Tunny is an optimization plugin for Grasshopper that leverages the Optuna framework to perform complex single and multi-objective design solvers. This computational tool integrates the Optuna optimization framework directly into the Grasshopper environment, allowing for advanced mathematical problem-solving within parametric workflows. It provides a robust interface for handling complex design variables, enabling users to define objectives and constraints for automated generation. The plugin architecture is designed for high-performance iteration, supporting both single-objective and multi-objective optimization tasks. By bridging Grasshopper with external solver capabilities, it facilitates data-driven decision-making, helping designers to refine geometry based on specific performance criteria, environmental factors, or structural requirements without manual intervention for every single design variation.

The download is a 1.8 MB file containing .ZIP files.

To integrate this tool, load the plugin into the Grasshopper components folder to access the solver nodes. Users define input parameters and output objectives within the canvas, connecting them to the solver component. The workflow involves setting up the optimization study, defining the search space, and executing the solver to iterate through design permutations. Results are typically harvested for post-processing and analysis, allowing for the selection of optimal configurations based on the generated data sets.

This tool is essential for architectural and engineering firms engaged in performance-driven design, facade rationalization, and structural form-finding. It is frequently employed during the early design stages to explore vast solution spaces for building massing, environmental daylighting optimization, or material efficiency. By automating the search for the best design outcomes, it assists professionals in balancing conflicting project requirements. It is a critical asset for computational designers working on complex geometric projects where traditional manual modelling methods are insufficient for achieving high-performance results.