Stately Agent: Build agents as state machines (github.com)

Stately AI has unveiled Stately Agent, a new framework that allows developers to build AI agents as state machines, offering enhanced control and inspection over their functionality. This innovative approach integrates model requests and decisions within the XState framework, enabling users to design agent workflows that are explicitly defined and easily tested. The state machines dictate the actions of the agents, with events proposed by the AI model, while the state machine governs the permitted actions and transitions based on pre-defined limits, such as monetary thresholds for refund approvals. The significance of Stately Agent lies in its robust modeling capabilities, allowing complex decision-making processes to be encapsulated within easily manageable structures. With this framework, developers can create agents that feature advanced functionalities such as approval states, personalized decision paths, and the ability to pause for human input while maintaining operational continuity. Furthermore, the alpha version allows for broad applications, whether it's for automating customer service transactions, gaming interactions, or structured data handling. The capacity to simulate, test, and visualize agent behavior without constant API calls enhances both efficiency and reliability in AI-driven systems.