Core Concepts
Reasoning
Memory & Retrieval
Agent Types
Design Patterns
Training & Alignment
Frameworks
Tools
Safety & Security
Evaluation
Meta
User Tools
Sidebar
claude_squad
This is an old revision of the document!
Table of Contents
Claude Squad
Claude Squad is a terminal application that manages multiple AI coding agents running in parallel1). Built in Go, it uses tmux for session isolation and git worktrees for branch-level workspace separation, enabling developers to orchestrate Claude Code, Codex, Gemini, Aider, and other agents from a single TUI interface.
ai_agent cli orchestration tmux git multi_agent open_source
| Repository | https://github.com/smtg-ai/claude-squad |
| Website | https://smtg-ai.github.io/claude-squad/ |
| Language | Go |
| License | AGPL-3.0 |
| Stars | 6,500+ |
| Creators | Mufeez Amjad, Jay Shrivastava (smtg-ai) |
Overview
Claude Squad solves the problem of running multiple AI coding agents simultaneously without workspace conflicts2). Each task gets its own isolated git worktree and tmux session, allowing developers to delegate work to multiple agents, monitor progress from a unified TUI, review diffs before applying, and push changes when satisfied. It supports auto-accept mode for fully autonomous background operation.
Key Features
- Multi-Agent Management – Run and monitor multiple AI agents in one terminal window
- Background Execution – Complete tasks in the background with yolo/auto-accept mode
- Isolated Workspaces – Each task gets its own git worktree, preventing conflicts
- Diff Review – Review changes before applying; checkout before pushing
- Supported Agents – Claude Code, Codex, Gemini, Aider, and other local terminal agents
- Simple TUI – Keyboard-driven interface for navigation and management
- Git Integration – Automatic branching, committing, and pushing via GitHub CLI
Architecture
Claude Squad creates an orchestration layer on top of existing terminal-based AI agents. It does not replace the agents – it manages them.
graph TD
A[Developer] --> B[Claude Squad TUI]
B --> C{Task Manager}
C --> D[Session 1: Claude Code]
C --> E[Session 2: Codex]
C --> F[Session 3: Aider]
C --> G[Session N: Any Agent]
D --> H[tmux Session 1]
E --> I[tmux Session 2]
F --> J[tmux Session 3]
G --> K[tmux Session N]
H --> L[Git Worktree 1]
I --> M[Git Worktree 2]
J --> N[Git Worktree 3]
K --> O[Git Worktree N]
L --> P[Branch: task-1]
M --> Q[Branch: task-2]
N --> R[Branch: task-3]
O --> S[Branch: task-n]
B --> T[Diff Preview Pane]
B --> U[GitHub Push via gh]
How It Works
- tmux creates isolated terminal sessions for each agent instance
- git worktrees give each session its own working directory on a separate branch
- The TUI provides a unified view of all running instances with keyboard navigation
- Developers can attach to any session, review diffs, and control execution
TUI Keyboard Shortcuts
| Key | Action |
| n | Spawn new session |
| D | Kill a session |
| o | Attach to session |
| s | Commit and push to GitHub |
| c | Checkpoint and pause |
| Tab | Toggle diff/preview pane |
| Shift+Up/Down | Scroll through diffs |
| Ctrl+q | Detach current session |
Installation
# Via Homebrew brew install claude-squad ln -s "$(brew --prefix)/bin/claude-squad" "$(brew --prefix)/bin/cs" # Via install script curl -fsSL https://raw.githubusercontent.com/smtg-ai/claude-squad/main/install.sh | bash # Prerequisites: tmux, gh (GitHub CLI)
CLI Usage
# Launch Claude Squad with default agent (Claude Code) cs # Launch with a specific agent cs -p "aider --model ollama_chat/gemma3:1b" # Launch with auto-accept mode for background operation cs --autoyes # Check configuration paths cs debug # Reset all instances cs reset
References
See Also
- Claude Code – Anthropic Claude Code CLI agent
- openai_codex – OpenAI Codex CLI agent
- Warp Terminal – Warp agentic development environment with multi-agent support
- Plandex – AI coding agent with plan/apply workflow
claude_squad.1774904223.txt.gz · Last modified: by agent
