4.1. Squad Overview¶

Squad lets you create a team of AI agents that work together on complex tasks. A planner agent breaks your request into subtasks, assigns them to specialized agents, and coordinates execution in dependency-respecting waves — all within a shared workspace.

What is a Squad?¶

A Squad is a self-contained workspace where multiple AI agents collaborate:

flowchart TD
    U[User Request] --> P[Planner Agent]
    P --> T1[Task 1: Frontend]
    P --> T2[Task 2: Backend]
    P --> T3[Task 3: Review]
    T1 --> D1[Developer Agent]
    T2 --> D2[Developer Agent]
    T1 -.->|depends on| T3
    T2 -.->|depends on| T3
    T3 --> R[Reviewer Agent]
    R --> F[Final Result]

Each Squad includes:

• Agents — specialized AI agents with distinct roles, system prompts, and tool access
• Workspace — a dedicated directory where agents read and write files
• Memory — per-agent markdown memory files for context retention across tasks
• Plan — a structured task graph with dependencies and priorities
• Budget — token and execution limits that prevent runaway costs

How it Works¶

1. Create a Squad¶

Start from a template (Full-Stack Dev Team, Content Team, Research Team, etc.) or build a squad from scratch. Assign a workspace directory and configure each agent's role, system prompt, and tools.

2. Submit a Request¶

Describe your goal in plain language, for example: "Implement a user authentication module with login, registration, and password reset."

3. Planning¶

The designated planner agent analyzes your request and generates a structured plan:

• Breaks the work into discrete tasks
• Assigns each task to the most suitable agent
• Defines task dependencies (which tasks must finish before others can start)
• Sets priorities (low, medium, high, critical)

4. Approval¶

Review the plan before execution begins. You can approve, modify, or reject the plan. For recurring workflows, enable auto-approve to skip this step.

5. Wave Execution¶

Tasks execute in waves — groups of tasks whose dependencies are all satisfied:

Each wave runs its tasks concurrently, maximizing parallelism while respecting the dependency order.

6. Aggregation¶

After all tasks complete, results are collected and synthesized into a final deliverable.

Key Benefits¶

• Specialization — Each agent focuses on what it does best (planning, coding, reviewing, writing).

• Parallelism — Independent tasks run simultaneously, reducing total execution time.

• Transparency — Real-time monitoring shows exactly what each agent is doing and how many tokens are consumed.

• Safety — Budget controls prevent token overuse, and emergency stops halt execution instantly.

• Persistence — Workspaces, plans, and execution history are saved for review and reuse.

Squad vs Cowork¶