Agent Swarm logo

AI AGENTS

Agent Swarm

by desplega-ai

agentcoding-agentMCP

Overview

Open-source, MCP-powered framework for orchestrating teams of autonomous AI coding agents with lead-worker delegation, Docker isolation, GitHub integrations, and PR-oriented delivery.

Details

Agent Swarm is an open-source, MCP-powered agent company OS built by desplega.sh for orchestrating teams of autonomous AI coding agents. Its documentation describes a hub-and-spoke MCP API server that coordinates Docker-isolated worker agents, uses SQLite state and vector memory, supports Claude Code or pi-mono harnesses, and includes integrations, scheduling, workflows, and heartbeat recovery. The project is positioned around lead-worker delegation and PR-oriented delivery with GitHub integration.

When to Use

Use Agent Swarm when you need to coordinate multiple autonomous coding agents around lead-worker delegation and pull-request-oriented delivery. Use it when Docker-isolated worker agents MCP-based orchestration state/memory scheduling workflows and GitHub-oriented delivery are important to your agent workflow.

Getting Started

  1. Visit the official Agent Swarm site at https://www.agent-swarm.dev/ for the product overview and links to docs and GitHub.
  2. Follow the official Getting Started guide at https://docs.agent-swarm.dev/docs/getting-started
  3. which provides installation commands for @desplega.ai/agent-swarm.
  4. Review the architecture overview to understand the MCP API server
  5. Docker-isolated workers
  6. SQLite state
  7. vector memory
  8. harness options
  9. integrations
  10. scheduling
  11. workflows
  12. and heartbeat recovery.
  13. Inspect the GitHub repository at https://github.com/desplega-ai/agent-swarm before adopting it in a production workflow.

Key Features

  • MCP-powered multi-agent orchestration for autonomous AI coding agents
  • Lead-worker delegation model for coordinating agent teams
  • Docker-isolated worker agents
  • GitHub integrations and PR-oriented delivery
  • SQLite state and vector memory
  • Scheduling
  • workflows
  • integrations
  • and heartbeat recovery described in the architecture documentation
  • Supports Claude Code or pi-mono harnesses according to the architecture page

Capabilities

  • multi-agent orchestration
  • coding agent coordination
  • MCP server orchestration
  • Docker-isolated agent execution
  • GitHub-integrated PR delivery
  • workflow scheduling

Last updated Jun 27, 2026