🤖 Docker Agent: The AI Agent Builder & Runtime

Docker Logo

Developed by Docker Engineering, the docker-agent is a powerful framework designed to build, execute, and distribute AI agents. By utilizing a declarative YAML configuration and a robust tool ecosystem, it enables the creation of multi-agent systems capable of solving intricate problems.

The Core Promise: Create and deploy intelligent AI agents that collaborate seamlessly—~~without writing a single line of code~~—using only configuration files.

🛠️ System Architecture & Logic

The Docker Agent operates as a bridge between high-level AI models and actionable tools. The relationship can be expressed mathematically as:

$Agent = (Model + Instructions) \times (Tools + Memory) \rightarrow Solution$

Below is the operational flow of the runtime:

🚀 Key Capabilities

The platform is designed for flexibility and scalability. Its primary features include:

Feature	Description
Multi-Agent Design	Orchestrate teams of specialized agents that delegate tasks to one another.
Tool Ecosystem	Access built-in utilities or any MCP server (whether local, remote, or containerized).
Provider Agnostic	Supports OpenAI, Anthropic, Gemini, AWS Bedrock, Mistral, xAI, and the Docker Model Runner.
Declarative Config	Uses YAML for version-controlled, shareable agent definitions.
Advanced Logic	Integrated tools for `think`, `todo`, and persistent `memory`.
RAG Integration	Pluggable retrieval using BM25, embeddings, hybrid search, and reranking.
OCI Distribution	Push agents to any OCI-compliant registry to pull and run them globally.

⚙️ Configuration Example

Agents are defined using a simple schema. Here is a basic example of an agent configuration:

agents:
  root:
    model: openai/gpt-5-mini
    description: A helpful AI assistant
    instruction: |
      You are a knowledgeable assistant that helps users with various tasks.

toolsets:
  - type: mcp
    ref: docker:duckduckgo

To launch this specific configuration, you would use: docker agent run agent.yaml.

📦 Installation & Setup

Prerequisites

Install Docker Desktop (v4.63 or newer) — the docker-agent CLI plugin is included by default.

Manual Installation

If you are not using the bundled version, you can:

Download the docker-agent binary.
Symlink it to your plugins folder: ln -s /path/to/docker-agent ~/.docker/cli-plugins/docker-agent

Environment Configuration

You must provide an API key for your chosen provider. For example: export OPENAI_API_KEY=sk-...

Note: Alternatively, you can utilize the Docker Model Runner for fully local model execution.

⚡ Quick Start Guide

Get up and running quickly with these common commands:

Launch Default: docker agent run
Run from Registry: docker agent run myorg/agent:tag
Interactive Creation: docker agent new
Custom Config: docker agent run agent.yaml

For more complex scenarios, refer to the examples/ directory in the repository.

📖 Documentation & Community

The project is licensed under the Apache-2.0 license.

Resource Map

Guides: Installation $\rightarrow$ Model Setup $\rightarrow$ Quick Start.
Technical Refs: Agents, Models, Tools, Multi-Agent Config, TUI, CLI, MCP Mode, and RAG.
Contributing: See the Contributing guide to help improve the tool.

Fun Fact: The developers actually use the tool to maintain itself: docker agent run ./golang_developer.yaml

Get Involved

Community: Join the Docker Community Slack in the #docker-agent channel.
Telemetry: Anonymous usage data is collected to enhance the runtime experience.