Agency

Agency is an open source Python library that provides an actor model framework for building agent integrated systems. It gives developers a minimal foundation for connecting AI agents with traditional software, so they can design whatever architecture their application needs rather than adopting a fixed agent pattern. It is aimed at Python developers who want to build and experiment with custom agent systems. In Agency every entity, whether an AI agent, a software interface or a human user, is an instance of an Agent class, and agents expose actions that other agents can discover and invoke at run time by sending messages. Actions can carry access policies, for example allowing an action at any time or requiring review before it runs, and agents can define callbacks around actions and lifecycle events for observability and control. The library supports multiprocessing and multithreading for concurrency and AMQP for networked agent systems, with detailed logging. A demo application includes OpenAI and Hugging Face transformers agent examples, an agent with operating system access, a Gradio interface and a Docker configuration, and the website hosts help articles, API documentation and an example application walkthrough. Agency is free under the MIT license and there is no paid tier. The most recent tagged release on GitHub dates from April 2024, so teams should check the repository's current activity before building on it.