Agents Squads
Coordinate multi-model AI teams with persistent memory, verification, and feedback loops.
- Source repo
- agents-squads/squads-cli
- Stars
- ★ 56
- Last updated
- 1mo ago
- License
- MIT
- Primary language
- TypeScript
- FA score
- 63/100 · Some gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformClaude Code
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- An engineering lead wants planners, implementers, and verifiers to collaborate on one goal while retaining an auditable Git history.
- Not a fit if
- Teams unable to install Claude Code
- Teams requiring state management without Git or local files
- Users seeking a hosted, serverless platform
- Website
- agents-squads.com ↗ set on GitHub
- Maintenance
- Commits in the last 6 monthslast commit 2026-10-20latest release v0.10.1 (2026-10-20)
- Source review
- 63/100 · Some gaps
What does this agent do, and when should you use it?
Agents Squads is a TypeScript CLI framework for organizing AI teams across engineering, marketing, finance, and operations. It stores business context, system rules, squad identities, goals, learnings, and feedback as Markdown under .agents/ in a Git repository, with no database, server, or custom DSL. Before each run, it builds a role-sensitive context cascade from company strategy, squad goals, previous feedback, and active work; leads plan, workers execute, and verifiers inspect the result. The CLI shells out to native tools such as claude, gemini, and aider, allowing different agents to use different models. Users start squad or bounded tasks with squads run and resolve approval, rejection, or deferral decisions through squads inbox; the repository description also identifies status, memory, goals, feedback, and a dashboard as management features. Execution and state remain on the user's machine, although the setup requires Node.js, Git, Claude Code, and credentials for the selected AI providers.
Squads reads business context from .agents/BUSINESS_BRIEF.md, shared rules from .agents/config/SYSTEM.md, SQUAD.md and individual agent definitions from .agents/squads/, and strategy, goals, learnings, and feedback from .agents/memory/. It assembles these inputs into a role-tuned context cascade, keeping scanner contexts lighter while giving leads a broader view. squads run research performs a squad conversation covering plan, work, review, and verification, while squads run intelligence --task "Scan X" starts a directed, bounded task. During execution it shells out to native AI CLIs including claude, gemini, and aider, so roles can be routed to different models. Feedback from one cycle is injected into the next, while Markdown files and Git history retain team state and its audit trail. squads inbox gathers items that require the user to approve, reject, or defer them.
- An engineering lead wants planners, implementers, and verifiers to collaborate on one goal while retaining an auditable Git history.
- A research team wants a lower-cost model to scan, a deeper reasoning model to build, and another model to verify the output.
- A marketing or operations lead needs business context and squad goals to persist across runs so agents do not repeat completed work.
- A local-first team wants multi-agent workflows to run on its own machine and API keys without deploying a database or server.
- A manager wants a single inbox for agent work that still requires a human approval, rejection, or deferral decision.
Key setup facts
Pulled from the install and usage notes below. The project README has the current values.
- Install / run
npm install -g squads-cli
- Runtime
- Node.js 20
How do you install or deploy this agent?
Install Node.js 20 or newer, Git, and Claude Code first. Claude Code must be installed and authenticated with claude /login; other providers are optional. Then install the CLI and initialize a Git-backed workspace:
npm install -g squads-cli
mkdir my-workforce && cd my-workforce
git init
squads init
squads run demo hello-worldCheck the machine and required Claude Code setup with:
squads doctorFor source development, use:
git clone https://github.com/agents-squads/squads-cli.git
cd squads-cli
npm install
npm run build && npm link
npm testHow do you use this agent?
After initialization and the demo run, start the first practical agent:
squads run research/analystRun the complete research squad conversation, including planning, work, review, and verification:
squads run researchGive the intelligence squad a bounded task:
squads run intelligence --task "Scan X"Open the queue of decisions awaiting human action:
squads inboxSquad definitions, business context, rules, and memory live under .agents/ and can be managed through Markdown and normal Git workflows.
What are this agent's strengths and limitations?
- Markdown and Git provide both state and an audit trail without requiring a database, persistent server, or custom DSL.
- The documented lead-worker-verifier structure implements a plan-work-review-verify loop and feeds each cycle's feedback into the next.
- Native
claude,gemini, andaiderCLI execution permits different models to be assigned to different roles. - The context cascade varies by role, supporting lightweight scanners without depriving leads of broader team context.
- The project states that execution, data, and keys stay on the user's machine, with no project-specific cloud login or telemetry.
- The first run explicitly requires Claude Code to be installed and authenticated, even though other providers are available as options.
- Adoption requires Node.js 20 or newer, Git, local filesystem access, and provider credentials, making setup heavier than a hosted chat product.
- Markdown and Git are the state layer, so teams requiring database transactions, centralized service state, or Git-free workflows must change their operating model.
- Mixed-model use depends on multiple external native CLIs, leaving provider installation, authentication, and usage charges to the user.
