AI Agent Cluster Command Center — A desktop application to manage multiple AI assistants simultaneously.
Note
🚧 Project Status: Beta 🚧
Check out our Development Roadmap to see the latest progress and future plans!
Neural Hive is a desktop application that allows you to manage multiple AI programming assistants (like Claude Code, Gemini CLI, Aider, etc.) simultaneously, monitoring their status, viewing their output, and sending commands through a unified interface.
When you need to use multiple AI assistants for different tasks:
- ❌ Opening multiple terminal windows is hard to track.
- ❌ It's difficult to quickly grasp the status of each assistant.
- ❌ You have to switch between windows to send the same command.
- ❌ You can't see at a glance "who is thinking, who is working, and who has an error."
Neural Hive centralizes all AI assistants on one screen, providing:
- ✅ Status Overview — See the status of all agents at a glance.
- ✅ Focus Mode — Click on any agent to enter a full-screen terminal.
- ✅ Batch Operations — Send the same command to multiple agents at once.
- ✅ Intelligent Parsing — Automatically identifies "THINKING," "WORKING," and "ERROR" states.
┌─────────────────────────────────────────────────────────────┐
│ 🐝 Neural Hive [＋] [Settings] │
├─────────────────────────────────────────────────────────────┤
│ │
│ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │
│ │ Claude-01 │ │ Gemini-02 │ │ Aider-03 │ │
│ │ 🟡 THINKING │ │ 🟢 WORKING │ │ 🔵 IDLE │ │
│ │ │ │ │ │ │ │
│ │ Analyzing │ │ Reading │ │ Waiting... │ │
│ │ codebase... │ │ files... │ │ │ │
│ └─────────────┘ └─────────────┘ └─────────────┘ │
│ │
│ Status: 🟡 1 Thinking | 🟢 1 Working | 🔵 1 Idle | 🔴 0 Error │
└─────────────────────────────────────────────────────────────┘
Neural Hive uses an Electron-based architecture that separates the core logic (Main Process) from the UI (Renderer Process). It interfaces with external command-line tools via a pseudo-terminal (node-pty) to ensure full compatibility.
┌─────────────────────────────────────────────────────────────┐
│ Electron Application │
├─────────────────────────────┬───────────────────────────────┤
│ Main Process │ Renderer Process │
│ ┌───────────────────────┐ │ ┌─────────────────────────┐ │
│ │ Node.js + TypeScript │ │ │ Vue 3 + TypeScript │ │
│ │ ─────────────────────│ │ │ ───────────────────── │ │
│ │ • node-pty (TTY) │ │ │ • Pinia (State Mgmt) │ │
│ │ • Process Manager │◄─┼──┤ • Tailwind CSS │ │
│ │ • Regex Parser │──┼─►│ • Shadcn/ui │ │
│ │ • IPC Handler │ │ │ • xterm.js (Terminal) │ │
│ └───────────────────────┘ │ └─────────────────────────┘ │
└─────────────────────────────┴───────────────────────────────┘
│
▼
┌───────────────────────────────┐
│ External CLI Processes │
│ ┌─────────┐ ┌──────────┐ │
│ │ Claude │ │ Gemini │ │
│ └─────────┘ └──────────┘ │
└───────────────────────────────┘
- Add Agent — Launch Claude, Gemini, Aider, or custom CLI tools.
- Custom Naming — Assign meaningful names (e.g., "Frontend", "Backend") to each agent for better organization.
- Multi-instance Support — Run 20+ agents simultaneously.
- Lifecycle Control — Start, pause, restart, and terminate any agent.
- True PTY Simulation — Deceives CLI tools into thinking they're in a real terminal, preserving colors, prompts (y/n), and cursor movements.
- Focus Mode — Click a card to enter a full-screen terminal with a dedicated "Chain of Thought" and "Skills Used" panel.
- Context Awareness — Always see the agent's Current Working Directory (CWD) at the bottom of the card.
- Rich State Detection — Automatically identifies states like
IDLE,THINKING,WORKING,ERROR, and evenWAITING_USERfor interactive prompts. - Skill Tracking — Intelligently extracts the tool an agent is using (e.g., "Reading file...", "Searching...") and displays it.
- Visual Feedback — A "breathing light" effect on the card border shows you exactly who is working.
- Instant Error Alerts — Agent cards turn red and a badge appears on the taskbar icon if an error occurs.
- Responsive Layout — The grid automatically adjusts its columns to fit your window size.
