Skip to content

Index

agentic-team

Orchestrate teams of AI coding agents working in parallel inside tmux sessions.

A team lead agent runs interactively and delegates tasks to worker agents (Claude, Codex, Gemini) that execute in their own tmux windows. Workers can run in oneshot or interactive mode, each with independent working directories, providers, and models.

User ──> team CLI ──> tmux session
                       ├── window 0: Team Lead (interactive)
                       ├── window 1: worker "fix-auth"
                       ├── window 2: worker "add-tests"
                       └── ...

Demo

agentic-team demo

Features

  • Conversational delegation: Talk to the lead agent like a project manager — it breaks down work, spawns workers, monitors progress, and synthesizes results
  • Multi-provider: Supports Claude Code, Codex, and Gemini CLI agents
  • Parallel execution: Workers run in isolated tmux windows with automatic logging
  • Two worker modes: Interactive (persistent, supports follow-ups) and oneshot (fire-and-forget)
  • Task files: Define batch tasks in markdown with checkbox syntax
  • Smart routing: team "your prompt" sends directly to the lead agent
  • Completion detection: Automatic status tracking via tmux pane analysis
  • Multi-window dashboard: team attach --multi for a tiled view of all workers
  • Session resumption: Resume Claude and Gemini sessions manually; completed Claude oneshot workers can also be resumed automatically

Installation

pip install agentic-team

See Getting Started for prerequisites and first-run setup.