↓ Skip to main content
9 – 17 UHR +49 8031 3508270 LUITPOLDSTR. 9, 83022 ROSENHEIM
DE / EN
Orchestrator smtg-ai (open-source project) Terminal (TUI)

Claude Squad

Claude Squad: open-source terminal app that runs Claude Code, Codex, Gemini CLI or Aider in parallel in tmux sessions and separate Git worktrees.

Licence AGPL-3.0
First release April 2025
Interfaces Terminal UI (cs command)
Platforms macOS, Linux

The seven building blocks in Claude Squad

How Claude Squad implements the building blocks of an AI harness

  1. 1

    Agent loop

    Claude Squad does not run an agent loop of its own; for every instance, it starts the chosen agent CLI, `claude` by default, in its own tmux session.

  2. 2

    Tools and execution environment

    Tools come from the respective coding agent; Claude Squad adds no tools of its own, no MCP servers and no API, only keyboard commands in the interface for creating, attaching, reviewing, committing and pushing.

  3. 3

    Context management

    Every instance works in its own Git worktree under `~/.claude-squad/worktrees` on its own branch, whose prefix is the user name by default.

  4. 4

    State and memory

    Claude Squad stores instances locally and restores them at the next start; with `c` you pause an instance after a commit, with `r` you resume it. Since v1.0.20, according to the release notes, instances whose tmux session ended with the server can also be recovered.

  5. 5

    Permissions

    By default, Claude Squad makes no approval decisions. The experimental `-y` mode (`auto_yes`) automatically confirms detected approval prompts from Claude Code, Aider and Gemini CLI by pressing Enter, also through a background process.

  6. 6

    Verification and feedback

    A diff tab shows the changes of each instance; with `c` you commit locally to check out the branch, with `s` you commit and push the branch to GitHub, and the actual review runs in your usual pull request process.

  7. 7

    Orchestration

    Claude Squad's core contribution: several agents in parallel, one tmux session and one worktree per instance, with a list that shows which instance is working or ready. Agents do not control each other.

Models and model access

Which models the harness uses and how they are obtained

Models and model access
Models Model-agnostic: Claude Squad starts the agents; the models are configured there
Access routes Model-agnostic (depends on the coding agent started)
Free choice of model Yes
Custom endpoint / LLM gateway No
Claude Squad does not call language models itself; the coding agent that is started determines model, provider and endpoint. The default program is `claude`; you start other agents with `-p` or via profiles.
Features
Resume sessions
Agents supported by herdr

Enterprise use

Data processing, controls and pricing model according to the vendor

Data processing and location

Claude Squad runs entirely locally; configuration, instance state and worktrees are stored under `~/.claude-squad`. Network access happens via Git and the GitHub CLI when you push a branch, and via the launched agents, which send source code and prompts to their model providers.

Permissions and auto-accept

Auto-accept is off by default. With `-y` or `auto_yes` in the configuration, Claude Squad confirms detected approval prompts from Claude Code, Aider and Gemini CLI by pressing Enter, and after you quit the interface it keeps doing so through a background process. The README marks the mode as experimental; a worktree only isolates files, not processes, network access or credentials.

Telemetry

The project does not document any telemetry. Based on our review of the v1.0.20 source code, Claude Squad contains no network calls of its own for analytics or error reports.

Identity and team

There is no user management, SSO or central admin control; every developer maintains their own `config.json`. Sign-in to GitHub runs through the existing GitHub CLI.

License, price and maintenance

Claude Squad is free and licensed under AGPL-3.0; anyone who distributes a modified version or makes it available over a network must disclose its source code. The latest release, v1.0.20, dates from 20 August 2026, and the latest commit from 13 November 2026 (GitHub, as of 22 November 2026); the project does not document commercial support.

