Agency is a Python framework for building agentic systems, targeting developers who need to integrate AI agents with traditional software. It offers a minimal, flexible foundation for creating custom agent-based applications, enabling scalable and adaptable architectures.

How It Works

Agency employs an Actor model where all entities (AI agents, software interfaces, human users) are instances of the Agent class. Agents expose callable actions (methods decorated with @action) that other agents can discover and invoke via messages. This allows for dynamic interaction and integration. Actions can have access policies for controlled invocation, and agents support lifecycle callbacks (before_action, after_action) for custom logic. Agents communicate within a Space, with LocalSpace for in-process communication and AMQPSpace for networked systems using AMQP.

Quick Start & Requirements

• Install: pip install agency or poetry add agency
• Prerequisites: Python. AMQPSpace requires an AMQP server (e.g., RabbitMQ).
• Demo: Available at examples/demo with Gradio UI.

Highlighted Details

• Actor model for agent communication.
• Supports multiprocessing and multithreading.
• AMQP support for networked agent systems.
• Action access policies and lifecycle callbacks.
• LocalSpace and AMQPSpace for communication.

Maintenance & Community

• Contributions are welcomed via the contributing guide.
• Planned work is tracked on the issues page. Discussions are encouraged.

Licensing & Compatibility

• License: Not explicitly stated in the README.

Limitations & Caveats

The licensing is not specified, which may pose a risk for commercial adoption or closed-source integration. The project appears to be in active development, with planned work and an experimental demo application.