- Command Broadcasting — Send the same command to multiple agents at once using a flexible tagging system.
- Variable Injection — Use variables in your broadcasted commands for dynamic prompts (e.g.,
Analyze {file_path}).
- Config-Driven Parsers — Add support for new AI CLI tools yourself by simply editing a JSON file—no code changes needed.
- High-Performance Rendering — A built-in output throttler prevents UI freezes even when agents produce a high volume of text, ensuring a smooth experience.
- Node.js 18.x or higher
- npm 9.x or higher
- Python 3.x (Optional, for running scripts in
execution) - OS Windows 10+, macOS 10.15+, Linux
# 1. Clone the project
git clone https://github.com/your-org/neural-hive.git
cd neural-hive
# 2. Install Node.js dependencies
npm install
# 3. (Optional) Install Python dependencies
# If execution/parse_lint.py requires extra packages, creating a requirements.txt is recommended.
# pip install -r execution/requirements.txt# Development Mode (Hot-reloading)
npm run dev # Starts the Vite dev server
npm run dev:electron # Starts Electron (in a separate terminal)
# Production Build
npm run build # Builds the frontend
npm run build:electron # Packages the Electron appneural-hive/
├── electron/ # Electron Main Process
│ ├── main.ts # Main process entry point
│ ├── preload.ts # Preload script (IPC Bridge)
│ ├── config-loader.ts # Parser rules loader
│ ├── output-throttler.ts # Output throttler (100ms)
│ ├── broadcast-manager.ts # Command broadcast manager
│ └── __tests__/ # Backend tests
│
├── src/ # Vue Frontend
│ ├── App.vue # Main app component
│ ├── main.ts # Vue entry point
│ ├── components/ # UI Components
│ │ ├── AgentCard.vue # Agent card
│ │ ├── AgentGrid.vue # Agent grid layout
│ │ ├── FocusMode.vue # Focus mode (full-screen)
│ │ ├── TerminalView.vue # xterm.js terminal
│ │ └── SkillsPanel.vue # Skills/thought chain panel
│ ├── composables/
│ │ └── useTerminal.ts # xterm.js integration
│ ├── stores/
│ │ └── agentStore.ts # Pinia state management
│ ├── utils/
│ │ └── parser.ts # Output parsing utility
│ └── __tests__/ # Frontend tests
│
├── config/
│ └── parsers.json # Parsing rules configuration
│
├── scripts/
│ └── mock-agent.js # Mock agent for development
│
├── package.json
├── vite.config.ts
├── tailwind.config.js
├── tsconfig.json
└── vitest.config.ts
Defines how to identify agent states from CLI output:
{
"parsers": {
"claude": {
"name": "Claude Code",
"command": "claude",
"patterns": {
"thinking": "Thinking|Analyzing|Considering",
"tool_use": "\[Tool Use\]|Reading|Writing|Searching",
"error": "Error|Exception|Failed"
}
}
}
}| Variable | Description | Default |
|---|---|---|
NODE_ENV |
Environment mode | development |
VITE_PORT |
Dev server port | 5173 |
# Run all tests
npm test
# Run tests once (no watch mode)
npm test -- --run
# Run tests and generate a coverage report
npm run test:coverage| Module | Tests |
|---|---|
parser.ts |
31 |
agentStore.ts |
21 |
output-throttler.ts |
15 |
broadcast-manager.ts |
25 |
config-loader.ts |
20 |
| Total | 112 |
- Edit
config/parsers.jsonto add new parsing rules. - Update the
AgentTypeinsrc/types/shared.ts. - Update
generateNameinsrc/stores/agentStore.ts.
- Add the channel name in
src/types/shared.ts. - Implement the handler in
electron/main.ts. - Expose the API in
electron/preload.ts. - Call
window.electronAPI.xxx()in the frontend.
A: Check if the regular expressions in config/parsers.json match the output format of that CLI tool.
A: Add a new parser configuration in config/parsers.json that defines the output patterns for the tool.
A: It is designed to support 20+ concurrent agents, but the actual number depends on system resources.
- ✅ Electron Sandbox Mode (
sandbox: true) - ✅ Context Isolation (
contextIsolation: true) - ✅ Node Integration Disabled (
nodeIntegration: false) - ✅ Full Input Validation
- ✅ No Hard-coded Sensitive Information
MIT License
Issues and Pull Requests are welcome!
Neural Hive — Let your AI assistants work as a cluster.
Made with ❤️ by the Neural Hive Team