Strengths
  • Lean Go binary with a terminal interface, without Electron and without cloud components
  • One tmux session and one Git worktree with its own branch per instance
  • Diff view per instance plus commit, pause and push directly from the interface
  • Any terminal CLI can be started via `-p` or profiles
  • No telemetry in the source code, all data stays local
  • Free and installable via Homebrew
Limitations
  • Auto-accept (`-y`) confirms approval prompts without review, also through a background process
  • State detection only via screen text and only for Claude Code, Aider and Gemini CLI
  • No API, no hooks and no agent-to-agent control
  • Requires tmux and the GitHub CLI; Windows is not documented
  • Push feature is designed for GitHub
  • No new release since August 2026 (as of 22 November 2026)
  • AGPL-3.0 can conflict with some companies' open-source policies

Use cases

Typical tasks for this harness

Hand several tasks to Claude Code or Codex in parallel, each on its own branch
Review the results of parallel agents in the diff before they are pushed
Let long-running tasks continue in the background and reattach later
Put different agents on the same codebase via profiles
Test local models via Aider and Ollama in a separate instance

Claude Squad is an open-source terminal app written in Go that runs several AI coding agents such as Claude Code, Codex, Gemini CLI or Aider in parallel. Every instance runs in its own tmux session and its own Git worktree with its own branch; you review changes in the diff before you commit and push them. As of 22 November 2026.

What is Claude Squad?

Claude Squad is a free, open-source project of the GitHub organization smtg-ai, licensed under AGPL-3.0. The smtg-ai/claude-squad repository was created in March 2025, and the first release, v1.0.0, came out on 3 April 2025. The current version is v1.0.20 from 20 August 2026 according to the release list, the latest commit dates from 13 November 2026, and the repository has 8,848 GitHub stars (GitHub API, 22 November 2026). Claude Squad is installed via Homebrew (brew install claude-squad) or an install script as the cs command. It requires tmux and the GitHub CLI gh. macOS and Linux are documented; the release files do include Windows builds, but the README does not describe Windows, and the install script refers to WSL for tmux on Windows.

Claude Squad as an AI harness

As an AI harness, Claude Squad covers only the orchestration building block: it starts agents, isolates their work and shows their state in a list. According to the README, the tool rests on three parts: tmux for a separate terminal session per agent, Git worktrees for a separate branch per session, and a lean terminal interface. With n you create a new instance, with N directly with a prompt. With Enter or o you attach to an instance to give further instructions, and ctrl-q detaches again. A preview tab shows the terminal output, a diff tab the changes in the worktree. c commits the changes and pauses the instance so you can check out the branch locally, r resumes it, and s commits and pushes the branch to GitHub. Profiles in ~/.claude-squad/config.json define which agents you can choose when creating an instance. Claude Squad offers no hooks, plugins, MCP or API for control by other agents.

Models and model access

Claude Squad is model-agnostic: it starts the chosen agent CLI, and that agent determines model, provider and access route. The default program is claude; with -p or a profile you start other agents, and the README gives aider --model ollama_chat/gemma3:1b as an example for a local model via Ollama. For Codex, the README refers to the OPENAI_API_KEY environment variable. Subscription use is not documented explicitly; since Claude Squad starts the agents under your user account, an existing login of the agent stays in effect. Whether model access runs through a central proxy such as innFactory’s AI Gateway depends on the configuration of the respective agent.

Company use: data protection and governance

Claude Squad runs entirely locally; the critical point for companies is the auto-accept mode. With the -y parameter or the auto_yes setting, Claude Squad automatically confirms detected approval prompts from Claude Code, Aider and Gemini CLI by sending Enter. When you quit the interface, it starts a background process for this that keeps monitoring all instances. The README marks the mode as experimental, and it is off by default. A worktree only isolates files in the repository, not processes, network access or credentials; in companies, auto-accept should therefore stay off or only run in an isolated environment.

Configuration, instance state and worktrees are stored under ~/.claude-squad. The project documents no telemetry, and based on our review of the v1.0.20 source code, Claude Squad opens no network connections apart from Git and the GitHub CLI. There is no user management, SSO or central policy. The AGPL-3.0 license permits internal use; anyone who distributes a modified version or makes it available over a network must disclose its source code. Check compatibility with your open-source policy. Between the release on 20 August 2026 and 22 November 2026, there was only one further commit. Source code and prompts of the launched agents go to their model providers and belong in your records of processing.

Which agents does Claude Squad support?

