Orchestrate multiple AI Agents to work together like a human team Complete complex software development tasks through reliable messaging protocols
Features | Quick Start | Architecture | Examples | Contributing
AgentHub is an open-source Multi-Agent Orchestration Framework that enables reliable communication and coordinated work between multiple AI Agents through a message routing architecture.
With a single command, launch a complete AI development team with:
- 1 Coordinator Agent (MAIN) - task planning, coordination, and review
- 4 Executor Agents (A/B/C/D) - parallel task execution
Manage AI collaborative development just like managing a human team.
| Traditional AI Development | AgentHub |
|---|---|
| Single AI working alone | Multi-Agent Parallel Collaboration |
| No division of labor | Automatic Task Breakdown & Assignment |
| High communication overhead | Standardized Messaging Protocol |
| State easily lost | Persistence + Crash Recovery |
| Hard to track progress | Complete Task State Management |
./scripts/start_team.sh- Auto-start Router (message hub)
- Open 5 independent terminal windows
- Generate standard documentation templates
- Inject AI role prompts
# Analyze project before adding features
team analyze --path . --feature "user login"
# Generate design document
team design --requirement "user login with OAuth"
# Smart execution with contract-first design
team run --task "implement login" --design-approved- Complexity Judgment - Auto-detect simple/complex tasks
- Task Decomposition - Break down into subtasks by domain
- Contract-First Design - Generate interface contracts before coding
- Specialized Agents - Frontend(A), Backend(B), Database(C), Support(D)
# View progress board
team board
# Report progress
team progress --task TASK-001 --percent 50 --step "implementing API"
# Lock files to prevent conflicts
team lock --files "src/api.py" --task TASK-001
# Notify interface changes
team notify --task TASK-001 --interface "POST /api/login" --change-type modify- Progress Board - Real-time visibility of all agents
- File Locking - Prevent code conflicts
- Change Broadcasting - Auto-notify when interfaces change
- Dependency Tracking - Track and alert on blocked tasks
- ACK Confirmation - Dual acknowledgment at delivery and application layer
- Auto Retry - Exponential backoff retry strategy
- Timeout Handling - Automatic timeout detection and handling
- Idempotency - Message deduplication to avoid duplicate execution
analyze -> design -> confirm -> schedule -> execute -> aggregate
Standardized AI-to-AI communication protocol supporting:
- Project analysis & impact assessment
- Design document generation
- Contract-first task decomposition
- Parallel execution with coordination
- Result aggregation
- Message logs (JSONL format)
- Inbox state persistence
- Auto-recovery after crash
- Session/epoch management support
- Customizable agent count
- Support for different AI CLI tools (Codex, Claude Code, etc.)
- Environment variable configuration
Watch how AgentHub orchestrates 5 AI agents working together
Coming soon: Video demonstration of multi-agent collaboration
Router Server
(Message Routing / State Management / Delivery)
│
┌───────┬───────┼───────┬───────┐
│ │ │ │ │
┌───▼───┐ ┌─▼───┐ ┌─▼───┐ ┌─▼───┐ ┌─▼───┐
│ MAIN │ │ A │ │ B │ │ C │ │ D │
│Coord. │ │Exec │ │Exec │ │Exec │ │Exec │
│Agent │ │Agent│ │Agent│ │Agent│ │Agent│
└───────┘ └─────┘ └─────┘ └─────┘ └─────┘
Role Responsibilities:
| Agent | Role | Specialty | Responsibilities |
|---|---|---|---|
| MAIN | Coordinator | - | Task planning, design review, problem solving |
| A | Frontend Expert | UI/UX | React, Vue, CSS, components, pages |
| B | Backend Expert | API | FastAPI, business logic, services |
| C | Database Expert | Data | Models, migrations, queries |
| D | Support Expert | DevOps | Testing, docs, deployment |
AgentHub defines a complete AI-to-AI communication protocol:
| Message Type | Direction | Purpose |
|---|---|---|
review |
MAIN->Members | Review documents/code |
report |
Members->MAIN | Feedback review results |
assign |
MAIN->Members | Assign tasks |
clarify |
Members->MAIN | Ask questions |
answer |
MAIN->Members | Answer questions |
verify |
MAIN->Members | Verify changes |
done |
Members->MAIN | Task complete |
fail |
Members->MAIN | Task failed |
See docs/main-members-workflow.md for complete protocol specification.
