Report Abuse

Basic Information

Agency is a Python library that implements an Actor model framework for building agent-integrated systems. It is designed for developers who want to connect AI-driven agents, human users, and software interfaces into flexible, scalable applications. The library provides a minimal foundation for experimenting with and constructing custom agent architectures, enabling agents to expose discoverable actions, enforce access policies, and communicate within shared environments called Spaces. Two Space implementations are provided for local and networked communication. The repository includes a demo application showcasing multiple agent examples and integrations with model providers and tooling. The project aims to reduce boilerplate for building agent systems while offering hooks for observability, lifecycle management, and integration with existing software stacks.

Links

App Details

Features
The repository documents an easy to use class and method based API for defining Agent classes and actions with decorators. It supports action access policies and lifecycle callbacks for before and after action handling. Concurrency is supported through multiprocessing and multithreading and AMQP support enables networked agent systems. Observability features include detailed logging and action/lifecycle callbacks. Included artifacts and examples demonstrate integration with external model providers and tools, a Gradio demo UI, and Docker configuration for development. The package is installable via pip or poetry and ships example code to help developers get started quickly.
Use Cases
Agency helps developers build and experiment with custom agent-based applications by providing a compact, extensible framework that handles agent registration, action discovery, message passing, and lifecycle hooks. It enables local and distributed communication patterns via LocalSpace and AMQPSpace to scale from single-process prototypes to networked deployments. Access policy decorators and callback hooks support safety, access control, and monitoring workflows. Concurrency primitives and AMQP integration aid performance and distribution. The demo application, examples, and documentation lower the barrier to integrate LLMs and other model providers, prototype agent interactions, and adopt observable patterns for production systems.

Please fill the required fields*