# Docker AI
Docker AI is [[Docker]]'s push to make the existing container workflow (`docker`, `docker compose`, Docker Hub, OCI artifacts) the substrate for building, running, and shipping agentic AI apps. The mental model is three layers: **models** (reasoning), **agents** (orchestration), and **tools** (integration), all wireable through a single `docker compose up`, with cloud GPU available when local hardware falls short.
## The pieces
**Models**
- [[Docker Model Runner]]: run and serve open models locally behind an OpenAI-compatible API
**Agents**
- [[Docker Agent]] (formerly cagent): build teams of specialized agents from YAML
- [[Gordon (Docker AI Agent)]]: Docker's built-in assistant ("Ask Gordon")
- [[Docker Compose for AI]]: `models` as a first-class Compose element
**Tools**
- [[Docker MCP Catalog]]: 300+ verified MCP servers as container images
- [[Docker MCP Toolkit]]: configure MCP servers once, share across clients
- [[Docker MCP Gateway]]: proxy that runs MCP servers in isolated containers with credential injection
**Execution and isolation**
- [[Docker Sandboxes]]: microVM isolation for running agents safely
- [[Docker Offload]]: ephemeral cloud GPU when local hardware isn't enough
## Related
- [[Docker]]
- [[Model Context Protocol (MCP)]]
- [[AI Agents]]
- [[Agentic Engineering]]
- [[Large Language Models (LLMs)]]