A desktop application for orchestrating multi-agent AI coding sessions. Launch coordinated teams of AI coding assistants (Claude, Codex, Gemini, etc.) that work together on complex software tasks.
- Hive Mode: Queen + Workers architecture for coordinated development
- Swarm Mode: Hierarchical planners with domain-specific mini-hives
- Fusion Mode: Parallel competing implementations with best-pick resolution
- Session Persistence: Save and resume sessions across app restarts
- Multi-CLI Support: Works with Claude Code, Codex, OpenCode, Gemini CLI, and more
- Real-time Monitoring: Watch all agents work simultaneously with live terminal output
- Git Integration: Automatic branch management and coordination
Download the latest release from the Releases page:
- NSIS Installer:
Hive Manager_x.x.x_x64-setup.exe(recommended) - MSI Installer:
Hive Manager_x.x.x_x64_en-US.msi
Requirements:
- Node.js 18+
- Rust 1.70+
- Windows 10/11
# Clone the repository
git clone https://github.com/rdfitted/hive-manager.git
cd hive-manager
# Install dependencies
npm install
# Development mode
npm run tauri dev
# Production build
npm run tauri build- Launch Hive Manager
- Click New Session in the sidebar
- Select your project directory
- Choose a session type (Hive, Swarm, or Fusion)
- Configure your agents (CLI, model, roles)
- Click Launch
Traditional Queen + Workers setup. The Queen coordinates and delegates tasks to specialized Workers.
Hierarchical architecture with Planners that each manage their own mini-hive of Workers. Great for large, multi-domain projects.
Launch multiple agents working on the same task in parallel. Compare approaches and pick the best solution.
| CLI | Behavior | Notes |
|---|---|---|
| Claude Code | Action-Prone | Anthropic's official CLI. Needs role hardening for worker agents. |
| Gemini CLI | Action-Prone | Google's CLI. Similar behavior to Claude. |
| Codex | Explicit-Polling | OpenAI's CLI. Uses bash loops for coordination. |
| OpenCode | Explicit-Polling | Open-source alternative. |
| Qwen | Instruction-Following | Follows instructions literally, respects role boundaries naturally. |
| Droid | Interactive | TUI mode with /model command for model selection. |
| Cursor | Interactive | Runs via WSL. Uses global model setting. |
Behavior profiles determine how Hive Manager prompts each agent:
- Action-Prone: Proactive agents that need strong constraints to stay in their lane
- Instruction-Following: Literal interpreters that respect role boundaries naturally
- Explicit-Polling: Agents that need bash loops for coordination
- Interactive: TUI-based agents with different prompt injection
Sessions are stored in %APPDATA%/hive-manager/sessions/.
App configuration is in %APPDATA%/hive-manager/config.json.
# Run in development mode with hot reload
npm run tauri dev
# Type checking
npm run check
# Build for production
npm run tauri build- Frontend: SvelteKit 5, TypeScript
- Backend: Rust, Tauri 2
- Terminal: xterm.js with PTY support
Contributions are welcome! Please feel free to submit a Pull Request.
MIT License - see LICENSE for details.