Docker Launches docker-agent to Build and Run AI Agents

Docker Engineering has launched docker-agent, an open-source AI agent builder and runtime designed to simplify the creation and deployment of intelligent, collaborative AI systems. Operating as a Docker CLI plugin, the tool allows developers to define specialized AI agents using declarative YAML configurations without writing any code. This release aims to streamline how developers package, run, and share multi-agent architectures across different environments.

The runtime is highly flexible, supporting a wide range of major AI providers including OpenAI, Anthropic, Gemini, AWS Bedrock, Mistral, and xAI, as well as local models via the Docker Model Runner. It features a multi-agent architecture where specialized agents can automatically delegate tasks to one another. For tool integration, docker-agent connects with built-in utilities and any Model Context Protocol (MCP) server, whether it is hosted locally, remotely, or within a Docker container.

Practitioners can leverage advanced reasoning capabilities, such as built-in think, todo, and memory tools. The platform also supports pluggable retrieval-augmented generation (RAG) using BM25, embeddings, hybrid search, and reranking. Once built, these agents can be packaged and pushed to any OCI registry, allowing other developers to pull and run them anywhere.

For practitioners, docker-agent is already pre-installed in Docker Desktop version 4.63 and higher, accessible via the simple "docker agent" command. It can also be installed via Homebrew or downloaded as a binary release from GitHub. To begin, users only need to configure an API key, such as "OPENAI_API_KEY", or utilize the Docker Model Runner to execute models locally. This integration brings AI agent orchestration directly into the standard containerization workflow.