Show HN: Gutsy, a 0.8B Jev-compatible decision model that runs on your CPU (github.com)

The newly announced Gutsy model is a local, decision-making AI tool built on a compact 0.8 billion parameter architecture, specifically optimized to run on standard CPUs. This model offers calibrated probabilities for various decision options based on input states and questions, all without the need for network calls or additional costs. Gutsy operates on the Jev-style Decisions format, making it easy for existing clients to integrate it into their systems with minimal changes, primarily requiring just a new base URL. The significance of Gutsy in the AI/ML community lies in its local processing capability and deterministic behavior, ensuring data privacy and consistent outputs for repeated requests. Furthermore, it boasts impressive accuracy, particularly in handling multiple options—up to 255—and has been fine-tuned using rigorous validation techniques to reduce calibration error to an impressive 0.021. With the model's ability to cache states and handle option ordering robustly, Gutsy presents a powerful new tool for developers aiming to incorporate AI decision-making into their applications without compromising performance or privacy.