Multea

Terminal-based multi-agent orchestrator for Claude Code
Run multiple AI coding agents across repositories simultaneously with a unified TUI and AI-powered task coordination.

Features • Quick Start • Architecture • Usage • Configuration • Shortcuts • Contributing

---

Why Multea?

Modern software teams work across multiple repositories simultaneously. AI coding assistants are powerful, but managing separate agent sessions for each repo is chaotic. Multea solves this by providing a single terminal interface that orchestrates multiple Claude Code agents working in parallel — with dependency-aware task scheduling, real-time streaming output, and an AI coordinator that breaks complex instructions into subtasks.

Think of it as tmux for AI agents — but with built-in task orchestration, dependency graphs, and an AI brain coordinating everything.

Features

Multi-Agent Orchestration

• Concurrent agents — Spawn independent Claude Code agents per project, all running in parallel
• Task queue with dependency resolution — DAG-based scheduling ensures tasks execute in the correct order
• Priority levels — Critical, high, normal, and low priority task support
• Auto-dispatch — Idle agents automatically pick up the next eligible task
• Pause / Resume / Abort — Full lifecycle control over individual agents

AI-Powered Coordination

• Top-level orchestrator — An AI coordinator that understands high-level instructions and decomposes them into subtasks
• Slash commands — /dispatch, /broadcast, /status and more for direct agent control
• DAG planner — Topological sort with cycle detection for complex multi-step workflows
• Cross-repo awareness — Coordinate changes that span multiple codebases

Rich Terminal UI (TUI)

• Multi-pane layout — Agent output, orchestrator chat, task queue, and questions — all visible at once
• Real-time streaming — Watch agent output as it arrives, not after it completes
• Keyboard-driven navigation — Vim-inspired shortcuts for fast context switching
• Directory browser — Browse project files without leaving the TUI
• Markdown rendering — Rich text in the orchestrator pane

Developer Experience

• Session persistence — Resume where you left off across restarts
• Dual authentication — API key or Claude Code native auth
• Config-driven setup — Single JSON file to register all your projects
• File logging — All agent outputs logged to ./logs/ for post-session review

Quick Start

Prerequisites

• Node.js >= 18
• Claude Code CLI installed (get it here)
• An Anthropic API key or active Claude Code authentication

Installation

# Clone the repository
git clone https://github.com/ashokDevs/multea.git
cd multea

# Install dependencies
npm install

Run

# Development mode
npm run dev

# Or with a custom config path
npx tsx src/index.tsx path/to/config.json

# Production build
npm run build && node dist/index.js

On first launch, Multea will prompt you to choose an authentication method (API key or Claude Code auth). Your choice is saved for future sessions.

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                     Terminal UI (React + Ink)                    │
│  ┌──────────────┐ ┌──────────────┐ ┌──────────┐ ┌───────────┐  │
│  │ Agent Panes  │ │ Orchestrator │ │  Tasks   │ │ Questions │  │
│  │ (per project)│ │   Chat Pane  │ │  Queue   │ │   Pane    │  │
│  └──────┬───────┘ └──────┬───────┘ └────┬─────┘ └─────┬─────┘  │
│         │                │              │              │         │
├─────────┴────────────────┴──────────────┴──────────────┴────────┤
│                                                                  │
│  ┌──────────────────────┐    ┌──────────────────────────────┐   │
│  │    AgentManager      │◄──►│       Orchestrator           │   │
│  │  ┌────────────────┐  │    │  - AI coordinator            │   │
│  │  │  AgentRunner A │  │    │  - Slash command parser      │   │
│  │  │  AgentRunner B │  │    │  - DAG planner               │   │
│  │  │  AgentRunner C │  │    │  - Conversation history      │   │
│  │  └────────────────┘  │    └──────────────────────────────┘   │
│  │  ┌────────────────┐  │                                       │
│  │  │   TaskQueue    │  │    ┌──────────────────────────────┐   │
│  │  │ (DAG-based)    │  │    │     Persistence Layer        │   │
│  │  └────────────────┘  │    │  .multea-state.json          │   │
│  └──────────────────────┘    └──────────────────────────────┘   │
│                                                                  │
├──────────────────────────────────────────────────────────────────┤
│              @anthropic-ai/claude-agent-sdk                      │
└──────────────────────────────────────────────────────────────────┘

Core Components

Component	Responsibility
AgentManager	Central registry for all agents. Handles task queuing, dependency resolution, and event emission.
AgentRunner	Execution engine per project. Wraps the Claude Agent SDK, manages sessions, buffers output.
TaskQueue	DAG-based dependency resolution with priority ordering and blocked-state tracking.
Orchestrator	Top-level AI coordinator. Translates high-level instructions into structured task dispatches.
Persistence	Saves orchestrator state, task queue, and session data to disk for resume capability.

