Models are advancing quickly. GPT-5, Claude Sonnet, Gemini. Each release gives us more capabilities. But most real work isn\u2019t solved by a single model.<\/p>\n\n
Developers are realizing they need a system of agents:<\/strong> different types of agents working together to accomplish more complex tasks. For example, a researcher to find information, a writer to summarize, a planner to coordinate, and a reviewer to check accuracy.<\/p>\n\n The challenge is that today, building a multi-agent system is harder than it should be.<\/strong> Context doesn\u2019t flow cleanly between agents. Tools require custom integration. Sharing with a teammate means sending instructions and hoping they can re-create your setup.<\/p>\n That\u2019s the problem cagent<\/a> solves.<\/p>\n\n In this blog, we\u2019ll walk you through the basics, how to create a multi-agent AI system in minutes, and how cagent makes this possible.\u00a0<\/p>\n\n A multi-agent system<\/strong> is a coordinated group of AI agents that collaborate to complete complex tasks. Using cagent, you can build and run these systems declaratively, no complex wiring or reconfiguration needed.<\/p>\n\n Figure 1: cagent workflow for multi-agent orchestration.\u00a0<\/strong><\/p>\n\n cagent<\/a> is an open-source tool for building agents<\/strong> and a part of Docker\u2019s growing ecosystem of AI tools<\/a>.\u00a0<\/p>\n\n Instead of writing glue code to wire up models, tools, and workflows, describe an agent (or a team of agents) in a single YAML file<\/strong>:<\/p>\n Which model the agent uses (OpenAI, Anthropic, Gemini, or a local one)<\/p>\n What its role or instructions are<\/p>\n Which tools it can use (like GitHub, search, or the filesystem)<\/p>\n And, if needed, which sub-agents it delegates to<\/p>\n This turns agents into portable, reproducible artifacts you can run anywhere and share with anyone.\u00a0<\/p>\n\n Create, run, and share multi-agent AI systems more easily with cagent.<\/p>\n\n Orchestrate agents (and sub-agents) more easily <\/strong>\u2013 Define roles and delegation (sub-agents). cagent manages calls and context.<\/p>\n Let agents use tools with guardrails <\/strong>\u2013 Grant capabilities with MCP: search, GitHub, files, databases. Each agent gets only the tools you list and is auditable.<\/p>\n Use (and swap) models<\/strong> \u2013 OpenAI, Anthropic, Gemini, or local models through Docker Model Runner<\/a>. Swap providers without rewriting your system.<\/p>\n Treat agents like artifacts <\/strong>\u2013 Package, version, and share agents like containers.<\/p>\n Here\u2019s what that looks like in practice.<\/p>\n\n agents: researcher: writer: instruction: Write a concise, clear summary from the researcher\u2019s notes.\n<\/p><\/div>\n\n The coordinator delegates, the researcher gathers, and the writer drafts. You now have a functioning team of agents.<\/p>\n\n Now, anyone on your team can run the exact same setup with:<\/p>\n\n That\u2019s a full multi-agent workflow, built and shared in under 5 minutes.<\/p>\n\n These principles keep cagent an easy-to-use and customizable multi-agent runtime to orchestrate AI agents.<\/p>\n\n Declarative > imperative.<\/strong> Multi-agent systems are mostly wiring: roles, tools, and topology. YAML keeps that wiring declarative, making it easy to define, read, and review.<\/p>\n Agents as artifacts.<\/strong> Agents become portable artifacts you can pull, pin, and trust.<\/p>\n Small surface area.<\/strong> A thin runtime that does one job well: coordinate agents.<\/p>\n Developers are already exploring different multi-agent use cases with cagent. Here are some examples:<\/p>\n\n 1. PR and issue triaging<\/strong><\/p>\n Collector<\/em> reads PRs\/issues, labels, failing checks<\/p>\n Writer<\/em> drafts comments or changelogs<\/p>\n Coordinator<\/em> enforces rules, routes edge cases<\/p>\n 2. Research summarizing<\/strong><\/p>\n Researcher<\/em> finds and cites sources<\/p>\n Writer<\/em> produces a clean summary<\/p>\n Reviewer<\/em> checks for hallucinations and tone<\/p>\n 3. Knowledge routing<\/strong><\/p>\n Router<\/em> classifies requests<\/p>\n KB agent<\/em> queries internal docs<\/p>\n Redactor<\/em> strips PII before escalation<\/p>\n Each one starts the same way: a YAML file and an idea. And they can be pushed to a registry and run by anyone.<\/p>\n\n cagent gives you the fastest path forward to build multi-agent systems. It\u2019s open-source<\/strong>, easy to use, and built for the way developers already work. Define your agents, run them locally, and share them, all in a few lines of YAML.<\/p>\n\n YAML in, agents out.<\/strong><\/p>\n\n Run the following to get started:<\/p>\n\n Get the technical details from our cagent documentation<\/a>.\u00a0<\/p>\n We\u2019d love to hear what you think. Join us in the Docker Community Slack<\/a>.\u00a0<\/p>\n Dive into more topics about AI and Docker<\/a>.\u00a0<\/p>\nWhat\u2019s a multi-agent system?<\/strong><\/h2>\n\n
Meet cagent: The best (and open source) way to build multi-agent systems<\/strong><\/h2>\n
Multi-agent challenges that cagent is solving<\/strong><\/h3>\n\n
How to build a multi-agent system with Docker cagent<\/strong><\/h2>\n\n
Step 1: Define your multi-agent system<\/strong><\/h3>\n\n
\n root:
\n model: anthropic\/claude-sonnet-4-0
\n instruction: |
\n Break down a user request.
\n Ask the researcher to gather facts, then pass them to the writer.
\n sub_agents: [“researcher”, “writer”]<\/p>\n
\n model: openai\/gpt-5-mini
\n description: Agent to research and gather information.
\n instruction: Collect sources and return bullet points with links.
\n toolsets:
\n – type: mcp
\n ref: docker:duckduckgo<\/p>\n
\n model: dmr\/ai\/qwen3
\n description: Agent to summarize notes.<\/p>\nStep 2: Run the YAML file<\/strong><\/h3>\n\n
Step 3: Share it on Docker Hub<\/strong><\/h3>\n
First principles: Why cagent works<\/h3>\n\n
What developers are building with cagent<\/strong><\/h2>\n
Get started<\/h3>\n\n
\ncagent new
\ncagent run agent.yaml\n<\/div>\n\nLearn more<\/h3>\n\n