Show HN: Statewright – Visual state machines that make AI agents reliable (github.com)

Statewright has introduced a novel approach to enhance the reliability of AI agents through visual state machines, which act as guardrails that define which tools can be utilized at each phase of a problem-solving process. Instead of relying solely on larger models and extensive prompts, Statewright constrains the AI's tool and solution space, allowing it to focus on specific tasks. For example, during the planning phase, only read-only tools are accessible, while editing tools become available in the implementation phase, thus preventing irreversible errors. This structured approach enables models to reason more effectively and reduces the risk of catastrophic failures, such as executing unauthorized actions. The significance of this innovation lies in its ability to boost the performance of smaller and local models, demonstrating that with Statewright's constraints, models previously unable to complete tasks improved dramatically in their output. Tested on models ranging from 13.8GB to 42.5GB, it was found that these structured workflows led to superior task completion rates, reinforcing the idea that effective problem framing can lead to better AI performance. Additionally, Statewright's deterministic Rust engine optimizes these state machine definitions, offering a tangible solution to the brittleness commonly associated with AI agents, and it provides a user-friendly integration with existing coding platforms.