Skip to main content

General

Agents Squads is a framework for building transparent AI agent systems. Instead of black-box AI, you get agents defined in simple markdown files that your team can understand, modify, and own.
Most frameworks focus on capability. We focus on transparency and trust:
  • Agents are markdown files, not code
  • Full visibility into what agents do and why
  • No vendor lock-in—you own everything
  • Designed for teams, not just developers
No. Agents are defined in markdown. If you can write a document, you can create an agent. The CLI handles execution.
Currently optimized for Claude via Claude Code. Support for other models is planned.

Getting Started

npm install -g @agents-squads/cli
See the Quickstart for full setup instructions.
Create a markdown file in .agents/squads/<squad>/<agent>.md with Purpose, Inputs, Outputs, and Instructions sections. See Creating Agents.
Yes. Run squads init in your project root. It creates the .agents/ directory structure without touching your existing code.

Architecture

  • Agent: A single AI worker with a specific purpose (defined in a .md file)
  • Squad: A team of agents organized around a business domain (defined in SQUAD.md)
Squads share memory and coordinate on domain-specific tasks.
Memory is stored in .agents/memory/ as markdown files. When an agent runs, relevant memory is loaded into context. After execution, new learnings are stored. This gives agents continuity across sessions.
Agents in the same squad share memory, which enables coordination. For direct handoffs, you can chain agent executions or use an orchestrator pattern.

Pricing & Licensing

The CLI and framework are open source. You pay only for the LLM usage (Claude API costs).
Yes. Enterprise features include dedicated support, custom deployments, and SLA guarantees. Contact us for details.
Yes. The open source version is available under MIT license.

Troubleshooting

Common causes:
  1. Vague instructions — Be specific about steps and output format
  2. Missing context — Check if memory is loaded correctly
  3. Scope too broad — Break into smaller, focused agents
Run with squads run <agent> --dry-run to preview without executing.
Ensure you’re running squads memory update <squad> after significant work. Check that .agents/memory/ directory exists and has write permissions.
Verify installation: npm list -g @agents-squads/cliIf not installed: npm install -g @agents-squads/cli

Still Have Questions?