Agent Teams

Positioning#

Agent Teams is an Electron-based AI Agent team orchestration desktop app where users act as CTO, managing AI agent development teams via kanban boards and high-level instructions.

Core Capabilities#

Multi-Agent Orchestration#

• Create parallel agent teams across 75+ LLM providers (Claude, Codex, OpenCode, etc.)
• Solo mode: single agent team that autonomously creates tasks and shows progress, saving tokens
• Flexible autonomy control: from fully autonomous to per-operation approval, with notification popups

Inter-Agent Communication & Collaboration#

• Built-in real-time email system for agents to message each other, create/manage tasks, review, and comment
• Cross-team communication: agents from different teams can fully interact with configurable collaboration and messaging rules

Kanban & Task Management#

• 5-column real-time kanban with automatic task status changes
• Task dependencies (blocked by) and link references
• Task attachments: agents automatically attach files, can read and attach files

Code Review & Editing#

• Cursor-like experience with hunk-level accept/reject/comment
• Built-in inter-agent review workflow
• Built-in code editor with Git support

Monitoring & Observability#

• Execution log viewer: per-task isolated display of agent/runtime logs, tool calls, reasoning, and timeline
• Real-time process monitor: view running agent processes, open URLs in browser, stop processes
• Session analytics: 6-category token tracking

Isolation & Security#

• Git worktree isolation: optionally enable, each agent works in an independent worktree
• IPC and HTTP handlers validate IDs/paths/payloads at boundaries
• Write operations restricted to selected project root, blocking path traversal and sensitive config access

Architecture Highlights#

The project uses a pnpm workspace monorepo structure:

• Desktop framework: Electron 40, built with electron-vite
• Frontend UI: React 19 + TypeScript 5 + Tailwind CSS 3 + Zustand 4
• Agent orchestration: agent-teams-controller subpackage (core orchestration engine)
• MCP integration: mcp-server / agent-teams-mcp subpackages
• Visualization: packages/agent-graph (force-directed graph + kanban zone layout)
• Web mode: vite.web.config.ts + dev:web script
• Containerization: docker/ directory with Dockerfile
• Code quality: ESLint + Prettier + Vitest + knip + Husky + CodeRabbit auto review
• CI/CD: GitHub Actions (including Ubuntu runtime tests)

Key mechanisms: Inter-agent communication via built-in real-time email system; data reads from local ~/.claude/ directory (session logs, todos, tasks); auto-detection of runtime/provider without prerequisite installation.

Use Cases#

• Dev team simulation: Assign AI agents to different roles (frontend, backend, testing) for a complete development pipeline
• Automated code review: Agents auto-review each other's output, humans only do final approval
• Multi-project parallel management: Manage multiple projects and teams simultaneously
• Freelancer/small team productivity: Leverage existing LLM subscriptions/API keys at zero extra cost for multi-agent collaboration

Installation & Getting Started#

Download platform-specific installers from GitHub Releases (macOS / Windows / Linux). On first launch, a setup wizard auto-detects runtime and guides provider authentication, then select a project, define roles, write provisioning prompts to get started.

Build from source:

git clone https://github.com/777genius/claude_agent_teams_ui.git
pnpm install
pnpm dev          # Development mode
pnpm build        # Build
pnpm dist         # Package distribution

Competitive Comparison#

Compared to Vibe Kanban, Aperant, Cursor, and Claude Code CLI, Agent Teams offers unique advantages in cross-team communication, native agent-to-agent messaging, built-in review workflows, and flexible autonomy control.

Pending Confirmation#

• Full list of 75+ LLM providers not detailed in README
• v1.3.0 version status: package.json shows 1.3.0 but GitHub Releases latest is v1.2.0
• No public web deployment URL found
• MCP Server specific exposed tools/resources/prompts scope to be confirmed