smtg-ai/claude-squad
Developer ToolsA terminal TUI that runs multiple AI coding agents in isolated tmux and git worktree workspaces.
What it does
Claude Squad is a terminal app for running several local AI coding agents at once. Instead of opening multiple terminals by hand, you manage sessions in one TUI, and each task gets its own tmux session plus a git worktree, so agents work on separate branches instead of editing the same checkout.
It supports Claude Code by default and can launch Codex, Gemini, Aider, or any other local agent through a shell command. The interface lets you create sessions (n), attach and reprompt (Enter/o), detach (Ctrl-Q), commit and push (s), or commit and pause a session (c) so you can inspect changes before moving them forward.
Install and first run
Prerequisites are tmux and the GitHub CLI gh (tmux install notes, GitHub CLI). Either install through Homebrew:
brew install claude-squad
ln -s "$(brew --prefix)/bin/claude-squad" "$(brew --prefix)/bin/cs"
or use the manual install script, which places cs in ~/.local/bin:
curl -fsSL https://raw.githubusercontent.com/smtg-ai/claude-squad/main/install.sh | bash
Then start it from a git repository:
cs
By default this launches the claude program. Use cs -p "codex" for Codex (after exporting OPENAI_API_KEY), cs -p "gemini" for Gemini, or another program command for Aider and other agents. Create a session, attach to it, and when work looks right use the commit/checkout actions to keep changes.
Configuration
Settings live in ~/.claude-squad/config.json; run cs debug to print the exact path. A profiles array defines named launch commands, and default_program picks one by default. When multiple profiles exist, the session-creation overlay shows a picker.
{
"default_program": "claude",
"profiles": [
{ "name": "claude", "program": "claude" },
{ "name": "codex", "program": "codex" },
{ "name": "aider", "program": "aider --model ollama_chat/gemma3:1b" }
]
}
Constraints
Each agent must already be installed and authenticated in your shell; Claude Squad launches commands rather than providing model access. The -y/--autoyes flag is marked experimental and auto-accepts prompts for Claude Code and Aider, so background runs should be reviewed before changes are pushed. It requires tmux and git worktrees, and Homebrew users must create the cs symlink themselves. A session timeout can occur when the underlying agent is out of date, so keep it current. The project is licensed under AGPL-3.0 (license).
HotRepo's take
On 2026-11-12, the repository recorded 8,789 stars and 631 forks, gaining 9 stars on the Go daily list, where it reached position 13. That is a single snapshot, not a trend, but it shows steady rather than explosive interest. The project's appeal is practical: it treats parallel agent work as a workspace-management problem by combining tmux isolation, git worktrees, and a review-and-commit loop in one TUI, rather than as a new model or chat client.
Written by AI from the project's README, updated . The project's own docs are the source of truth.
Hacker News
You might also like
- max-sixty/worktrunkA CLI that turns Git worktrees into branch-like commands, so parallel AI coding agents each get an isolated directory.
- openai/codexA terminal coding agent from OpenAI that reads your repo, edits files and runs commands from a prompt.
- getpaseo/paseoOrchestrate multiple coding agent CLIs across desktop, mobile, web, and terminal from one self-hosted daemon.
- Yeachan-Heo/oh-my-claudecodeRun a coordinated team of Claude Code agents on one coding task instead of prompting a single agent repeatedly.
- ZiChuanLan/agent-purgeReplace or append the system prompt for several coding agents from one local persona, with reversible switches.