Agency

Agency is a refreshingly minimal framework for building multi-agent systems. It strips away the complexity of larger platforms like LangChain or AutoGen, letting you focus on agent logic. If you're a Python developer comfortable with async programming and want to prototype agent interactions quickly, this is a great choice. When should you pick Agency? When you need a thin, understandable abstraction for agent communication and lifecycle management. The local space works great for single-process experiments, and the AMQP space allows distributed agents. The audit logging and resource management are nice touches without adding bloat. When should you pass? If you need a rich ecosystem of pre-built tools, integrations with LLM providers, or a graphical interface. Agency expects you to handle those yourself. It's not for production systems requiring robust monitoring or enterprise support. Compared to alternatives: LangChain gives you hundreds of integrations and prompt templates but can feel heavy. AutoGen from Microsoft offers more built-in patterns for conversation and code execution. Agency is closer to a raw bones abstraction—less opinionated, more transparent, but with a steeper climb to a working system. In practice, we'd reach for Agency when we want to understand exactly how agent messaging works under the hood. It's a fantastic learning tool. But for a shipping product, we'd probably gravitate to a more mature framework unless Agency's minimalism is a hard requirement.