docker/docker-agent

Docker Agent – AI agents as a Docker CLI plugin

What it is – Docker Agent is a genuine open‑source tool that lets you define, run, and share AI agents using a simple YAML file. It ships as a Docker CLI plugin (docker agent) so you can start an agent with the same command line you already use for containers.

How it works – You write a declarative configuration that specifies:

• The LLM model to use (OpenAI, Anthropic, Gemini, AWS Bedrock, Mistral, Docker Model Runner, etc.)
• A description and system prompt for the agent
• One or more toolsets (e.g., a DuckDuckGo search tool via the Model Context Protocol, RAG utilities, memory, todo, etc.)

Docker Agent then launches the model, wires the requested tools, and, if you define multiple agents, orchestrates them so they can delegate tasks to each other automatically.

Key capabilities

• Multi‑agent orchestration – Build a team of specialized agents that hand‑off work without writing code.
• Tool ecosystem – Built‑in utilities plus any MCP‑compatible service (local, remote, or Docker‑based).
• Provider‑agnostic – Works with cloud APIs and with locally‑run models via Docker Model Runner.
• YAML‑only configuration – Easy to version, share, and store in OCI registries.
• RAG support – Plug‑in retrieval using BM25, embeddings, hybrid search, and reranking.
• Package & share – Push an agent image to any OCI registry and run it anywhere with docker agent run <image>.

Installation

• Docker Desktop 4.63+ includes the plugin automatically.
• Homebrew: brew install docker-agent (or symlink the binary to ~/.docker/cli-plugins/docker-agent).
• Binary releases from the GitHub Releases page.
• Set at least one provider API key (e.g., export OPENAI_API_KEY=…).

Quick start

# Run the built‑in default agent
docker agent run

# Run an agent stored in an OCI registry
docker agent run myorg/agent:tag

# Create a new agent interactively
docker agent new

# Run a custom YAML definition
docker agent run agent.yaml

Documentation & community – Full docs are hosted at https://docker.github.io/docker-agent/ covering installation, model setup, configuration reference, TUI/CLI usage, MCP mode, RAG tools, and provider details. Community support lives on Docker Slack (the #docker-agent channel) and a collection of community projects, videos, and articles.

Typical use‑cases – Rapid prototyping of AI assistants, building multi‑agent pipelines for data‑analysis or automation, packaging reproducible AI workflows as OCI images, and experimenting with RAG or custom tool integrations without writing code.