Docker Agent

Canonical version: Docker Agent.

Docker Agent is Docker's open-source framework for running teams of specialized AI agents that delegate to each other, configured entirely in YAML and shipped as OCI artifacts. Built on the cagent engine; runs with docker agent run config.yaml.

The bet: one generalist agent context-switching across a complex task does it worse than a root agent that delegates to focused sub-agents, each with its own model, instructions, and tools. Docker Agent gives you that hierarchy without building the orchestration loop yourself.

How it works

• Root + sub-agents. You talk to a root agent; it delegates to sub-agents based on the config. Hierarchies nest.
• Per-agent models. Each agent picks its own LLM and parameters from Anthropic (Claude), OpenAI, or Google.
• Isolated contexts. Agents keep separate knowledge instead of sharing one context window.
• Built-in tools. Todo lists, memory, task delegation out of the box.
• External tools via Model Context Protocol (MCP). Wire in MCP servers, filesystem, and shell.

Distribution is the Docker-native part

Package an agent config as an OCI artifact and push it to Docker Hub or any compatible registry, the same way you ship an image. That's how a team shares and versions agents.

Running it

1. Set API keys for your provider.
2. Write the YAML (models, roles, instructions, sub-agent list, toolsets).
3. docker agent run config.yaml, then interact in the terminal.

Ships in Docker Desktop 4.63+, or install via Homebrew, Winget, or a prebuilt binary in ~/.docker/cli-plugins.

References

• Documentation: https://docs.docker.com/ai/docker-agent/
• Source code: https://github.com/docker/docker-agent

Related

• Docker AI
• Docker
• Docker MCP Catalog
• Docker MCP Gateway
• Docker Model Runner
• Docker Compose for AI
• Gordon (Docker AI Agent)
• Model Context Protocol (MCP)
• AI Agents

About Sébastien

Ready to get to the next level?