Claude Squad starts any CLI that runs in a terminal via the -p parameter or a profile. The README names Claude Code as the default as well as Codex, Gemini CLI and Aider; the project website names Claude Code, Codex and Aider, and the repository description also names OpenCode and Amp. According to the source code, Claude Squad detects approval prompts via the screen text only for Claude Code, Aider and Gemini CLI, and auto-accept only applies there. Other agents run as ordinary terminal processes in their instance.

How it fits: Claude Squad, herdr, Orca and the AI Gateway

Claude Squad is the leanest approach among the orchestrators: tmux and Git worktrees with an interface on top, without cloud services and without telemetry. herdr also relies on the terminal, detects the state of many agents more precisely and offers a CLI and socket API through which agents control each other. Orca shares the “one task, one worktree” principle with Claude Squad but brings a desktop app with editor, browser and pull requests. cmux is a native macOS terminal with notifications and a socket API. You will find a detailed comparison in the article herdr vs. Orca. None of the orchestrators captures developers’ costs and agent time across teams and projects; the AI Gateway does that in proxy and companion mode. For business users outside development, CompanyGPT is the right harness; our AI harness overview gives an overview.

From developer harness to enterprise harness

Coding harnesses are tools for developers. Business users, cost control and governance need two more layers.

CompanyGPT – the harness for business users

CompanyGPT gives any language model tools via MCP servers, integrations and n8n, company knowledge via companyRAG, roles from Entra ID or Keycloak, and audit logs – operated in your own cloud tenant (Azure, AWS or Google Cloud) or sovereignly on STACKIT.

Discover CompanyGPT
AI Gateway – cost and governance for every agent

The AI Gateway bundles the model access of your tools and coding agents behind an OpenAI-compatible API: sign-in via Entra ID, Keycloak or Cognito, budgets and hard limits per team and user, cost attribution per project and repository via the Companion, and logging of every request.

View the AI Gateway

Frequently Asked Questions

What is Claude Squad?

Claude Squad is a free, open-source terminal app written in Go that manages several AI coding agents in parallel. Every instance gets its own tmux session and its own Git worktree with its own branch, so parallel agents do not get in each other's way in the repository. A list shows the state of all instances, and you review changes in the diff and commit or push directly from the interface.

Is Claude Squad open source and free?

Yes. The smtg-ai/claude-squad repository is licensed under AGPL-3.0, and the software is free and available via Homebrew or an install script. The project does not document a commercial offer or support. Anyone who modifies Claude Squad and distributes it or makes it available over a network must observe the AGPL obligations; contributions to the project require a contributor license agreement.

Which coding agents does Claude Squad support?

By default, Claude Squad starts Claude Code. According to the README, Codex, Gemini CLI and Aider can be started as well, and the project description on GitHub also names OpenCode and Amp. In principle, any terminal CLI runs via the `-p` parameter or a profile. According to the source code, Claude Squad only detects approval prompts for Claude Code, Aider and Gemini CLI; other agents run without this detection.

Does Claude Squad use my Claude or ChatGPT subscription?

Claude Squad starts the agent CLIs in tmux sessions under your user account, so an existing login of the agent stays in effect, including a subscription. Subscription use is not documented explicitly; for Codex, the README names the `OPENAI_API_KEY` environment variable. Claude Squad itself manages no accounts, shows no usage and makes no model calls of its own.

Can Claude Squad be used in a GDPR-compliant way?

Claude Squad itself processes everything locally: configuration, instances and worktrees are stored in the `~/.claude-squad` directory. The project documents no telemetry, and based on our review of the v1.0.20 source code, there are no network calls apart from Git and the GitHub CLI. What matters for data protection are the launched agents, which send source code and prompts to their model providers, and the push to GitHub; both belong in your records of processing.

Does Claude Squad work with innFactory's AI Gateway?

Yes, through the agents it launches. Claude Squad makes no model calls of its own. Claude Code, Codex or OpenCode can be routed through the AI Gateway via configuration, and that configuration also applies in Claude Squad's tmux sessions. The gateway's companion measures tokens and agent time from the local session logs, including subscription usage and regardless of whether Claude Squad started the session.

Introduce coding agents safely in your company

We support selection, model access in the EU, permission design and cost control.