Skip to content

Documentation

AgentCrew is an open-source AI agent orchestrator. Just as Kubernetes orchestrates containers regardless of what application runs inside, AgentCrew orchestrates AI agents regardless of their purpose.

Create teams of specialized agents for any domain: software engineering, marketing, finance, data analysis, content creation, operations, or anything else. Agent specialization is driven entirely by documentation: write Markdown instructions describing what each agent should do, and AgentCrew handles deployment, communication, and coordination.

Supports multiple AI providers including Claude Code and OpenCode. Choose the provider that best fits your needs when creating a team.

Key Concepts

Teams

A team is a group of AI agents that work together on a shared mission. Each team runs in its own isolated environment (Docker containers with a dedicated NATS messaging bus). You can have multiple teams running simultaneously, each focused on a different project or domain.

Agents: Leader + Workers

Every team has exactly one leader and one or more workers. The leader receives your messages, breaks down the work, and delegates to workers. Workers are specialist agents defined by Markdown files. Their role, instructions, and behavior are entirely determined by the documentation you write. A worker can be anything: a "Content Strategist", a "Financial Analyst", a "Backend Developer", or a "Legal Reviewer". The same orchestration platform, infinite use cases.

Documentation-Driven Specialization

The core of AgentCrew is that agents are defined by text. The system generates structured files that determine each agent's behavior:

  • CLAUDE.md: The leader's instructions, team context, and delegation protocol.
  • agents/*.md: One file per worker, with YAML frontmatter (name, model, skills) and a Markdown body containing detailed instructions.
  • Skills: Installable extensions from GitHub repos that give agents additional capabilities.

This documentation-driven approach means you can create agents for any purpose without writing a single line of code. Describe what the agent should do, and it becomes that specialist.

Skills

Skills are extensions that give agents specialized capabilities sourced from GitHub repositories. Skills can be installed on individual workers (agent-specific) or on the leader (global, shared with all agents in the team). They allow you to extend what your agents can do without modifying their core instructions.

Workspaces

A workspace is a directory mounted into the agent container at /workspace. It gives agents access to your project files, including source code, documents, data, and configuration, so they can read, edit, and create files on your behalf. Workspaces persist across sessions.

Documentation Sections

  • Quick Start: Get AgentCrew running in one command with Docker Compose.
  • Configuration: Environment variables, application settings, and storage options.
  • Providers: Supported AI providers (Claude Code, OpenCode) and how to configure each.
  • Skills: How to extend agent capabilities with installable skills.
  • Schedules: Automate recurring tasks with cron-based schedules and track run history.
  • Webhooks: Trigger agent teams from external systems via HTTP with token-authenticated endpoints and prompt templates.
  • Post-Actions: Reusable HTTP callbacks that fire automatically after webhook or schedule runs to notify external systems.
  • MCP Servers: Connect agents to external tools (databases, APIs, Kubernetes) via the Model Context Protocol.
  • Authentication: Built-in user authentication, organizations, invites, and role-based access control.
  • Architecture: System components, messaging, container internals, and runtime design.

License

AgentCrew is open source software licensed under the GNU Affero General Public License v3.0 (AGPL-3.0).