- macOS (Linux support planned)
- Python 3.8+
- Terminal.app or iTerm2
- AI CLI tool (Codex, Claude Code, or compatible)
# Clone repository
git clone https://github.com/Dmatut7/AgentHub.git
cd AgentHub# Start in your project directory
./scripts/start_team.shThe system will automatically:
- Start Router (default port 8765)
- Generate standard documentation templates
- Open 5 terminal windows for each agent
# MAIN writes code -> A/B/C/D review in parallel -> MAIN consolidates feedback# MAIN breaks down feature -> A/B/C/D implement components -> MAIN integrates# MAIN outlines -> A/B/C/D write sections -> MAIN finalizes# MAIN describes bug -> A/B/C/D investigate & propose fixes -> MAIN verifiesSee EXAMPLES.md for detailed use cases.
AgentHub/
├── scripts/ # Launch scripts
│ ├── start_team.sh # One-command launch
│ ├── stop_team.sh # Stop system
│ └── status_team.sh # Check status
├── src/
│ ├── api/ # HTTP server
│ ├── cli/ # CLI tools
│ ├── router/ # Message routing core
│ ├── protocol/ # Protocol definitions
│ ├── state/ # State management
│ ├── storage/ # Persistent storage
│ └── launcher/ # Terminal launcher
├── prompts/ # AI prompt templates
├── doc/ # Documentation templates
├── docs/ # Design documents
│ ├── design.md # System architecture
│ └── main-members-workflow.md # Protocol spec
└── README.md
# === System Management ===
./scripts/start_team.sh # Start system
./scripts/status_team.sh # Check status
./scripts/stop_team.sh # Stop system
# === Smart Task Flow (NEW) ===
team analyze --path . --feature "new feature" # Analyze impact
team design --requirement "feature description" # Generate design
team run --task "feature" --design-approved # Execute with contracts
team schedule --task "feature description" # Preview schedule only
# === Collaboration (NEW) ===
team board # View progress board
team progress --task T1 --percent 50 --step "..."# Report progress
team lock --files "src/api.py" --task T1 # Lock files
team notify --task T1 --interface "API" -c modify # Notify changes
# === Messaging ===
team say --from MAIN --to A --text "Start task" # Send message
team review --to A,B,C,D --task T1 --file doc.md # Request review
team assign --to B --task T1 --files "src/*" # Assign task
# === Monitoring ===
team status --tasks # View task status
team trace --task T1 # Trace task messages
curl http://127.0.0.1:8765/status | jq # Raw API status| Environment Variable | Description | Default |
|---|---|---|
TERMINAL_ADAPTER |
Terminal type (terminal/iterm) |
terminal |
CODEX_PATH |
AI CLI executable path | codex |
We welcome contributions! Please see CONTRIBUTING.md for details.
Quick steps:
- Fork the repository
- Create your feature branch (
git checkout -b feature/AmazingFeature) - Commit your changes (
git commit -m 'Add some AmazingFeature') - Push to the branch (
git push origin feature/AmazingFeature) - Open a Pull Request
- Design Doc - System architecture
- Protocol Spec - Message protocol details
- Examples - Use cases and examples
- Contributing - Contribution guide
- Support - Help & troubleshooting
- Changelog - Version history
- Linux support
- Windows support
- Web dashboard for monitoring
- More AI model integrations (GPT-4, Claude, Gemini)
- Plugin system for custom protocols
- Distributed agent support (across machines)
MIT License © 2026 Dmatut7
AgentHub - Making AI team collaboration simpler.
GitHub | Issues | Discussions
⭐ If you find this project helpful, please give it a Star!