ZeoAgent: Autonomous Design of Zeolite Frameworks through Pore-Topology-Guided Generation and Evaluation

Abstract

Autonomous agents are emerging as a promising approach to assist and accelerate materials design and discovery. For zeolites, an important class of crystalline porous materials, this paradigm could address a long-standing challenge in framework discovery: the lack of rational strategies for coordinating framework generation with property objectives. Yet building such an autonomous agent system for zeolites remains challenging, as performance is governed by periodic pore topology and strict framework connectivity constraints. This requires the agent to jointly reason over pore architectures and generate valid frameworks. Here we present ZeoAgent, a zeolite-specific system that converts natural-language objectives into computational operations for zeolite structure analysis and property prediction as well as framework design. ZeoAgent uses a point-cloud representation of pore architecture to learn from existing zeolite frameworks and guide the generation of framework candidates, which are then evaluated against the design objective within the same loop. We demonstrate ZeoAgent in representative design tasks, including the design of frameworks with prescribed structural constraints, enhanced diffusion performance, and pore dimensions suited for CO₂/CH₄ separation. This work establishes a route for goal-directed zeolite design, showing how AI agents can be adapted to periodic porous materials governed by constrained pore topology.