docker/docker-agent
AI Agent Builder and Runtime by Docker Engineering
What it solves
Docker Agent provides a simplified, no-code way to build, run, and share AI agents. It removes the complexity of coding agentic workflows by using a declarative YAML configuration to define agent behavior, tools, and orchestration. It also solves the problem of portability, allowing agents to be packaged and pushed to OCI registries for distribution.
How it works
It operates as a Docker CLI plugin. Users define agents in YAML files where they specify the model (e.g., OpenAI, Anthropic, Gemini), the agent's instructions, and the toolsets they can use. The system supports multi-agent architecture, allowing specialized agents to delegate tasks to one another. It integrates with the Model Context Protocol (MCP) to provide a rich tool ecosystem and includes built-in support for RAG (Retrieval-Augmented Generation) with various search and reranking methods.
Who it’s for
Developers and AI practitioners who want to create complex AI agent teams without writing extensive code, and those who who want to run local or cloud-based models via a standardized interface.
Highlights
- Multi-agent orchestration: Automatically delegate tasks between specialized agents.
- OCI Registry support: Package and share agents as images in any OCI registry.
- MCP integration: Use any Model Context Protocol server for tools.
- AI provider agnostic: Compatible with a wide range of cloud providers (OpenAI, Anthropic, Gemini, etc.) and local models via Docker Model Runner.
- Built-in RAG: Pluggable retrieval using BM25, embeddings, and hybrid search.
- Declarative config: Use YAML for versionable and shareable agent definitions.