Squad
Open-source CLI that gives your repo a persistent, human-led AI dev team (lead, frontend, backend, tester) on GitHub Copilot and other coding agents
Overview
Squad bills itself as "human-led AI agent teams for any project." You describe what you are building, and one command creates a team of specialists (for example a lead, frontend, backend and tester) that live in your repo as plain files. Each agent keeps its own history and the team shares a decisions log, so the team remembers past work across sessions. It runs first on GitHub Copilot (CLI and VS Code chat), and also on Claude Code, Cursor, OpenAI Codex, OpenCode, Kiro and Google Antigravity. A watch mode called "Ralph" can poll GitHub issues and pick up work. Squad is maintained by Brady Gaster and Tamir Dresher. Note: trysquad.ai is a separate product from TrySquad, Inc. (see TrySquad), not this project.
The Verdict
Who Should Use Squad?
Best For
- GitHub Copilot users who want a multi-agent team instead of one assistant
- Developers who want team memory and decisions stored as reviewable files in the repo
- Teams that move between Copilot, Claude Code, Cursor and Codex and want one team setup
Not Ideal For
- Anyone uneasy about running an agent with broad auto-approve permissions
- Teams that need formal approval gates, cost or security review built into the flow
- Production use of the TypeScript SDK mode, which is still marked experimental
What's Great
- One command (
squad init) sets up a whole team in your repo - Persistent per-agent history and a shared decisions log survive across sessions
- Works with seven agent hosts, led by GitHub Copilot
- Governance options: file-write guards, PII scrubbing and reviewer lockout
- Free, MIT licensed, with very active development and strong community adoption (see badge)
Watch Out For
- The README recommends running with
--yolobecause Squad makes many tool calls - SDK-first mode is experimental with known bugs; markdown-first is the production default
- Per-agent model choice works only in the Copilot CLI and app; VS Code runs every agent on one model
- Needs Node.js 22.5 or newer
- The old interactive shell is deprecated in favor of
copilot --agent squad
Pricing
View all features & details
Key Features
- Specialist agent roles generated from your project description
- Team stored in a
.squad/folder: agent histories, decisions and ceremonies config - Parallel agents working on separate parts of a task
- "Ralph" watch mode that polls GitHub issues
- GitHub Agentic Workflows support
- Optional TypeScript SDK (
squad.config.ts) - Install via npm, Homebrew, WinGet or an install script
Use Cases
- Adding a standing AI dev team to an existing repo
- Splitting feature work across frontend, backend and test agents
- Working a GitHub issue backlog with agents
- Keeping agent context and decisions across many sessions
How It Compares
| Feature | Squad | Software Factory | BMAD-METHOD |
|---|---|---|---|
| Approach | Persistent repo-resident agent team | Agent team with approval gates | Agile persona agents and workflows |
| Agents supported | Copilot, Claude Code, Cursor, Codex, OpenCode, Kiro, Antigravity | Claude Code, Codex | Many AI IDEs and agents |
| Human gates | Human-led; governance hooks, no fixed gates | PRD, design and cost, release, plus irreversible actions | Checkpoints between phases |
| Team memory | Per-agent history plus shared decisions log | Per-project agent memory and factory brain | Project docs |
| License | MIT | Not stated | MIT |
Sources: Squad · Software Factory · BMAD-METHOD
User Reviews
Is this your tool? Add a badge