cagent

Docker cagent is an official AI agent builder and runtime framework from Docker. It enables zero-code agent development through declarative YAML configuration, lowering the barrier to AI agent creation.

Core Features#

Multi-Agent Orchestration#

• Define agent teams with coordinators and executors
• Automatic task delegation and coordination via orchestrator mode
• Simulate real-world dev team collaboration (PMs, Designers, Engineers, QA)

Rich Tool Ecosystem#

• Built-in Tools: filesystem, shell, memory, think, todo
• MCP Protocol Support: Integrate any MCP (Model Context Protocol) server
• Docker Integration: Native support for containerized MCP tools (e.g., docker:duckduckgo)

Multi-Model Provider Support#

• OpenAI, Anthropic (Claude), Google Gemini
• AWS Bedrock, Mistral, xAI
• Docker Model Runner (local models)

RAG Capabilities#

• BM25 keyword retrieval and Embeddings vector search
• Hybrid search and reranking mechanisms

OCI Image Distribution#

• Package agents as OCI-compliant container images
• Push to standard registries like Docker Hub
• Pull and run agents just like container images

Multiple Interfaces#

• TUI: Interactive terminal user interface
• CLI: Command-line tools (cagent run, cagent new)
• HTTP API: For integration into other applications
• MCP Mode: Interoperate with other tools

Installation#

• Pre-installed in Docker Desktop 4.49+
• Homebrew: brew install cagent
• Binary download from GitHub Releases

Configuration Example#

agents:
  root:
    model: openai/gpt-4o
    description: A helpful AI assistant
    instruction: |
      You are a knowledgeable assistant.
    toolsets:
      - type: filesystem
      - type: mcp
        ref: docker:duckduckgo
    sub_agents: [designer]

Use Cases#

• CLI assistants and code assistants
• DevOps automation (Kubernetes debugging, Docker container management)
• Multi-role collaboration simulation and documentation generation
• Third-party service integration (GitHub, Notion, Airbnb)