AgentBreeder

AgentBreeder is a tool for defining AI agents in a portable YAML file and deploying them across local and cloud targets. Its own documentation frames it around three core actions: build agents, deploy anywhere, and govern automatically. The same configuration is intended to work with any framework and any cloud, with governance, RBAC, cost tracking, and an audit trail applied automatically on every deploy.

The system includes a versioned and cached prompts layer, a hybrid RAG setup with vector and graph search, and support for A2A agent-to-agent communication over JSON-RPC with authentication. It also mentions an MCP capability with auto-discovered tools and a shared organization registry for prompts, tools, knowledge bases, and MCP servers. A deployment shown in the docs uses a single agent.yaml file and a terminal command, and the output indicates validation, building, deploying, prompt caching, RAG indexing, and MCP wiring.

AgentBreeder lists support for LangGraph, CrewAI, Claude SDK, Google ADK, OpenAI Agents, and custom agents. It also describes multiple deployment targets, including Local Docker Compose, AWS ECS Fargate, GCP Cloud Run, Azure Container Apps, App Runner, and Kubernetes. The page says business users can drag and drop, engineers can write YAML, and researchers can use the full SDK, with all three paths compiling to the same pipeline and governance model. It is delivered as a Python package installable with pip3, and the documentation presents it as open source on GitHub.

The product is described as having an 8-step atomic deployment pipeline with all-or-nothing behavior and no partial deploys. A separate note says AgentBreeder Cloud is coming, with managed infrastructure, zero DevOps, and automatic governance. The entry also shows a brief example configuration that includes a name, version, team, framework, model settings, and a local runtime setting.