An official Docker AI Agent builder and runtime that enables multi-agent orchestration, model-agnostic scheduling, and OCI-standard distribution through declarative YAML/HCL configuration.
Docker Agent is an official Docker tool for building and running AI Agents, centered on declarative YAML/HCL configuration that eliminates the need for framework code when creating, running, and sharing AI Agents.
Multi-Agent Orchestration & Tool Ecosystem Create specialized agent teams with automatic task delegation through coordinators. Built-in tools include filesystem, shell, think (chain-of-thought), todo (task planning), and memory (memory management), with support for any MCP Server (local/remote/Dockerized) as additional tool sources.
Model Agnosticism & Local Execution Supports OpenAI, Anthropic, Google Gemini, AWS Bedrock, Mistral, xAI, Docker Model Runner, and custom providers. When paired with Docker Model Runner, models can run entirely locally without external APIs.
OCI-Standard Distribution Push agent configurations to any OCI-compatible registry and pull/run them like container images, enabling enterprise-grade standardized agent distribution. A built-in Agent Catalog allows direct execution of remote community agents.
Multiple Interfaces & Protocol Support Five interaction modes: TUI (interactive terminal), CLI (scripted), HTTP API (programmatic integration), MCP mode (ecosystem interoperability), and A2A Server mode (agent-to-agent communication). Supports RAG with BM25, embeddings, hybrid search, and reranking.
Engineering Features Built in Go (98.8%) with cross-platform support for Linux/macOS/Windows (amd64/arm64). Configuration follows strict JSON Schema with versioned management. Built-in OpenTelemetry tracing and debug logging. Delivered as a Docker CLI plugin, standalone binary, or Docker image; pre-installed in Docker Desktop 4.63+. Latest version v1.57.0, Apache-2.0 license.