cagent

Version: 1.9.10
OS: Ubuntu 24.04
Category: Developer Tools

Description

cagent is Docker's AI agent framework for building and running intelligent agents with YAML-first configuration and a flexible tool ecosystem. It supports both single-agent and multi-agent architectures, integrates with cloud and local model providers, and can route tasks across specialized agent roles.

What is cagent?

cagent provides a composable runtime for agent workflows:

Multi-agent architecture for domain-specific specialists
Smart delegation between specialist sub-agents
Built-in reasoning helpers such as think, todo, and memory patterns
MCP-based tool integration for external capabilities
Portable YAML configuration for repeatable agent definitions

Key Features

Build intelligent AI agents with specialized capabilities
Create multi-agent teams for complex workflows
Use MCP tools for extensible integrations
Support both cloud providers and local runtimes
Push and pull agents through Docker Hub
Container-oriented deployment model

Software Included

Package	Version	License
cagent	v1.9.10	Apache License 2.0

System Requirements

cagent is installed as a binary on Ubuntu 24.04 and requires Docker for containerized models and tools.

Usage Level	RAM	CPU
Basic agents with cloud APIs	1 GB	1 CPU
Local models (small)	4 GB	2 CPU
Local models (medium)	8 GB	4 CPU
Local models (large)	16 GB+	8 CPU+

Note: Local model inference requirements depend on model size.

Getting Started

Quick Start

Deploy cagent on an EasyCloudify VPS.
Connect via SSH.


bash
ssh root@your-vps-ip

Configure API keys for your selected providers.
Run an example agent to validate installation.


bash
cagent run /opt/cagent/examples/basic_agent.yaml

Setting Up API Keys


bash
# For OpenAI models
export OPENAI_API_KEY=your_openai_key_here

# For Anthropic models
export ANTHROPIC_API_KEY=your_anthropic_key_here

# For Google Gemini models
export GOOGLE_API_KEY=your_google_key_here

Running Your First Agent


bash
# Run a basic agent
cagent run /opt/cagent/examples/basic_agent.yaml

# Run with local Docker Model Runner
cagent run /opt/cagent/examples/dmr.yaml

# Other examples
cagent run /opt/cagent/examples/pirate.yaml
cagent run /opt/cagent/examples/pythonist.yaml
cagent run /opt/cagent/examples/todo.yaml

Creating Custom Agents


bash
cagent new
cagent new --model openai/gpt-4o-mini
cagent new --model dmr/ai/gemma3:2B-Q4_0

Using Docker Model Runner


yaml
version: "2"

agents:
	root:
		model: local-model
		description: A helpful AI assistant
		instruction: You are a knowledgeable assistant.

models:
	local-model:
		provider: dmr
		model: ai/gemma3:2B-Q4_0
		max_tokens: 8192

Agent Store


bash
cagent pull docker.io/username/my-agent:latest
cagent push ./my-agent.yaml docker.io/username/my-agent:latest
cagent run creek/pirate

Configuration

Basic Agent Configuration


yaml
version: "2"

agents:
	root:
		model: openai/gpt-4o-mini
		description: A helpful AI assistant
		instruction: |
			You are a knowledgeable assistant that helps users with various tasks.

models:
	openai:
		provider: openai
		model: gpt-4o-mini
		max_tokens: 4096

Multi-Agent Teams


yaml
version: "2"

agents:
	root:
		model: coordinator
		description: Main coordinator agent
		instruction: |
			You coordinate tasks and delegate to specialized agents.
		sub_agents: ["researcher", "writer"]

	researcher:
		model: research-model
		description: Research specialist
		instruction: |
			You research topics and gather information.

	writer:
		model: writing-model
		description: Writing specialist
		instruction: |
			You create well-written content based on research.

models:
	coordinator:
		provider: anthropic
		model: claude-sonnet-4-0
	research-model:
		provider: openai
		model: gpt-4o
	writing-model:
		provider: anthropic
		model: claude-sonnet-4-0

Adding Tools via MCP


yaml
version: "2"

agents:
	root:
		model: assistant
		description: Assistant with web search capabilities
		instruction: You help users by searching the web when needed.
		toolsets:
			- type: mcp
				ref: docker:duckduckgo

models:
	assistant:
		provider: openai
		model: gpt-4o-mini
		max_tokens: 4096

Common Commands


bash
cagent --help
cagent run ./my-agent.yaml
cagent new
cagent build ./my-agent.yaml my-agent:latest
cagent pull creek/pirate
cagent push ./my-agent.yaml username/my-agent:latest
cagent readme ./my-agent.yaml

Examples and Documentation

Local examples: /opt/cagent/examples/
Quick reference: /opt/cagent/README.txt
GitHub examples: https://github.com/docker/cagent/tree/main/examples

Example categories:

Basic single-agent examples
Advanced tool-enabled examples
Multi-agent collaboration examples

Use Cases

Code assistance workflows
Research and summarization pipelines
Content creation with multi-agent teams
Task automation with system and tool integrations
Custom specialist teams for internal operations

Post-Deployment Notes

After deployment, the instance includes:

cagent binary at /usr/local/bin/cagent
Docker pre-installed
Example agents in /opt/cagent/examples/
Quick guide in /opt/cagent/README.txt

Important notes:

Store API keys in environment variables
Local models require additional RAM
Some MCP tools may require additional dependencies
Use cagent --version to verify installed version

Resources

GitHub: https://github.com/docker/cagent
Usage Docs: https://github.com/docker/cagent/blob/main/docs/USAGE.md
Contributing: https://github.com/docker/cagent/blob/main/docs/CONTRIBUTING.md
Docker Community Slack: https://dockercommunity.slack.com/archives/C09DASHHRU4

Cagent