Docker · 2026-10-07 · notable
Docker Agent 1.149 — Docker's YAML agent runtime loads skills from GitHub
Docker Agent is Docker's Apache-2.0 tool for building AI agents and agent teams in YAML and running them with `docker agent run`. Version 1.149.0 loads skills from public GitHub repos and adds an evaluator backend.
Write an agent as a YAML file, give it MCP tools, and run or share it like a container.
Key specs
| GitHub stars | 3,670 |
|---|
What is it?
Release 1.149.0 of Docker Agent, out on 7 October 2026, lets a `skills:` entry point straight at a public GitHub repository and adds a dedicated evaluator service as an alternative to LLM-as-judge. Docker Agent itself is Docker Engineering's agent builder and runtime, a plugin for the docker CLI that Docker Desktop 4.63+ already includes. It reached the Hacker News front page the same day.
How does it work?
Each agent in a Docker Agent YAML file has a model, a description, an instruction and toolsets; a root agent can delegate work to specialist agents. Tools come from built-in think, todo and memory tools or any MCP server, including ones that run as Docker containers. Models can be OpenAI, Anthropic, Gemini, Bedrock, Mistral, xAI or local ones through Docker Model Runner, and finished agents can be pushed to and pulled from any OCI registry.
Why does it matter?
Teams that already use Docker get an agent runtime inside a tool they know, with agents that are versioned files rather than code. Loading skills by GitHub URL removes a copy step when reusing shared skill libraries, and the new evaluator backend gives agent tests a judge that does not depend on another LLM call.
Who is it for?
developers already working with Docker who want to build agents
Try it
docker agent run