Data Flow

User Input → App → AgentManager / Orchestrator → TaskQueue → AgentRunner → Claude Agent SDK
                                                                    ↓
Terminal UI ← React State Updates ← Event Emitters ← Agent Output Stream

Usage

Slash Commands

Command	Description
/dispatch <project> <prompt>	Send a task to a specific project agent
/broadcast <prompt>	Send the same task to all agents
/status	View the status of all agents and queued tasks
/help	Show all available commands
/clear	Clear the orchestrator chat history
/context	View current context and session info

Task Dependencies

Tasks can declare dependencies using the AFTER: prefix:

/dispatch backend "Create user API endpoint"
/dispatch frontend "Build user form component AFTER:1"
/dispatch e2e-tests "Write integration tests AFTER:1,2"

Task 2 waits for task 1 to complete. Task 3 waits for both tasks 1 and 2.

Configuration

Create a multea.config.json in your project root:

{
  "projects": [
    {
      "name": "backend",
      "path": "/absolute/path/to/backend-repo"
    },
    {
      "name": "frontend",
      "path": "/absolute/path/to/frontend-repo"
    },
    {
      "name": "shared-lib",
      "path": "/absolute/path/to/shared-library"
    }
  ],
  "initialTasks": [
    {
      "projectName": "backend",
      "prompt": "Review the current API endpoints and suggest improvements"
    }
  ]
}

Field	Required	Description
projects[].name	Yes	Unique identifier for the project agent
projects[].path	Yes	Absolute path to the project directory
initialTasks	No	Tasks to auto-dispatch on startup

Keyboard Shortcuts

Key	Action
q	Quit Multea
1 / 2 / 3	Switch between panes (agents / orchestrator / questions)
h / l	Navigate panes left / right
Tab / Shift+Tab	Cycle through agent panes
i / Enter	Enter input mode (type commands)
Esc	Exit input mode
o	Trigger orchestrator evaluation
x	Stop the focused agent
?	Show help popup

Tech Stack

Layer	Technology
Runtime	Node.js
Language	TypeScript 5.9+ (strict mode)
UI Framework	React 19 + Ink (React for terminals)
AI Integration	@anthropic-ai/claude-agent-sdk
Build	tsup (ESM output)
Dev Server	tsx
Linting	ESLint 9 + TypeScript ESLint
Formatting	Prettier

Project Structure

src/
├── index.tsx                  # Entry point — auth selection & app bootstrap
├── app.tsx                    # Main React app component
├── config.ts                  # Config loading & validation
├── types.ts                   # Shared TypeScript type definitions
│
├── core/
│   ├── agent-manager.ts       # Multi-agent registry & task dispatcher
│   ├── agent-runner.ts        # Single agent execution engine
│   ├── orchestrator.ts        # High-level AI task coordinator
│   ├── task-queue.ts          # DAG-based dependency-aware task queue
│   ├── persistence.ts         # State serialization & resume
│   └── sdk-env.ts             # Authentication & environment setup
│
├── components/                # React + Ink terminal UI components
│   ├── agent-pane.tsx         # Individual agent output display
│   ├── orchestrator-pane.tsx  # Orchestrator chat interface
│   ├── task-pane.tsx          # Task queue visualization
│   ├── command-bar.tsx        # User input bar
│   ├── questions-pane.tsx     # Agent-posed questions
│   ├── sidebar-pane.tsx       # Navigation sidebar
│   ├── status-bar.tsx         # System status display
│   ├── help-popup.tsx         # Keyboard shortcut reference
│   ├── directory-browser.tsx  # In-TUI file browser
│   └── sessions-pane.tsx      # Session management
│
├── orchestrator/              # Advanced orchestration subsystem
│   ├── index.ts               # Orchestrator entry point
│   ├── core/                  # Core orchestration logic
│   ├── scheduler/             # DAG planner & topological sort
│   ├── connectors/            # Extensible task executors
│   └── questions/             # Agent question routing
│
└── utils/                     # Shared utilities

Use Cases

• Monorepo development — Coordinate changes across frontend, backend, and shared libraries simultaneously
• Microservices — Update multiple services in parallel with dependency-aware ordering
• Migration projects — Run migration scripts across dozens of repos with a single command
• Code review assistance — Have agents review different parts of a large changeset concurrently
• Refactoring at scale — Rename, restructure, or update patterns across your entire stack

Roadmap

• Custom themes for the TUI
• Plugin system for user-defined commands
• Vim mode for power-user navigation
• Web dashboard companion
• Agent-to-agent communication
• Metrics and cost tracking dashboard

Contributing

Contributions are welcome! Whether it's bug reports, feature requests, or pull requests — all contributions help make Multea better.

# Fork & clone
git clone https://github.com/<your-username>/multea.git

# Install dependencies
npm install

# Start development
npm run dev

# Lint & format
npm run lint
npm run format

License

MIT © ashokDevs

---

Built with Claude Code and the Claude Agent SDK