- The supplied material gives no dashboard launch command, interface details, or deployment procedure, so that feature cannot be evaluated in depth.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Agents Squads This agent | 63 · Some gaps | CLIFree + model costs | ★ 56 | 1mo ago | TypeScript | Claude Code |
| AI DevOps Framework (aidevops) | 63 · Some gaps | CLIFree + model costs | ★ 468 | today | Shell | OpenAI API · Claude API |
| Manifold | 55 · Major gaps | Self-hosted serviceFree | ★ 504 | 1mo ago | Go | OpenAI API · Claude API |
| VCP ToolBox | 50 · Major gaps | Self-hosted serviceFree + model costs | ★ 2.5k | today | JavaScript | OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
The docs describe a limited default toolchain with per-agent opt-in tools; inbox decisions, PR gates, timeouts, and budget caps provide approval and constraint mechanisms. The README and configuration guide describe local storage, anonymous telemetry fields and opt-outs, and secret and commit/push handling. Deductions: agents can still run commands, and the files recommend rather than demonstrate sandbox isolation; telemetry destination and transport details are unspecified; security advice does not establish encryption or enforced access controls; the source sample does not show dependency audit results. Git worktrees, auto-saving, and opt-in commits/pushes provide recovery cues, but the recovery procedure is not fully described. Author and maintenance contact details are present, but publisher identity is unverified and is not treated as established.
The README, package metadata, and CI configuration show Node 20+, native AI CLI dependencies, lint/typecheck/build/test and install smoke workflows; a source-contract test checks for a nonzero exit code when a CLI is missing. Deduction: the README example describes a local commit without a push, and the configuration guide explains opt-in auto-commit; the seed skill also claims every run uses its own worktree and never leaves loose working-tree changes. The supplied material does not fully resolve the scope of these statements. External CLI and API availability depends on the user's environment, and static evidence cannot establish successful operation.
The README, configuration guide, and skill address human operators and agents across several business domains and providers such as Claude and Gemini; they document task, dry-run, timeout, budget, provider, tool-grant, and local-default options. Deductions: the product primarily targets Node, Git, and terminal users; boundaries for nontechnical users and environments without supported native CLIs are limited, and configurable tools allow capabilities to expand.
Repository guidance, README, configuration docs, package metadata, and security policy provide a recognizable structure, install and development commands, examples, MIT license text, and a vulnerability reporting channel. Deductions: some command examples in the supplied skill excerpt are incomplete, FAQ coverage is limited, and although a 0.x and package version are shown, the packet provides no change history. AGENTS.md identifies itself as contributor guidance, while product operating guidance is spread across the README, docs, and templates.
The CLI manages squads through status, memory, goals, feedback, inbox, and dashboard; JSON output, reviewable artifacts, cost/time bounds, and multi-provider orchestration indicate practical value for automation and human decisions. Deductions: utility and cost-saving claims are primarily documented assertions without independent outcome evidence in this sample; reliance on several local CLIs and configuration adds setup and maintenance cost.
Specific feature claims can be traced across the README, configuration guide, package metadata, seed skill, CI configuration, and test source; CI declares test workflows and the test files show some assertions. Deductions: this is a static sample review, so CI configuration and test source do not establish that this revision passed; some product-effect claims lack independent corroboration. The rationale distinguishes direct file evidence from conclusions inferred from the docs.
- Review agent definitions, granted tools, and available credentials before running; the docs say agents can execute system commands, and isolation depends on the deployment environment.
- Treat project and user-level .env files and memory files as sensitive; the docs recommend ignoring .env, restricting file permissions, and handling memory carefully.
- Auto-push also pushes other unpushed commits on the current branch; inspect branch state before enabling it.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
What's Changed * release: v0.10.1 — security patch by @agents-squads[bot] in https://github.com/agents-squads/squads-cli/pull/1326 Full Changelog: https://github.com/agents-squads/squads-cli/compare/v0.10.0...v0.10.1View on GitHub ↗ -
Release notes excerpt
What's Changed * release: v0.10.0 by @agents-squads[bot] in https://github.com/agents-squads/squads-cli/pull/1286 Full Changelog: https://github.com/agents-squads/squads-cli/compare/v0.9.0...v0.10.0View on GitHub ↗ -
Release notes excerpt
What's Changed * release: v0.9.0 — the loop release by @kokevidaurre in https://github.com/agents-squads/squads-cli/pull/1033 Full Changelog: https://github.com/agents-squads/squads-cli/compare/v0.8.3...v0.9.0View on GitHub ↗
FAQ
Is the software itself paid?
Can I use it without Claude Code?
claude /login before the first run. Gemini, Aider, and other CLIs can also be invoked, but the source does not document removing the initial Claude Code requirement.Does it need a database or background server?
Does project data go to the vendor's cloud?
How do I diagnose missing dependencies or authentication?
squads doctor. It checks the machine and Claude Code setup and reports what is missing; further recovery behavior is not documented in the supplied material.Is Agents Squads open source?
Is Agents Squads safe to use?
Is Agents Squads still maintained?
Who makes Agents Squads?
What is Agents Squads's official website?
Thanks — we use this to improve the page.