agent-tower/core

What it solves

Agent Tower provides a centralized command center for managing multiple AI agents simultaneously. It eliminates the visual chaos of juggling multiple terminal windows, prevents code conflicts through automatic workspace isolation, and allows users to monitor and manage agent tasks from any device, including mobile phones.

How it works

The system acts as an orchestration layer over existing agent CLI tools (like Claude Code, Gemini CLI, and Cursor Agent). It uses a Kanban-style dashboard to track task status (Review, Running, Done). To prevent conflicts, it automatically creates isolated Git Worktrees for each task, ensuring agents work on separate branches. It employs a AgentPipeline architecture using pseudo-terminals (PTY) and output parsers to stream real-time progress to a web interface.

Who it’s for

Developers who use multiple AI coding agents and want to organize their workflows, avoid merge conflicts, and reduce costs by assigning specific models to specific tasks based on their capabilities.

Highlights

• Multi-Agent Dashboard: A single interface to manage projects, tasks, and agents across different providers.
• Git Worktree Isolation: Automatic creation of isolated branches for every task to eliminate edit conflicts.
• **Team Mode: Enables multi-agent collaboration in a shared group chat where agents can take on specific roles (lead, implementer, reviewer, tester).
• Provider Flexibility: Ability to pick different AI providers per task to optimize for cost and performance.
• MCP Integration: Built-in Model Context Protocol server allowing agents to proactively claim and report progress on tasks.
• Remote Access: One-click Cloudflare tunnels for mobile monitoring and notifications via desktop or Lark webhooks.