Show HN: Theus – I built a framework to make AI-generated code safe to run (github.com)

Theus, a newly developed framework, introduces a Process-Oriented Programming (POP) paradigm designed to enhance the safety and integrity of AI-generated code. By transforming applications into deterministic workflows, Theus addresses common pitfalls in state management, such as implicit mutations and race conditions, making it particularly significant for complex systems like AI agents and financial applications. Theus employs a micro-kernel architecture that enforces strict runtime invariants, ensuring processes can only access explicitly declared data, thereby reducing potential bugs and enhancing security. Key technical features of Theus include a structured approach to state management, where state is treated as a three-dimensional space defined by layers, semantic types, and zones. This framework supports advanced typing in Python 3.12+, facilitating more rigorous input validation and error handling. Theus’s hybrid transaction model combines fast updates with strong rollback safeguards for collections while allowing lower overhead for high-performance AI workloads. Additionally, automated tools for auditing, schema generation, and project scaffolding streamline the development process. Overall, Theus reflects a significant step toward safer and more reliable AI programming practices, promising to simplify complex workflows and improve transparency in system operations.