Neuro-symbolic AI agents combine two approaches to artificial intelligence: on the one hand a neural (data-driven) part that works on the basis of large language models and/or deep learning and on the other hand a symbolic (rule and knowledge driven) part, which works on the basis of logic, knowledge graphs and ontologies.
The neural part recognizes patterns, processes language or images, can generalize, and handle incomplete information. The symbolic part ensures logical reasoning, structure, traceable decisions, and compliance with rules. This combination results in AI agents that are less affected by uncontrolled output, because reasoning is tested against formal rules.