cagent
Build, run, and share AI agents with a declarative YAML config — Docker's agent builder and runtime.
About
cagent is Docker's agent builder and runtime: you define agents in YAML — model, instructions, toolsets — and run them with cagent run agent.yaml. No application code required. It supports multi-agent setups where specialized agents delegate to each other, plugs into any MCP server (local, remote, or Docker-based), and ships with built-in think, todo, and memory tools plus pluggable RAG (BM25, embeddings, hybrid, reranking).
It is provider-agnostic — OpenAI, Anthropic, Gemini, Bedrock, Mistral, xAI, and Docker Model Runner for local models — and agents can be packaged and pushed to any OCI registry, which is the Docker-native way of sharing them. It ships pre-installed with Docker Desktop 4.49+, and is also on Homebrew and as standalone binaries. The repo was renamed from docker/cagent to docker/docker-agent in 2026; the binary and concept are unchanged.
The honest caveat: YAML-defined agents are great for standard assistant shapes, but the moment you need custom control flow you will wish you had code. cagent covers the common ground well; exotic orchestration still belongs in a framework like LangGraph or the BeeAI framework.
What it can do
- Declarative YAML agent definitions — no code required
- Multi-agent architecture with automatic task delegation
- Rich tool ecosystem: built-in tools plus any MCP server
- Provider-agnostic: OpenAI, Anthropic, Gemini, Bedrock, Mistral, xAI, Docker Model Runner
- Built-in reasoning tools (think, todo, memory) and pluggable RAG
- Package and share agents via any OCI registry
Setup
- Install: ships with Docker Desktop 4.49+ (
cagentis on PATH), orbrew install cagent, or download a binary from GitHub Releases. - Set at least one provider API key (e.g.
export OPENAI_API_KEY=...), or use Docker Model Runner for local models. - Write an
agent.yamldeclaring the agent, model, instructions, and toolsets. - Run it:
cagent run agent.yaml.
Prerequisites
Docker Desktop 4.49+ (or Homebrew / standalone binary); at least one LLM provider API key, or Docker Model Runner for local models.
Use cases
Works with
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Adoption 7: starScore from 4,514 stars on docker/docker-agent, formerly docker/cagent (log scale vs 298k peer max); Maintenance 9: pushed 2026-10-15 (recency 10); ships with Docker Desktop 4.49+, active releases; Openness 10: Apache-2.0 (0.7*10) + official docs verified (0.3*10); Manual 8: Docker-official agent runtime corroborated by repo, Docker Desktop release notes; docked: YAML-only config limits exotic orchestration. License: Apache-2.0.
★ 4,514 GitHub stars
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 9.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · Adoption 7: starScore from 4,514 stars on docker/docker-agent, formerly docker/cagent (log scale vs 298k peer max)
- · Maintenance 9: pushed 2026-10-15 (recency 10); ships with Docker Desktop 4.49+, active releases
- · Openness 10: Apache-2.0 (0.7*10) + official docs verified (0.3*10)
- · Manual 8: Docker-official agent runtime corroborated by repo, Docker Desktop release notes; docked: YAML-only config limits exotic orchestration
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.