TECHNOLOGY
Docker releases docker-agent, a CLI plugin for building and running AI agents from YAML files
Docker Engineering's docker-agent lets users define AI agents in a YAML config and run them with the docker agent command. According to its project page, it supports multiple model providers and teams of agents.
Image: IntraGoals Media
Docker has a project called docker-agent that lets people build and run AI agents from a YAML configuration file. It works as a plugin for the docker command line tool, so you start it with docker agent.
The project page says you can create agents that work together on complex problems without writing code. An agent is an AI program that can use tools to get a task done. YAML is a simple text format for settings, and here it describes what each agent should do.
The example on the page is short. It defines one agent called root, sets its model to openai/gpt-5-mini, and gives it a plain instruction to be helpful, accurate and concise. It also attaches a tool set of type mcp that points to docker:duckduckgo. You then run the file with docker agent run agent.yaml.
The page lists several features. You can set up teams of specialized agents that hand tasks to each other automatically. Agents can use built-in tools or any MCP server, whether it runs locally, remotely or in Docker. MCP, short for Model Context Protocol, is a common way to connect AI models to outside tools.
Docker says the tool is not tied to one AI provider. The page names OpenAI, Anthropic, Gemini, AWS Bedrock, Mistral, xAI and Docker Model Runner, and adds "and more". It also lists built-in think, todo and memory tools for reasoning, and retrieval support (RAG) using BM25, embeddings, hybrid search and reranking. RAG means the agent looks up relevant material before it answers.
Agents can also be packaged and shared. According to the page, you can push them to any OCI registry, the same kind of registry used for container images, and pull and run them elsewhere.
There are three ways to install it. Docker Desktop 4.63 or later has the plugin pre-installed. You can also use Homebrew with brew install docker-agent, or download binaries from GitHub Releases. For the manual routes, the page says to symlink the binary to ~/.docker/cli-plugins/docker-agent so that docker agent works.
You need at least one API key, such as OPENAI_API_KEY or ANTHROPIC_API_KEY, or you can use Docker Model Runner to run local models. The page points to a Set Up a Model guide for both options.