │  done   │ Generate OpenAPI docs...     │
│ reviewer    │ opus   │ running │ Review PR #42 for security.. │
└─────────────┴────────┴─────────┴──────────────────────────────┘

---

Features

🤖 1612+ Agent Templates	112 categories from code-dev to AEC to healthcare
⚡ Parallel Execution	Spawn hundreds of agents simultaneously in tmux
🔄 Bidirectional Feedback Loop	Agents report back; orchestrator iterates
🐳 Pipeline Orchestration	Planner → parallel Workers → Combiner, automated
📊 Three Interfaces	Terminal CLI, Textual TUI dashboard, React web UI
🔌 MCP + Agent Messaging	Integrate with any MCP client; inter-agent DMs
🧬 Nested Spawning	Agents create child agents up to 6 levels deep
🌐 Multi-model	Claude Opus / Sonnet / Haiku + local Ollama models

---

Quick Start

# Install
./install.sh            # Linux / macOS / WSL2

# Start a session
oa start

# Spawn your first agent
oa run "Write a Python web scraper" --model claude/sonnet --direct

# Check status
oa status

---

Architecture

┌─────────────────────────────────────────────────────┐
│                    oa CLI (Python)                   │
│  start · run · status · pipeline · delegate · web   │
├──────────┬──────────┬───────────┬───────────────────┤
│  tmux    │ Textual  │ React     │ MCP Server        │
│  session │ TUI      │ Web UI    │ (FastMCP)         │
├──────────┴──────────┴───────────┴───────────────────┤
│              Claude Code CLI (subscription)          │
│         Claude Opus · Sonnet · Haiku · Ollama        │
├─────────────────────────────────────────────────────┤
│  1612+ Agent Templates  │  22 Skills  │  Guardians  │
│  112 categories          │  7 domains  │  auto-run   │
└─────────────────────────────────────────────────────┘

---

Commands

Command	Description
oa start	Start tmux session
oa run "<task>"	Spawn an agent
oa status	Show all agents
oa watch <name>	Stream live output
oa collect <name>	Get completed output
oa pipeline "<task>"	Multi-agent pipeline (Planner → Workers → Combiner)
oa delegate "<task>"	Autonomous orchestrator agent
oa dashboard	Textual TUI dashboard
oa web	React web UI at localhost:5174
oa team create <n>	Create agent team
oa send <agent> "msg"	Inter-agent messaging
oa kill <name>	Stop an agent
oa clean	Remove completed workspaces

oa run Options

Option	Description
--name NAME	Agent name (auto-generated if omitted)
--model MODEL	claude/opus, claude/sonnet, claude/haiku, ollama/<model>
--parent NAME	Parent agent for nested hierarchies
--direct	Write directly to codebase (skip proposals)
--workspace DIR	Use existing workspace
--prompt-file FILE	Read task from file (avoids shell escaping issues)

Multi-line Prompts & Special Characters

When your prompt contains single quotes, backticks, $variables, or spans multiple lines, use --prompt-file to avoid shell parsing issues:

# Write your prompt to a file
cat > /tmp/task.txt << 'EOF'
Refactor the `parse_user()` function in src/utils.py.
It's broken when input contains $special chars or it's > 100 chars.
EOF

oa run --prompt-file /tmp/task.txt --model claude/sonnet

Or use a heredoc variable:

PROMPT=$(cat << 'EOF'
Your multi-line prompt with 'quotes', `backticks`, and $variables here.
EOF
)
oa run "$PROMPT" --model claude/sonnet

---

Pipeline Orchestration

oa pipeline "Build a CSV validator library with tests and README"

Planner (5 min) → plan.json → Subtask agents (parallel, 30 min each) → Combiner (10 min) → result.md

---

Agent Library

1612+ templates across 112 categories, including:

Domain	Categories	Examples
Development	code-dev, frontend, backend, testing, devops	Find bugs, generate tests, review code
AEC	blender, bonsai, ifcopenshell, sverchok	3D modeling, BIM authoring, IFC processing
Data	analytics, data-pipeline, ml-ops, database	Transform, visualize, validate
Business	finance, legal, marketing, hr, logistics	Compliance, content, operations
Infrastructure	cloud, security, monitoring, iot	Deploy, audit, observe

---

Installation

Linux / macOS / WSL2

./install.sh

Verify

oa version
oa doctor     # Check all dependencies

Requirements

Dependency	Version
Python	>= 3.10
tmux	any recent
Claude Code CLI	latest (active subscription)

Windows: Use WSL2. Native Windows is not supported.

---

Visual Canvas (Advanced)

Drag-and-drop React Flow canvas for building complex agent workflows:

pnpm install && pnpm dev    # localhost:5173

160+ pre-built agents, flow + pool patterns, assembly engine (NL → agent graph), safety & audit, multi-provider support.

---

Contributing

Contributions welcome! Open an issue or submit a PR.

1. Fork the repository
2. Create a feature branch
3. Submit a pull request

Organization

Entity	Role
OpenAEC Foundation	Open-source stewardship

License

MIT

---

Build free. Build together.