Docker Agent
The repository has about 3.4k stars and 481 forks. Version 1.149.0 shipped on October 7 and adds GitHub repository skill loading along with a dedicated evaluator judge backend. The tool works as a Docker CLI plugin. Developers define an agent in a YAML file, run it with one command, and share it through any OCI registry, the same way they handle container images. Docker first released the project as cagent. It is bundled with Docker Desktop 4.63 and later, and in versions 4.49 through 4.62 it carried the cagent name. A root agent hands tasks to sub-agents. Each sub-agent uses its own model and context and can reach external tools through MCP servers. Docker presents it as not a framework you write code in, and not a hosted SaaS. For AI builders, packaging agents as OCI artifacts means versioning, distribution and access control can run on the registries companies already have. Startups that sell agent orchestration now compete with a free default from a vendor already installed on most developer laptops. They will need to stand out on evals, hosting or domain depth.
